Our own findings
What we found ourselves
Things that came out of measuring the procurement record and were in no news story. Each one states which law it would engage, what is missing before it can be asserted, and who has to answer.
How this section is built
Three conditions, and one warning
To appear here a finding must satisfy three things at once: it came out of measuring this corpus, it has no prior press coverage, and it engages a specific law. If a finding is good but a newspaper already reported it, it goes to the relevant investigation, not here.
The warning is the important part. Naming a law is not accusing anyone of breaking it, and this page does not say a crime was committed: it says what would have to be looked at and what is missing to know. Sharing a phone number does not prove an agreement; a single-offer purchase may be a single-supplier market; an impossible unit price is usually a data-entry error and not a diversion of funds. Those who have the powers to find out — the Competition Commission, the Court of Auditors, the transparency board, a prosecutor — have powers we do not.
That is why every card carries the "what is missing" field. It is the field that keeps it honest and the first one anybody would want to trim. It does not get trimmed.
1 / 32
Firms competing against each other from the same doorway
- What we measured
- Eight pairs of companies bid on the same tender while sharing a phone number or an address, across 24 tenders in 2025 and 2026. In three of those pairs each firm declares the link on its own website: same phone and same office, and in one case one site states verbatim that it is "a company of the group" of the other. In 6 of the 24 tenders the purchase was awarded to both.
- Which law it would engage
Establecer, concertar o coordinar las ofertas o la abstención en licitaciones, concursos o subastas.
Law 18,159, article 4 BIS, item 4 — added by Law 19,833 of 2019. Investigated by the Competition Commission, on its own initiative or on a complaint.
- What is missing before asserting it
- Everything essential: sharing a doorway does not prove the offers were coordinated. Two firms of one group may legitimately bid on the same tender, and bidding is not colluding. It would take comparing the price offers — which are not public — and establishing whether there was an agreement. The Competition Commission can do that; we cannot.
- Who has to answer
- The public body that awarded each of the 24 tenders, which can say whether it knew two of its bidders answered the same phone.
Read the full investigation → Measured on 2026-08-13 · Re-measure it with: npx tsx tests/unit/competencia-aparente.verify.ts
2 / 32
One in five competitive tenders drew a single offer
- What we measured
- 852 of the 4,372 purchases with published bidders (19.5%) received exactly one offer, and all are procedures that exist to produce competition: price competition 522 of 2,781, abbreviated tender 323 of 1,489, public tender 7 of 102. It is the first measurement of its kind on Uruguayan procurement.
- Which law it would engage
No law is broken by receiving a single offer. It is the first of the eight risk indicators oversight bodies use worldwide, and what it opens is a question about the tender document, not a case file.
TOCAF, article 33 — defines which procurement procedure applies when.
- What is missing before asserting it
- Knowing why. A single-offer tender may reflect a market with one capable supplier, a specification written too closely to one product, a short deadline, or nobody hearing about it. Each explanation leads to a different action and none follows from the number.
- Who has to answer
- Each public body, about its own tenders: whether it re-tendered, reviewed the specification, or asked the suppliers who did not bid.
Read the full investigation → Measured on 2026-08-13 · Re-measure it with: npx tsx tests/unit/competencia-aparente.verify.ts
3 / 32
A public body publishes winners where offers should be
- What we measured
- Across the 65 inspected purchases of Montevideo's city government, the "participating suppliers" list matches the awarded-supplier list exactly, all 65 times: nobody who bid and lost ever appears. We checked it against an official record with three participants where all three were awarded — one item each — and the page does not say how many offers were received. Other bodies do show losers: the social security bank in 742 of 1,021 purchases.
- Which law it would engage
Procurement publicity exists so that it is possible to know who competed, not only who won. If the block publishes winners, the fact citizens need in order to watch competition is not published, even though the page looks complete.
TOCAF, article 33, and the publicity regime of the state procurement portal.
- What is missing before asserting it
- Knowing whether it is a decision, a limitation of the system that body publishes through, or an error. That city government's records enter the portal by a different route — its identifiers start with "i" — which suggests a different integration rather than a deliberate omission.
- Who has to answer
- Montevideo's city government and the state procurement agency, which runs the portal where that block is published.
Read the full investigation → Measured on 2026-08-13 · Re-measure it with: npx tsx src/jobs/refresh-bidder-competition.ts --dry-run
4 / 32
Whole contracts recorded as "one unit at one peso"
- What we measured
- 2.4% of award lines since 2023 (28,008 of 1,156,503) carry a unit price of one peso or less. They are not evenly spread: at the Presidency they are 64.5% of its lines and at the state power utility 43.5%, and those lines carry 6,390 million pesos. The practical effect is that in those purchases the unit price stops being a price, and no comparison between bodies is possible: measured without filtering, "integral cleaning of premises" showed a 17-million-fold gap between one body and another.
- Which law it would engage
None that is broken on its own: it is not a diversion of funds nor an irregular purchase, it is a way of entering the data. What it breaks is oversight: the published record stops allowing the comparison that justifies publishing it.
TOCAF, article 33, and the publicity regime of the state procurement portal.
- What is missing before asserting it
- Knowing whether the system allows another way to record a lump-sum service contract. It may be a limitation of the form rather than a decision by the body, in which case the request goes to whoever designs the system, not to whoever fills it in.
- Who has to answer
- The bodies where the practice is the majority, and the state procurement agency, which defines the record format.
Measured on 2026-08-13 · Re-measure it with: npx tsx tests/unit/sobreprecio-organismo.verify.ts
5 / 32
Purchases in dollars counted as if they were pesos
- What we measured
- When the feed publishes a line with no currency, the record counts it as Uruguayan pesos. We checked it against the official page of purchase 1126716, which reads literally "U$S B 27,500.00" per unit for six units — USD 165,000 — and which the corpus stores as 0.17 million pesos: about forty times less. It is not an isolated case: 39,607 awards since 2015 (4.1%) have at least one item with no declared currency.
- Which law it would engage
None: it is a defect in the published data, not an irregularity in the purchase. But it has public consequences, because the error is not random: it systematically penalises those who quote in dollars — vehicles, medical equipment, imported IT — against those who quote in pesos, and that distorts any supplier ranking built on the record, including ours.
Publicity regime of the state procurement portal; currency is part of the published datum.
- What is missing before asserting it
- The correct currency, which is not in the feed. It can be read one by one on each purchase's official page, but there are almost forty thousand. In the meantime, the honest thing is to say those purchases are miscounted and not to use them for rankings.
- Who has to answer
- The state procurement agency, which publishes the feed without the field; and ourselves, who had been adding them up as pesos without warning anyone.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/moneda-nula.verify.ts
6 / 32
When the direct-purchase ceiling rose to $654,000, the bunching of purchases just below the ceiling moved with it
- What we measured
- Across awards in pesos from January to July — 69,026 in 2024, 55,792 in 2025 and 37,466 in 2026, deduplicated by ocid — we measured how many fall just below the direct-purchase ceiling and how many just above. One correction has to come first: the corpus stores amounts excluding tax and the legal ceiling includes it, so at the standard 22% rate the 2025 ceiling ($263,000) appears in the database as $215,574 and the 2026 one ($654,000) as $536,066. In the ±1.1% band around the old point the below/above ratio was 0.51 in 2024, 2.98 in 2025 and 1.03 in 2026 — the bunching dissolved — and around the new point it was 0.37 in 2024, 0.69 in 2025 and 3.71 in 2026, with 126 purchases below against 34 above. The result holds across window widths of ±3,000 to ±15,000 pesos (4.40 · 3.76 · 2.93 · 2.38 in 2026 against 0.65 · 0.69 · 0.85 · 0.91 in 2025) and no single body drives it: excluding the Maciel Hospital the 2026 ratio rises to 4.09. Sixteen awards in 2026 land exactly on $536,066; we opened all 16 official records and 11 read «Monto Total de la Compra: $ 654.000,00» to the peso, the other 5 between $653,998 and $653,999, and all 16 are labelled «Compra Directa» (10 belong to the national education council's governing board); in 2025 and 2024 there are none. What we did not find is a stampede towards direct purchase: the band the law opened went from 9.09% of awards in 2025 to 10.40% in 2026 (8.33% to 9.40% without the Maciel), against 8.88% in 2024, and in a random sample of 30 awards from 2026 in the $6,000 just below the point, 27 are direct purchases, 2 price competitions and 1 an exception purchase.
- Which law it would engage
No law is broken by what we measured, and it is worth saying so up front: a body buying directly for $654,000 is using exactly the authorisation Parliament gave it. The rule that sets the line is Law 20,446 (National Budget 2025-2029, of 16 December 2025), article 28, which replaces item C) of article 482 of Law 15,903 (TOCAF art. 33 lit. C) and reads, verbatim: «Directamente el monto de la operación no exceda de $ 630.000 (seiscientos treinta mil pesos uruguayos) a excepción de los Gobiernos Departamentales cuyo monto máximo autorizado para la compra directa será de $ 987.000 (novecientos ochenta y siete mil pesos uruguayos) a precios enero 2025». Indexed by the statistics institute, that $630,000 floor at January 2025 prices becomes $654,000 for January-December 2026, which is the figure the procurement agency publishes. The rule that would actually be engaged, and only if the next step confirms it, is a different one: the TOCAF forbids splitting a contract to avoid the competitive procedure its size would require. We did not measure that: we saw repetitions — the Maciel Hospital awarded 7 times to SANYFICO S.A. and the education board 6 times to SAN PEDRO CARRATO MARIO RICARDO, all in the 3% just below the ceiling — but we did not verify that they are one purchase split up, and seven months of hospital purchases from one supplier are perfectly normal.
Ley 20.446 (Presupuesto Nacional 2025-2029, 16/12/2025), art. 28, que sustituye el literal C) del art. 482 de la Ley 15.903 (TOCAF art. 33 lit. C). Amount in force January-December 2026 ($654,000; departmental governments $1,024,000) published by the procurement agency, indexed per the statistics institute's technical bulletin of 30/12/2025 (table «Montos Límites de las Adquisiciones Estatales»). Competent body: Agencia Reguladora de Compras Estatales (ARCE); external oversight: Tribunal de Cuentas.
- What is missing before asserting it
- Four concrete things, none of which is «let the body explain it». (1) The procurement method is not in the open data: zero of the 42,544 awards from January to July 2026 carry tender.procurementMethodDetails and the merged OCDS record does not carry it either; the «Compra Directa» label exists only in the HTML of each record, and we downloaded 70 records by hand. To say what share of the 3,896 purchases in the band are direct purchases you have to download the 3,896 records, or the procurement agency has to publish the field in the award release: until then the «27 of 30» is a sample, not a rate. (2) The tax conversion is an approximation: we divided by 1.22, correct for the standard rate but not for the 10% reduced rate nor for exempt services, so the real cut-off is not a line but a band ($536,066 at 22%, $594,545 at 10%, $654,000 exempt) and our count of 126 purchases is a floor, not the total; closing it requires each purchase's tax-inclusive «Monto Total de la Compra», which is also only in the HTML. (3) 2026 is an incomplete year in the feed — 37,466 awards in pesos from January to July against 55,792 in the same stretch of 2025, 33% fewer, and one channel is missing —: we normalised proportionally and repeated everything over a panel of the 63 bodies with at least 150 awards in both periods (there the band goes from 5.23% to 9.18%), but we cannot rule out that what is missing is skewed. (4) Whether the repetitions are contract splitting: someone has to read the item detail of the 7 Maciel-SANYFICO awards and the 6 education-board ones and see whether it is one object split or separate purchases. That is 13 records and we did not read them.
- Who has to answer
- The state procurement agency, for two distinct things: why the procurement method does not travel in the award release of the open data, being the fact that decides whether a purchase was competitive, and whether it is monitoring the effect of the new ceiling, which is the natural counterpart of having raised it. The national education council's governing board, for the direct purchases pinned at exactly $654,000 to the peso: if it is an administrative practice of authorising up to the ceiling, saying so is enough. The Court of Auditors, as external auditor, on whether the jump in the ceiling entered its 2026 oversight plan.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-tope-directa.verify.ts
7 / 32
The purchasing channel Law 20,446 made mandatory has not one award and not one peso in the open data
- What we measured
- Across 2,184,332 releases of the procurement agency's OCDS corpus (2002 to 13/08/2026) there are 48 records with tender.procurementMethodDetails = «Convenio Marco»: one per purchase, all 48 tagged [tenderUpdate], all 48 with no buyer, no award, not a single supplier and an amount of 0 — total of amount.primaryAmount = $0 —, against the 193,339 million (2023), 217,103 million (2024) and 178,059 million pesos (2025) the corpus does record. By year of record they are 1 (2015), 1 (2016), 6 (2017), 2 (2018), 9 (2019), 0 (2020), 2 (2021), 2 (2022), 6 (2023), 16 (2024), 3 (2025) and 0 in 2026, equivalent to 1, 1, 3, 2, 4, 0, 2, 2, 5, 12, 3 and 0 distinct framework agreements with activity. The agency's own per-purchase OCDS endpoint returns 404 («El identificador de la compra especificado no se encuentra en el sistema») for all 48 framework-agreement ids, and 200 for 12 ordinary 2025 purchases taken at random as a control. The site's HTML record, by contrast, does publish the award: we checked Framework Agreement 4/2025 (milk, 9 participating suppliers with tax ids, «awarded in part»), 1/2024 (materials, 13 suppliers) and 3/2024 (hospital clothing, 7 suppliers), with unit prices by zone, a «Ver Órdenes de Compra» button and a link to the Virtual Store: none of that reaches the open data. According to figures the agency gave the weekly Búsqueda — a press figure, not a measurement of ours — 2,276 purchase orders were issued through that channel in 2023, 4,059 in 2024 and 6,485 in the first nine months of 2025: none of those orders exists in the corpus, and neither does the buyer that signs them (0 records with a buyer or party containing «UACM»).
- Which law it would engage
No law is being broken and it has to be said in full: there is no legal obligation to pour framework agreements into the OCDS open-data set, the agency publishes the tender and the award on the state procurement site and publishes aggregate amounts on the Public Procurement Observatory. The rule that changed the weight of the matter is article 33 of the TOCAF as redrafted by Law 20,446, which reads verbatim: «De conformidad con lo anterior, previamente a la elaboración de un procedimiento competitivo de adquisición, las administraciones públicas estatales deberán consultar los convenios marco y sistemas dinámicos vigentes, así como la existencia de nóminas vigentes de procedimientos especiales aprobados al amparo del artículo 483 de la Ley Nº 15.903, de 10 de noviembre de 1987, modificativas y concordantes, que impliquen la agregación de demanda e incluyan nóminas de proveedores habilitados», and in the following paragraph: «Si el objeto de la contratación se encuentra incluido en alguno de los procedimientos anteriores, las administraciones públicas estatales, a excepción de los Gobiernos Departamentales, deberán adquirir a través de estos, pudiendo contratar por otro procedimiento competitivo previsto en la normativa vigente únicamente cuando se cumpla alguna de las siguientes condiciones» (different technical characteristics, lack of availability, or a better price outside the agreement). What Law 20,446 did, as of 16/12/2025, was turn into a mandatory and preferred route a channel whose transactions are declared outside the scope of the open data; the measurable result today is that the procedure the law orders bodies to use first is the only one that cannot be audited with the data the State publishes in reusable format.
TOCAF (Decreto 150/012), art. 33, texto según Ley N° 20.446 de 16/12/2025, arts. 27 y 28. Framework-agreement regime: Decreto N° 367/018 and its particular regulation. Declared scope of the open data: the agency's entry in the OCP Data Registry — «No information on centralized purchases or framework agreements is included». Competent body: Agencia Reguladora de Compras Estatales (ARCE), Unidad de Adquisición de Convenios Marco (UACM).
- What is missing before asserting it
- Four concrete things, none of them «let somebody explain it in general». (1) THE AMOUNT: we cannot say how many pesos went through framework agreements. The agency publishes it only on the Observatory's «Convenios Marco» dashboard, in thousands of pesos by agreement, body and product, on a portal with no bulk download and no endpoint, and there is no framework-agreement dataset at all on the national open-data catalogue; someone has to extract the series from inside the dashboard or request them from the agency. Without that figure the story says «zero in the open data» but cannot say «zero against X billion». (2) WHERE THE 48 ENDED UP: we inspected 7 award records (3 awarded in part and 4 declared void); 41 HTML records remain to be read, one request each. (3) THE PURCHASE ORDERS: the official record has a «Ver Órdenes de Compra» button per item and supplier, but the modal is filled by AJAX and comes back empty in the static HTML; that endpoint has to be identified to know which body bought what and when. Until then the ~12,800 orders the agency says it issued between 2023 and September 2025 are a press figure, not a measurement of ours. (4) THE AGENCY'S ANSWER on whether it will incorporate framework agreements and purchase orders into the OCDS set, and by when. Also, a limit of the data itself: the 0 in 2026 is evidence of nothing — 2020 also had 0 tenders, tenders run from 1 to 12 per year and the 2026 corpus is distorted —; the finding is not «there were none in 2026», it is «in eleven years there is not one award and not one peso».
- Who has to answer
- The state procurement agency, which both runs the channel and publishes the open data. Specifically: the framework-agreement purchasing unit (UACM), which appears as buyer on the official records; the agency's management, which pushed in Parliament the preference for framework agreements that ended up in Law 20,446 and publicly gave the purchase-order figures; and the open-data area / Public Procurement Observatory, responsible for the «Datos Históricos de Compras» set on the national catalogue and for the scope declared to the OCP Data Registry. There is a single question and it is answered yes or no: will the agency publish in OCDS the framework-agreement awards and the Virtual Store purchase orders, now that the TOCAF requires buying through them? In the background, the Court of Auditors, which reviews these resolutions as the records themselves show (Resol. TCR 823/26 and 106/2026), and the public information access unit if the formal-request route proves necessary.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-convenios-marco.verify.ts
8 / 32
The state telecom published 11,641 million pesos awarded in 2025 without naming the winner: the name is only inside the scanned resolution
- What we measured
- In calendar year 2025 the corpus holds 109,465 releases with an award and a valid amount, worth 178,059 million pesos; 539 of them (11,642 million, 6.5% of the total) carry no supplier at all, and 538 — 11,641 million — belong to the state telecom company: 99.5% of everything it published as awarded that year (11,697 million, the State's third-largest buyer). It is not a hole in our ingestion and it is not new: downloading the agency's official archives (ocds-2021/2024/2025.zip), the government's own data has 593 of 626 records with no winner in 2021, 613 of 615 in 2024 and 587 of 613 in 2025 — the award carries a value and a PDF, but neither suppliers nor items — while the rest of the State adds up to 17, 2 and 1 case respectively. The control closes the question: in the same publication channel (ocid ocds-yfs5dr-i…) during 2025 the state insurance bank publishes a winner in 100% of its 2,806 records, the state bank in 100% of 2,117, Montevideo's city government in 98.1% of 15,457, the state oil company in 97.6% and the water utility in 92.2%; the telecom, in 0% of 587. We opened three official records and the winner is indeed published, but only inside the resolution: i477309 ($781,474,029, exception purchase under art. 33.3, Resol. 1045/2025) and i465874 ($661,791,664) to Compañía Ericsson Uruguay S.A., and i481126 ($366,000,000, public tender P103376/2022, Resol. 371/25) to A y M Santa Lucía Limitada; the procurement pages show the amount and «awarded in full» and nothing else, and the PDF for i477309 is a scanned image with no text layer. Of the 587 resolutions from 2025, 306 are PDF, 256 TXT, 2 ZIP and 23 purchases have no document at all (516 million pesos plus 5.27 million dollars), and there the winner is published nowhere on the agency's site: the largest is i477201, public tender P103975/2023 for $219,600,000, whose record offers no resolution file (we tried .pdf, .PDF, .txt and .zip: 404).
- Which law it would engage
TOCAF (Decreto 150/012), article 50, second paragraph, verbatim: «Todas las administraciones públicas estatales deberán dar publicidad en el sitio web de compras y contrataciones estatales, al acto de adjudicación, declaración de desierta o de rechazo de ofertas, a todos sus procedimientos de contratación de monto superior al 20% (veinte por ciento) del límite de su procedimiento de compra directa, incluidos los realizados por mecanismos de excepción, así como las ampliaciones y los actos de reiteración de gastos observados por el Tribunal de Cuentas, en la forma que disponga la reglamentación. Estos organismos contarán para ello con un plazo de diez días luego de producido el acto que se informa». And Law 18,381 on access to public information, article 5 item E, active transparency, verbatim: «Concesiones, licitaciones, permisos o autorizaciones otorgadas, especificando los titulares o beneficiarios de éstos». How they play: for the 564 purchases where the telecom did publish the resolution, TOCAF 50 is probably satisfied — what the rule requires is publicity of the act, and the resolution names the winner — and what is engaged there is art. 5 item E of Law 18,381, identifying the beneficiary rather than leaving it inside a scanned image, plus the agency's OCDS publication policy. For the 23 purchases from 2025 with no supplier and no resolution file at all — all far above 20% of any direct-purchase ceiling, the smallest being $16.1 million — TOCAF article 50 is directly engaged.
TOCAF Decreto 150/012 art. 50 (redacción dada por el art. 58 de la Ley 20.212 de 6/11/2023; fuente: art. 31 de la Ley 18.834). Ley 18.381 de 17/10/2008, art. 5 lit. E. Competent bodies: ARCE (runs the site and the OCDS publication), Unidad de Acceso a la Información Pública (UAIP/AGESIC) under Law 18,381, and the Tribunal de Cuentas, which per the resolutions themselves reviews these awards in advance.
- What is missing before asserting it
- Five concrete things, none of them a bare «let the body explain it». (1) An answer from the telecom and the agency to one specific question: why the telecom's integration with the procurement system sends header, amount and PDF but not the winner nor the items, since when, and whether they will backfill the 2014-2025 records. (2) Knowing whether the practice is still alive in 2026: we cannot claim it ended, because the 2026 annual archive is only published in January 2027 and our live scraper does not capture the «i…» ids; all we see is that the telecom's structured channel grew (24 awards with a supplier in December 2025, 184 so far in 2026 worth 2,327 million, including an exception purchase of USD 15,280,500 where the record does show «Proveedores participantes: KYNDRYL URUGUAY S.A. RUT 218778790016» and items with unit prices). Someone has to request the 2026 list from the agency or wait for the archive. (3) For the 23 purchases with no resolution (516 million pesos plus 5.27 million dollars), a freedom-of-information request for the award decision and the winner's name in each. (4) To say how much each company took someone has to read the 587 resolutions: we checked 3, and at least one is a scan with no text layer, meaning OCR and human review are needed; today we cannot claim concentration by supplier, only absence of supplier. (5) A warning about our own database, which the story has to carry: our corpus shows an amount of 0 for these pre-2025 telecom records, so our own series understate the telecom in 2014-2024; we measured the 2024 and 2021 amounts in the agency's raw archive, not in our database, and before publishing them the ingestion has to be fixed to read awards[].value when there are no lines.
- Who has to answer
- The telecom: its contracting division and board, which sign these resolutions, and its public information access unit. The state procurement agency, which runs the procurement site and the OCDS publication: why it accepts award records from a body with no winner and no items, and what it checks on receiving them. Subsidiarily, the public information access unit (active transparency, art. 5 item E of Law 18,381) and the Court of Auditors, which reviews these awards in advance and therefore does know the winners.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-antel-sin-adjudicatario.verify.ts
9 / 32
The amount of the state telecom's purchases has been on the State's website since at least 2010 and entered the open data on 2 January 2025
- What we measured
- Since 1 January 2025 there are 748 awards by the state telecom company with an amount in the corpus, worth 14,024,289,516 pesos (11,697,454,124 in 2025 and 2,326,835,392 in 2026 up to 13 August). Within the comparable channel — the monthly OCDS files a-MM-YYYY.json, same file and same reader — between 2008 and 2024 there were 2,362 telecom records and only 17 carried an amount (0.7%); in 2025 it was 499 of 533 (93.6%), worth 11,396 million pesos. The series starts on 2 January 2025, the first working day of the year and two months before the change of government on 1 March, so it does not coincide with the change of administration but with the calendar break. The amount, however, was never hidden: we opened 14 telecom award records on the agency's site between 2010 and 2024 and all 14 show «Monto Total de la Compra» (the 2018 one, i220663, reads $ 499.006,88; the 2024 one, i405283, reads U$S 20.326,00). What was missing was in the open data, not on the web, and 2,882 telecom awards from before 2025 remain with no amount, 2,345 of them inside the comparable channel. The opening is also half-done: 473 of those 499 awards from 2025 (11,339 million, 99.5% of the money) name no supplier and detail no items.
- Which law it would engage
No law is broken, and that has to be said up front. Article 50 of the TOCAF (Decreto 150/012) requires publicity on the state procurement website for the award decision, the declaration of a failed tender or the rejection of offers in every procedure above 20% of the body's direct-purchase ceiling, «incluidos los realizados por mecanismos de excepción», within ten days of the decision; the telecom complied: the records exist, are published and show the amount, which we checked on 14 of 14 pages between 2010 and 2024. Article 50 requires publishing the act, not delivering the amount as structured data in the open-data archive, which is exactly the only thing that was missing. There is also an express reason that works in the telecom's favour: the declared scope of Uruguay's OCDS publication excludes bodies operating in competitive markets — «Bodies that operate within competitive markets have the right to withhold information» — and gives as its literal example «purchases of the public cellular telephone company». The only rule left standing as an open question, not as a breach, is article 33 item 3 of the TOCAF, the ground the telecom invokes in the largest purchase of the period ($781,474,029, ocds-yfs5dr-i477309): «Para adquirir bienes o contratar servicios cuya fabricación o suministro sea exclusiva de quienes tengan privilegio para ello... La marca de fábrica no constituye por sí causal de exclusividad, salvo que técnicamente se demuestre que no hay sustitutos convenientes».
TOCAF, Decreto 150/012, art. 50 (publicidad del acto de adjudicación) and art. 33 numeral 3 (compra por excepción por exclusividad, quoted verbatim on the official record of purchase i477309). Scope and exclusions of Uruguay's OCDS publication, including the reservation for bodies in competitive markets and the express mention of the state cellular telephone company: OCP Data Registry, Uruguay / Agencia Reguladora de Compras Estatales. Bodies competent to answer: ARCE and ANTEL. Active transparency framework: Ley 18.381 and the Unidad de Acceso a la Información Pública (UAIP).
- What is missing before asserting it
- Four concrete things, none of them a bare «let the body explain it». (1) How big the hole is: we know 2,882 telecom awards from before 2025 have no amount in the open data and that the amount is on each one's web record, but nobody ever added them up; someone has to scrape those 2,882 records (the «Monto Total de la Compra» field, the same one our correct-lumpsum-artifacts job already scrapes). Without that we cannot claim the telecom «looked small»: we know the datum was missing, not what it was worth. (2) What happened on 2 January 2025: it is a calendar cut-off date and it affects only the telecom among the bodies publishing through the same route; there are two possible causes we cannot tell apart from outside — that the telecom changed how it enters the award in the procurement system or integrated a new one, or that it decided to lift the commercial reservation the OCDS scope itself allows it — and the difference between the two changes the meaning of the story. (3) Who was paid: 473 awards worth 11,339 million publish the amount but not the winner nor the items, and the name is inside the resolution PDF; that half of the problem is measured separately, in the neighbouring card on awards with no supplier. (4) Whether this continued in 2026: the telecom's 2,326 million from 2026 come from our own web scraping, a channel that only exists since 2025 and is not comparable with the historical series; the a-MM-2026.json files are missing. A technical detail to declare if the aggregate figure is published: part of the 14,024 million are awards in dollars converted to pesos with a single recent exchange rate, not with the rate of each award's day.
- Who has to answer
- The telecom, at two levels: the supplies / purchasing division, which enters the award in the procurement system, and the board, which can say whether there was a decision to lift the commercial information reservation or whether it was a system change. The concrete question: what changed on 2 January 2025 so that the amount began travelling in the open-data archive, and why is the winning supplier still omitted in 473 of 499 awards when the insurance bank, the state bank, Montevideo's city government and the water utility always publish it? The state procurement agency, which builds and publishes the OCDS export: why did the amount its own site shows on each purchase record not travel to the downloadable archive in the telecom's case, and was the telecom covered by the reservation for bodies under free competition that the declared scope of Uruguay's OCDS contemplates? As a third addressee, if the agency answers that it was a reservation invoked by the body, the public information access unit is the one that can say whether that reservation was formally classified.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-antel-monto.verify.ts
10 / 32
The social security bank bought 1,457 wheelchairs one tender at a time and the same firm won 94% of them
- What we measured
- The social security bank (buyer.id 28-1) has 1,457 awarded purchases with wheelchair items between 2016 and 2026, worth 289,630,923 pesos; 1,369 of them (94.0%) and 255,782,057 pesos (88.3%, apportioning the purchases with more than one winner) went to MEDI IMPORT S.A. (tax id 212291040019). These are not bulk purchases: 1,205 of the 1,216 purchases under catalogue code 15230 are for a single chair, and of the 1,125 purchases where the corpus records the method, 1,117 are price competitions (5 direct purchases, 3 abbreviated tenders), a more competitive procedure than the legal ceiling would require of it. Two things run against the finding and belong here: today there is competition and the price fell. Of the 129 tenders where the government record published the «participating suppliers» block, there was a single bidder in 1 of the 65 from 2025 (1.5%) and in 8 of the 64 from 2026 (12.5%), with 2.9 and 2.3 bidders per tender; the second firm by appearances, SAMURIO MOLINA CARMEN YOLANDA, bid 115 times against MEDI IMPORT's 125. And the median unit price MEDI IMPORT charges under code 15230 went from 233,607 pesos in 2022 to 131,148 in 2026.
- Which law it would engage
No law appears to be broken, and that comes first: the procedure the bank uses is the legal one and it is more competitive than the minimum required. The resolutions invoke the price competition of TOCAF article 33 item B «en la redacción dada por el art. 314 de la Ley 19.889», and with chairs from 150,000 to 300,000 pesos the bank could buy directly (item C, ceiling $630,000 at January 2025 prices) and opens a tender anyway. What the finding would engage, if it were confirmed that the specifications were written around one importer's catalogue, is the principle in TOCAF article 149, item B, which orders: «Publicidad, igualdad de los oferentes y concurrencia en los procedimientos competitivos para el llamado y la selección de las ofertas». Competitive access is the legal interest at stake, not the legality of the chosen procedure. In the background: the tender document weights price at 45 of 100 points (40 «lowest weight», 15 warranty period, up to −20 for supplier-registry sanctions) and awards «por renglón a la oferta que obtenga mayor puntaje», a legitimate discretionary decision by the body that the bank should be able to justify when the result is 94% of the purchases going to one supplier.
TOCAF (Decreto 150/012), art. 149 lit. B — principios de publicidad, igualdad de los oferentes y concurrencia; art. 33 lit. B y C — concurso de precios y compra directa, topes según Ley 20.446 de 16/12/2025. Reference tender document read in full: Concurso de Precios ODOO N° 2026-28-CR-10383, BPS Gerencia de Salud, sections 1.3, 1.4 and 1.5. Oversight bodies: Tribunal de Cuentas (reviews the expenditure, per the tender document itself, point 5.2) and ARCE. Benefit framework: Reglamento sobre Ayudas Extraordinarias, R.D. N° 3-33/2015 del BPS.
- What is missing before asserting it
- Five concrete things, none of them a bare «let the body explain it». (1) THE BIDDER HISTORY BEFORE 2025 DOES NOT EXIST IN THE DATABASE: only 129 of the 1,457 purchases have the «participating suppliers» block read, and none is earlier than 2025. So we cannot claim there was less competition before: it is not measured, and not measured is not zero. Closing it requires probing the 2016-2024 tenders with the job that already exists, after checking that no other process is running against the government site. (2) THE PRICES OF THE LOSING OFFERS: the corpus has the winner's price and nothing else, and the agency's online offers are not public; without that one cannot claim MEDI IMPORT wins expensively. It is resolved by reading the award resolutions (all at /Resoluciones/acta_{id}.pdf) or requesting the comparison sheets under Law 18,381. (3) THE PRICE SERIES IS NOT A COMPARABLE PRICE: the catalogue has a single code for every wheelchair, so the fall from 233,607 to 131,148 pesos mixes different models and may reflect a change of mix rather than a real drop. (4) THE 2019-2024 TENDER DOCUMENTS: the corpus holds only 224 of them, all from 2025-2026, so the question of whether the specifications narrowed in some period and widened afterwards cannot be answered without downloading the earlier ones and comparing them. (5) WHETHER MEDI IMPORT IS THE EXCLUSIVE REPRESENTATIVE of the awarded brands (Sunrise Medical/Quickie, Stealth, Leggero, Ki Mobility): it does not say so on its site and we did not find it published; if it is, exclusivity explains almost everything and the finding shrinks to a question about how chairs are prescribed.
- Who has to answer
- The social security bank — health purchasing section, health division, which runs the tenders. The outpatient care unit, which writes the technical specifications for each chair and signs the assessment that rejects offers. The health administration and control division, the competent officer that signs the award resolutions. And the bank's board, which approved the extraordinary aid regulation R.D. 3-33/2015 and the delegation R.D. 21-2/2023. Secondarily, the state procurement agency, on whether a series of more than a thousand price competitions for one unit each is the right way to buy a recurring benefit, or whether an open supplier register is called for, of the kind the bank itself already uses for compression stockings and CPAP machines.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-bps-sillas.verify.ts
11 / 32
The Army awarded 394.4 million pesos in pork to a family-farmer cooperative, in six tenders with three to five bidders each
- What we measured
- The Army General Command (buyer.id 3-4) awarded six public tenders under catalogue article 13960 «CARNE PORCINA (USO HUMANO)» for 394,423,507 pesos before tax — 433,865,858 with the 10% VAT, a figure that matches the totals the State itself publishes on each record — and 1,874,338 kilos, always to the same winner: the Uruguayan Pig Producers' Agrarian Cooperative (Caluprocerd), tax id 080194250014. The six tenders are 1111858, 1112103, 1187731, 1190452, 1287113 and 1287125, and each had between three and five bidders according to the record's «participating suppliers» block: Arroyal S.A. bid on all six, Junpol S.A. on four, and Sanquilco, Heba Ltda., Logística 21 SAS, Sardina Bazán Cecilia and Ocaño Virginia also competed. Across the whole State since 2018, that cooperative holds 926,770,715 pesos of the 1,076,373,124 spent on that same article (86.1% of the money and 84.2% of the kilos), against another 66 suppliers spread over 42 buying bodies, almost all of hospital scale. The Army's price is flat — 213.80 $/kg in one of the large tenders and 213.40 $/kg in the equivalent one a year later, that is, a fall in real terms — and in September 2025 the national rehabilitation institute bought 600,000 kilos from the same cooperative at 183.00 $/kg, though of «whole cut» rather than quarters. What did move is the volume: the Army went from 55,000 kilos awarded in 2022 to 485,690 in 2023 and to 665,814 contracted at the end of 2025 for the 2026 supply.
- Which law it would engage
No law appears to be broken by what we measured, and it is worth saying first: the law that exists pushes in exactly this direction. Law 19,292 (2014), article 3, establishes «un mecanismo de reserva mínima de mercado del 30% (treinta por ciento) para las compras centralizadas y del 100% (cien por ciento) para las no centralizadas» of food goods, «siempre que exista oferta», in favour of Qualified Organisations, which article 5 defines as any organisation «integrada por al menos cinco productores agropecuarios, de los cuales como mínimo el 70% deben ser productores familiares agropecuarios y/o pescadores artesanales». Caluprocerd — some 300 family producers based in San Jacinto, Canelones — is a qualified organisation under that regime. The parallel purchase by the rehabilitation institute rests on another express legal route: TOCAF article 33 item 16, «Adquisición de alimentos de producción nacional y de víveres frescos existentes en mercados, ferias o directamente a productores», which enables direct purchase by exception. The only rule that would come into play if a new fact appeared — that the private competitors' offers were coordinated — is Law 18,159 on competition, article 4, which forbids bid rigging: we have not a single indication of that, it is the hypothesis that would have to be ruled out, not an accusation.
Ley 19.292 de 16/12/2014, arts. 3 y 5, reglamentada por el Decreto 86/015; operating procedure in the procurement agency's manual «Producción Familiar Agropecuaria y Pesca Artesanal». TOCAF (Decreto 150/012), art. 33 num. 16, invoked in the rehabilitation institute's Compra por Excepción 6/2025. Ley 18.159, art. 4 — competent body: Comisión de Promoción y Defensa de la Competencia (MEF). Register of Qualified Organisations: MGAP, Dirección General de Desarrollo Rural.
- What is missing before asserting it
- Four concrete things, none of them a bare «let the body explain it». (1) The losers' offers: the award resolutions are PDFs scanned as images, so there is no reading what Arroyal, Junpol, Heba and Logística 21 bid; without that one cannot claim the cooperative won on price nor rule out that the others were excluded on formal grounds. It is resolved with OCR on those six PDFs or by requesting the comparison sheets under Law 18,381. (2) An external reference price per kilo of pork quarters for 2024-2025 (from the meat institute, the agriculture ministry or the statistics office): the 2025 agricultural yearbook on the pork chain is published as a graphic PDF and does not deliver the series, and without that figure saying 213 $/kg is expensive or cheap is opinion. (3) The explanation for the jump in volume: why the quartermaster service went from 55 tonnes awarded in 2022 to 666 contracted for 2026, how many daily rations that covers, for which units, and whether it absorbed the supply of other departments; with the Army's headcount (on the order of 16,000 people) those 666 tonnes work out to some 40 kilos of pork per person per year. (4) Whether the tender documents for these six invoked the Law 19,292 reservation or whether the cooperative won in open competition without invoking it: the documents are in .docx and were not read. A methodological detail that also has to be declared: the cooperative appears in the corpus under two spellings of the same name, so grouping has to be by tax id; grouping by name drops its share to 56.2% and the figure would be false.
- Who has to answer
- The Army General Command, quartermaster service — the release itself carries the buyer's contact details. On policy and volume, the Ministry of National Defence. On the price-per-kilo contrast between bodies, the national rehabilitation institute and the Ministry of the Interior, which through exception purchase 6/2025 paid 183.00 $/kg to the same cooperative. On the reservation regime and the qualified-organisation status, the agriculture ministry's rural development directorate and the procurement agency. And for the right of reply, the cooperative itself: Fernando Andrade chairs both the producers' association and Caluprocerd and has already spoken publicly about it — in a Radio Carve interview on 30 September 2025 he said that «más del 60% de la producción de la cooperativa se destina al Ministerio de Defensa y al Ministerio del Interior», a press statement we did not measure ourselves.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-ejercito-cerdo.verify.ts
12 / 32
A single distributor won all five of the food institute's food-parcel tenders since 2023, and the real price per parcel fell 17.6%
- What we measured
- Between 23/02/2023 and 14/05/2026 the National Food Institute (buyer.id 15-3, social development ministry) awarded catalogue code 31999 «CANASTA DE VIVERES SECOS» eight times for $193,095,700 before tax (187,400 parcels), and all eight times to the same supplier: DISTRIBUIDORA SANTA ANA S.A., tax id 211603980013. Those eight awards correspond to five procedures — abbreviated tender 15/2023, abbreviated tender 41/2023 (coeliac parcels), public tender 14/2024, exception direct purchase 173/2024 (coeliac parcels) and public tender 1/2026 —; the other three are second resolutions on the same tender, with identical unit prices, worth $58,135,850 (30.1% of the money) and with no new call. There was a competitive call in four of the five procedures and, read on each one's record, between 3 and 5 bidders appeared: 5 in tender 15/2023, 4 in tender 41/2023, 3 in public tender 14/2024, 3 in the exception purchase and 5 in public tender 1/2026. The price of the emergency parcel — with composition identical product by product, 11 items in the same quantities, verified in the 2023, 2024 and 2026 tender documents — fell from $1,179 (February 2023) to $1,066 (September 2024) and to $971 (May 2026) in constant pesos of August 2026, deflated by the indexed unit: 17.6% less in three years. In 2021 that same supply was split between the TA TA and Macromercado retail chains and Santa Ana.
- Which law it would engage
No law appears to be broken by what we measured. The only procedure without an open call — exception direct purchase 173/2024, $19,404,000 — invokes in its own tender document «ARTICULO 33 LITERAL D NUMERAL 2 DEL TOCAF», which enables direct contracting «cuando la licitación pública, abreviada o remate resultaren desiertos, o no se presentaren ofertas válidas o admisibles, o que las mismas sean manifiestamente inconvenientes», and requires that «la contratación deberá hacerse con bases y especificaciones idénticas a las del procedimiento fracasado y, en su caso, con invitación a los oferentes originales, además de los que estime necesarios la Administración». All three requirements can be checked from outside: public tender 20/2024 (the same 12,000 coeliac parcels) ended with all offers rejected by Resolution 188/2024 of 29/11/2024; the exception's tender document is titled «COMPRA DIRECTA DE EXCEPCIÓN DERIVADA DE LA LP 20/2024» and repeats the parcel composition item by item (14 products, same net contents); and three bidders appeared (Santa Ana, JUNPOL, SANQUILCO), not one. The other rule engaged, if anyone wanted to question the three second resolutions for the same amount, is TOCAF article 74 on contract extensions, which we could not read verbatim.
TOCAF (Decreto 150/012), art. 33 literal D numeral 2, invoked verbatim in the exception purchase's tender document (pedido_1206642.pdf). Art. 74 del TOCAF for the extensions. Competent bodies: Agencia Reguladora de Compras Estatales (ARCE) and Tribunal de Cuentas de la República (preventive expenditure oversight).
- What is missing before asserting it
- What is missing is the ability to read three documents that exist but are PDFs scanned without a text layer, and without them nothing can be asserted about the one doubtful stretch. (1) Resolution 188/2024 of 29/11/2024, which rejected every offer in public tender 20/2024 for coeliac parcels: the resolution downloaded from comprasestatales.gub.uy/Resoluciones/acta_1180071.pdf has 2 pages and exactly 1 extractable character, and without that text there is no knowing whether the offers were rejected for technical non-compliance or for a «manifestly inconvenient» price — that difference decides whether paying $1,617 per parcel a month later was the result of a thin market or the endorsement of a price the body had just deemed inconvenient. (2) For the same reason the losing bidders' prices are missing: the corpus keeps only the winner and the opening records are scans (acta_1206642.pdf: 4 pages, 74 characters), so there is no saying whether Santa Ana won by being the cheapest. (3) It remains to be confirmed that the three repeated awards (releases adjudicacion-1034087, adjudicacion-1100868 and adjudicacion-1252900, same unit prices as the original, between 2 and 10 months later) are article 74 extensions and not republications of the same expenditure: the record for purchase 1017145 states in text that «la compra incluyó una ampliación el 10/04/2023 por el mismo monto», but the other two only show a second resolution (175/2023 and 137/2025) on the same tender. If they were republications, the total drops from $193.1M to $135.0M. Besides, 2026 is incomplete in the corpus, so what the parcels represent of the institute's spending that year does not compare against previous years.
- Who has to answer
- The food institute's supply office and its directorate, on two concrete things: why every offer in public tender 20/2024 was rejected and at what prices, and whether the three second resolutions are article 74 extensions. The state procurement agency, for two publication problems this case exposes: the resolutions deciding awards of this size are published as scanned images with no text, and the OCDS feed lost the buyer in five 2021 parcel awards worth $336,257,896 (releases adjudicacion-856765, -865232, -868514, -893194 and -878322, all with no buyer field) that the government record does attribute to the institute.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-inda-canastas.verify.ts
13 / 32
The Maciel Hospital buys enoxaparin in direct purchases that land between 92% and 98.5% of the legal ceiling, and when the ceiling rose 2.5-fold the purchase rose 2.4-fold
- What we measured
- The Maciel Hospital (public health administration, buyer.id 29-5) has 342 awards of procurement article 69888 (injectable enoxaparin) between 2022 and 2026, the range where the feed publishes direct-purchase calls: 174 direct purchases with their own call and 168 contract extensions that the agency publishes under a different purchase id. The median of each direct purchase, measured with tax as TOCAF article 156 requires — a factor of 1.122, measured against the official page's «Monto Total de la Compra» and not assumed —, came to 91.8% of the direct-purchase ceiling in 2022, 97.2% in 2023, 96.6% in 2024, 98.5% in 2025 and 96.8% in 2026, with the ceiling moving from $220,000 to $654,000. On 1/1/2026 the ceiling went from $263,000 to $654,000 (×2.49, Law 20,446 art. 28) and the hospital's median enoxaparin purchase went from $230,868 to $564,344 (×2.44), at the same unit price ($206.70 per syringe) and from 1,100 to 2,500 syringes per purchase; against the official source, direct purchase 2903/2025 was $255,109.14 against a ceiling of $263,000 and 504/2026 was $649,368.72 against $654,000. Across the whole hospital, 403 of its 682 own purchases in 2026 (59%) exceed the ceiling that applied in 2025. In 2025 the hospital spent $20,130,820 before tax on this article alone, above that year's abbreviated-tender ceiling ($13,166,000), and the two competitive routes it attempted did not close: public tender 6028/2021 ended in a cancelled award on 8/8/2022 and public tender 9/2024 «Suministro de Enoxaparinas de 40mg y 60mg», published on 10/12/2024, only has an award record dated 30/7/2026 and with no winner.
- Which law it would engage
In Uruguay splitting a contract is not illegal in itself, and that is decisive for the framing. The TOCAF, article 43, reads: «Los ordenadores de gastos adoptarán las medidas necesarias para contratar los suministros o servicios por grupos de artículos o servicios, de forma de facilitar la presentación del mayor número posible de oferentes. (…) Los ordenadores, bajo su responsabilidad, podrán fraccionar las compras dejando expresa constancia de su fundamento y de su conveniencia para el servicio. Cuando el Tribunal de Cuentas observe reiteradamente el fraccionamiento, sin que se corrija tal situación, podrá suspender la facultad establecida en el inciso anterior a los ordenadores responsables y, de corresponder, a los organismos involucrados dando cuenta a la Asamblea General o a la Junta Departamental que corresponda. A los efectos de dicho control, no se considerará fraccionamiento de compra la adquisición de bienes o servicios, cuando el mismo se integre en un proceso de compra centralizada efectuado por la Agencia Reguladora de Compras Estatales o cuando la compra se realice mediante la aplicación de un convenio marco». It is a power of the spending officer conditioned on leaving a written record of the grounds, and the sanction provided is not nullity but that the Court of Auditors withdraws the power. What is compromised is article 33 item C (the direct-purchase ceiling) and the doctrine the procurement agency teaches in its own training material: «Si se constata que el fraccionamiento es artificial para que encuadre en determinados límites de contratación, se considera falta grave (Art. 32)». The measurement does not prove artificiality: it proves the size of the purchase is determined by the ceiling.
TOCAF (Decreto 150/012), art. 43 (fraccionamiento), art. 33 lit. C (tope de compra directa común) and art. 156 (amounts are adjusted by CPI and «para la determinación del monto de cada gasto se incluirá el Impuesto al Valor Agregado»). Yearly ceilings: statistics institute, «Montos Límites de las Adquisiciones Estatales», 2022 to 2026 editions. 2026 ceiling raised by Ley 20.446 art. 28. Body competent to flag contract splitting: Tribunal de Cuentas de la República.
- What is missing before asserting it
- Four concrete things, none of them in the open data. (1) The article 43 records: the law requires that in each split purchase the spending officer leave «expresa constancia de su fundamento y de su conveniencia para el servicio», and those records are neither in the OCDS feed nor on the purchase's public page; without reading them there is no saying whether the splitting is the kind the law authorises or the «artificial» kind the agency calls a serious offence. They have to be requested under Law 18,381 from the hospital and the health administration: the files of the 2022-2026 enoxaparin direct purchases, with the officer's resolution and its grounds. (2) What happened to public tender 9/2024: it was published on 10/12/2024 for 170,000 syringes and its award record only appears on 30/7/2026, empty; if it was stuck for 19 months, the direct purchases are the unavoidable bridge and the story changes subject. (3) Whether enoxaparin fell inside a centralised purchase or a framework agreement at any point in 2021-2026, because article 43 itself says that in that case «no se considerará fraccionamiento»: the corpus holds not one record of 69888 with a centralised buyer or with the «Convenio Marco» method, but the feed may not publish the centralised process. (4) Whether the Court of Auditors has already flagged these specific purchases: its resolutions are not in the corpus, they have to be looked up on tcr.gub.uy by officer and period. And a limitation of the measurement: 2021 does not enter the series because the feed carries only 1 own call for 82 awards that year, and 2026 is measured to August and with the feed incomplete in other channels.
- Who has to answer
- 1) The Maciel Hospital — purchasing / materials department, the contact the call itself publishes. Question: what are the grounds and the convenience recorded in writing, purchase by purchase, for buying enoxaparin 174 times by direct route instead of tendering it? 2) The public health administration — materials directorate and board: why does public tender 9/2024 for enoxaparin still have no winner 19 months after publication, and what was bought in the meantime? 3) The Court of Auditors, the body article 43 designates to oversee contract splitting and which already flagged the health administration's southern region for it — as the weekly Búsqueda reported on 14/12/2023, a press fact and not a measurement of ours —: are these purchases among those flagged and what became of those findings? 4) The state procurement agency: why is an anticoagulant of daily and predictable consumption, which a single hospital buys for over $20 million a year, neither in a centralised purchase nor in a framework agreement, the two figures article 43 says are not contract splitting?
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-maciel-enoxaparina.verify.ts
14 / 32
The power utility's purchase of drivers is published at $14,964 million and the resolution approving it says $32.9 million
- What we measured
- On 21/05/2026 the state procurement portal published award 1315467 by the state power utility (abbreviated tender 103519/2026, provision of staff for driver services, to ESTILO S.R.L., tax id 215529660017) with «Cantidad: 27.000.000,00 UNIDAD» at «Precio unitario sin impuestos: $ 454,29» and a «Monto Total de la Compra: $ 14.964.312.600,00»; our corpus stores it as $12,265,830,000 before VAT. The resolution approving that same purchase — acta_1315467.pdf, Resolution G.R. N°123/26 of 13/04/2026, linked from that very page — says «Valor hora/persona $ 454,29» and «Monto total a adjudicar por el periodo de contratación ITEM 1 Y 2: $ 27.000.000,00 / IVA 22% $ 5.940.000,00 / TOTAL incluido IVA $ 32.940.000,00»: what is published is exactly 454.29 times what was awarded, and 454.29 is the unit price, because the «quantity» field carries the total amount in pesos and the portal multiplies it by the hourly rate. There are 12 utility records from 2026 with that pattern (driver, administrative staff, security, cleaning) worth $29,485,387,226 as published; we downloaded the resolutions of 8 of them: $29,005,044,875 published against $67,278,080 in the resolutions, a factor of 431 and $28,937,766,795 of spending that does not exist. With that, 74% of everything the utility reports having bought in 2026 — $39,221,265,989 across 617 records — is a mis-entered field: corrected it comes to $10,283,499,194 and the utility stops being the year's largest buyer (the state oil company, $12,443,520,232). None of those records is in our anomalies collection: the 1,945 anomalies from 2026 are all price spikes and $454.29 an hour for a driver is a normal price, because our detector scores the unit price and uses quantity only to suppress false positives, never to detect. The utility itself uses the harmless convention in 2026 — unit price 1, quantity in pesos — in 360 lines worth $967,011,833.
- Which law it would engage
No law penalises a mistyped number, and it has to be put that way. What is engaged is the duty of publicity: TOCAF (Decreto 150/012) art. 50, verbatim: «Todas las administraciones públicas estatales deberán dar publicidad en el sitio web de compras y contrataciones estatales, al acto de adjudicación, declaración de desierta o de rechazo de ofertas, a todos sus procedimientos de contratación de monto superior al 20% (veinte por ciento) del límite de su procedimiento de compra directa, incluidos los realizados por mecanismos de excepción, así como las ampliaciones y los actos de reiteración de gastos observados por el Tribunal de Cuentas, en la forma que disponga la reglamentación. Estos organismos contarán para ello con un plazo de diez días luego de producido el acto que se informa». What is published has to be the act, and here the act says $32,940,000 and the site says $14,964,312,600. The second point is that the published record, read literally, describes an impossible procedure: it is an abbreviated tender, and the 2026 ceiling is $13,705,000 (art. 33 item A) or $68,525,000 under the extended regime of art. 44 that applies to an autonomous entity like the utility, so what is published ($12,265,830,000 before VAT) is 179 times the extended ceiling. What was actually awarded ($27,000,000) sits comfortably within the ceiling: the purchase is legal, the published figure is the one that cannot exist.
TOCAF, Decreto 150/012, art. 50 (fuente: Ley 18.834 art. 31, en la redacción del art. 58 de la Ley 19.889). Ceilings: statistics institute, «Montos Límites de las Adquisiciones Estatales, Enero–Diciembre 2026» (30/12/2025, under Ley 20.446 of 16/12/2025), item h) abbreviated tender $13,705,000 and item i) extended abbreviated tender art. 44 $68,525,000. Body competent over the system where the datum was entered: Agencia Reguladora de Compras Estatales (ARCE), owner of the SICE. Expenditure reiteration: TOCAF art. 114.
- What is missing before asserting it
- Four concrete things, none of them «let somebody explain it in general». (1) That the procurement agency answer whether the system has a field for «amount awarded» when a purchase is awarded by amount rather than by quantity — the resolution says «12 meses … o hasta agotar el monto adjudicado, lo que ocurra primero» and payment is by hours actually worked — and what the current instruction is for that case; without it we do not know whether the utility's operator improvised or followed the only way out the form offered. (2) That the agency say whether it will correct the 12 records and regenerate the OCDS feed, or whether the figure stays published: today anyone downloading the 2026 open data takes away the $28,937 million that do not exist. (3) Whether the official 2026 procurement statistics the agency publishes are computed from that same field, because then the official figure for the year's state spending is inflated too; that is asked through a freedom-of-information request, not by observation. (4) We are missing 4 of the 12 records in the class (adjudicacion-1103510, 1103395, 1158012 and 1335094): we did not download their resolutions, which is why the figure we publish is the one for the 8 verified. And one more piece of honesty: the resolution sets a spending cap, not an execution, so we do not know how much the utility actually paid ESTILO S.R.L. and we do not assert it. The percentages against the 2026 national total also carry the warning that the year is distorted in the feed.
- Who has to answer
- The state procurement agency, owner of the procurement system, which computes and publishes the «Monto Total de la Compra» and produces the open-data feed: it falls to it to say whether the system allows entering an awarded amount without inventing a quantity, and whether it will correct the records. The power utility, purchasing and contracting sub-management and the supplies and services division — the requesting unit the resolution itself names —, which entered the datum and which in 2026 uses both conventions at once. As context and not as a complaint: the Court of Auditors' delegate accountant at the utility already intervened in this purchase — flagging the expenditure for lack of budget availability, which the utility reiterated by Resolution G.R. N°124/26 under TOCAF article 114 —, meaning the oversight chain worked on the correct figure. The only place the $14,964 million appear is the publication.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-ute-choferes.verify.ts
15 / 32
The Budget repealed the procurement agency's price cap for buying food from family-farmer organisations: seven months in, the price paid is still below the market's
- What we measured
- Article 29 of Law 20,446 repealed the last paragraph of item 16) of letter D) of TOCAF article 33, which required that the price paid not exceed the one published by the procurement agency; that item does not cover direct food purchases in general, but those of nationally produced food and fresh provisions offered by family producers, worker cooperatives and organisations qualified under Law 19,292. We identified 33 tax ids of rural development societies and agrarian cooperatives with 4,357 award lines in the corpus, and for each line we computed the unit price divided by the market median for the same catalogue article, the same unit and the same half-year, excluding code 68715 («fruit and vegetable basket»), whose unit is not comparable. The median ratio was 0.905 in 2024-H1; 0.879 in 2024-H2; 0.886 in 2025-H1; 0.941 in 2025-H2; and 0.919 in 2026-H1 (n=105), the first full half-year with the rule in force. Aggregated it gives 0.916 before (July 2024 to December 2025, n=248) against 0.874 after (January to August 2026, n=127): the difference is −0.04 and a permutation test of 20,000 resamples gives p≈0.5, that is, indistinguishable from noise. The volume did not move either: 62 purchases from those organisations in 2025-H1, 79 in 2025-H2 and 49 in 2026-H1. Measured with data to 13/08/2026; 2026-H2 is incomplete (n=22) and must not be read as a half-year.
- Which law it would engage
No law is broken: the State may today legally pay above the agency's reference price in that channel, which is precisely what changed. The repealed text, TOCAF art. 33, item D), num. 16), last paragraph, read: «En cualquier caso, los precios a pagar no podrán superar los precios publicados por la Agencia Reguladora de Compras Estatales para ese producto». The item that contained it remains in force and describes the channel: «La adquisición de alimentos de producción nacional y de víveres frescos por parte del Poder Ejecutivo y los organismos comprendidos en el artículo 220 de la Constitución de la República y los Gobiernos Departamentales, con la finalidad de abastecer a sus dependencias, que sean ofrecidos directamente por: i) productores familiares (…); ii) cooperativas de trabajo (…); iii) toda organización habilitada creada al amparo del artículo 5º de la Ley Nº 19.292 (…)». The repeal is article 29 of Law 20,446: «Derógase el último inciso del numeral 16) del literal D) del artículo 482 de la Ley Nº 15.903, de 10 de noviembre de 1987, en la redacción dada por el artículo 35 de la Ley Nº 19.996». What would be engaged if the price shot up is not an illegality but article 3 of Law 19,292, which still empowers the Executive to set maximum-price conditions for the market reservation: a duty to regulate that today is not exercised for most fresh provisions.
TOCAF 2012 (Decreto 150/012), art. 33 lit. D num. 16, texto dado por el art. 35 de la Ley 19.996 del 3/11/2021. Ley 20.446 (Presupuesto Nacional 2025-2029), art. 29, enacted on 16/12/2025, published in the Official Gazette on 8/1/2026 and in force since 1/1/2026 under its art. 3. Ley 19.292 (Producción Familiar Agropecuaria y Pesca Artesanal), arts. 3 y 5. Competent bodies: ARCE (publishes the reference prices; for instance, MEF resolution 162/024 for fresh milk, agreed with the dairy industry chamber) and MGAP–DGDR (Registro Nacional de Organizaciones Habilitadas).
- What is missing before asserting it
- Four concrete things, none of them «let somebody explain it». (1) The agency's reference price series by article and date: it was the literal legal ceiling and it does not exist as open data — the agency publishes them as scattered announcements and finance ministry resolutions, and only for a few products —, so we are measuring against the market median, which is a proxy and not the ceiling; it has to be requested through freedom of information: reference prices published by article between 2023 and 2026. (2) The roll of the agriculture ministry's National Register of Qualified Organisations: our universe is 33 tax ids identified by the supplier's name, so it probably includes organisations that are not on the roll (CALNU and CALPRYCA are large agro-industries, not family producers) and misses qualified ones under another name. (3) The exception's item number, purchase by purchase: the OCDS feed only says «Compra por Excepción», never the article nor the item; of the purchases from those organisations since 2024, only 27 appear as exception purchases and most have no sibling tender release, so we cannot prove that any was processed under art. 33 item D num. 16 — it is inferred from supplier and product — and it is resolved by scraping the purchase's HTML record, which does show the grounds. (4) Time: seven and a half months in force and 127 lines after the cut-off; with that n, a change smaller than ~15% in the median cannot be told from noise, so the measurement has to be repeated in January 2027, with two full half-years. With this sample one can only assert «no change is detected», never «there was no effect».
- Who has to answer
- The state procurement agency, on two questions: whether it still publishes reference prices for fresh provisions after they stopped being binding, and whether public bodies still use them as a ceiling. The Ministry of Economy and Finance, which drafted article 29: why the cap was repealed and what replaces it. The agriculture ministry's rural development directorate, which keeps the Register of Qualified Organisations and oversees the 30% market reservation of Law 19,292: whether the Executive will exercise the article 3 power to set a maximum price now that the TOCAF ceiling is gone. And the public health administration, the channel's largest buyer: its occupational and psychosocial rehabilitation centre accounts for a large share of the lines bought from these organisations since 2025, far above the Air Force or the Paysandú city government.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-alimentos-tope.verify.ts
16 / 32
The tax office resolves 13% of its purchases on 31 December, and 93.5% of those resolutions are extensions of earlier contracts
- What we measured
- Across 212 public bodies with at least 300 awards between 2019 and 2025, the median signs 0.10% of its purchases with a resolution date of 31 December, and the 97th percentile reaches 1.56%. The Dirección General Impositiva, the tax office, signs 13.03%: 215 of 1,650 awards, and 472 of 1,807 million pesos, 26.1% of its money. The second body on the list stands at 4.64%. Within the tax office's own Decembers, 215 of 465 awards — 46.2% — fall on the 31st; across the rest of the State that share is 2.33%, and December is, on the national average, the month in which the LEAST is resolved on the last day of the month: 2.63%, against 3.08% to 4.93% for the others. We went to see what those 215 records are on the state procurement portal, one by one: 201 of the 215 — 93.5% — carry the «Ver Compra Original» banner, meaning they are extensions of an earlier purchase. What is being extended is an abbreviated tender in 92 cases, an exception purchase in 59, a public tender in 34, a price competition in 10 and a direct purchase in 6: 59 of those 201 extensions prolong a procedure that had already avoided competition. In a control group of the tax office's 100 largest awards that do not carry that date, the extensions number 18. The administrative explanation covers a good part of this and comes before any other reading: service contracts run from 1 January to 31 December, next year's extension has to be resolved before the fiscal year closes because the budget credit that pays for it dies with the year, and 31 December is the last day on which it can be signed. The data points that way: suppliers repeat year after year, and INFOTECH appears for the same amount, US$ 280,600, in 2022, 2023 and 2024. That is the shape of a contract renewal. Two things fall outside that explanation. The first is that the share of the annual value resolved on 31 December went from 7.3% in 2019 to 15.6%, 22.0%, 29.2%, 37.7% and 46.6% in 2024, whereas a stable portfolio of annual contracts would give a stable share. The second is that those resolutions are published with a median delay of 52 days: 142 of the 215 take more than a month, 34 more than six months and 14 more than a year, with a maximum of 1,120 days, while the rest of the tax office's own resolutions are published the same day or the next, and those of the rest of the State the next day. It is the same body, the same system and the same publication channel. Measurement details: the date used is the resolution date (the «Fecha Resolución» field of the official record, `awards[].date` in the feed), present in 789,454 of the period's 793,956 awards; the histogram by day of the year shows everything rests on a single day, with 215 records on 31/12 and 24 on 26/12, the next busiest; and the series is read up to 2024, because in 2025 the tax office shows 2 records dated 31 December and, with a 52-day median delay, those resolutions land in the following year's feed, which in 2026 is incomplete because one channel is missing.
- Which law it would engage
None. Resolving a purchase on 31 December breaks no rule, and extending a contract does not either: the TOCAF allows it and, for a service starting on 1 January, 31 December is the last possible date to sign the extension. What the date reveals is not an infringement but a sequence: the tax office contracts the same supplier again by extending an earlier tender instead of returning to the market, and it does so on the day the fiscal year closes. Two rules would be engaged: the one setting how far an awarded contract may grow before it stops being an extension, and the one defining when an exception purchase applies, because some of these extensions prolong precisely an exception. That is settled contract by contract, comparing the original amount against the accumulated extensions, and it is not settled with the date.
TOCAF (Decreto 150/012), art. 33: defines which procedure applies according to the amount and which are the grounds for exception to tendering. TOCAF, art. 74, on increasing contracted services: contracts may be increased «hasta un máximo del 20% (veinte por ciento)» of their original value and, with the awardee's agreement and under the same pre-established conditions, «en ningún caso los aumentos podrán exceder el 100% (cien por ciento) del objeto del contrato» (impo.com.uy/bases/tocaf2012/150-2012/74). For these cases we did not measure how much each contract accumulated: the record of an extension does not publish the running total, so the figure is neither on the portal nor in the feed.
- What is missing before asserting it
- Four things, and two of them are preconditions for publishing. (1) The prior-press-coverage check, which is not complete: the session in which this was measured ran out of its search budget and the alternative engines returned CAPTCHAs. What we did verify is that the site's own curated news collection has nothing on this, and that our corpus of cases and investigations mentions the tax office only in passing and never as a buyer; before publishing, one must search «DGI ampliación contrato», «DGI licitación informática» and the suppliers' names (ATEL, GXC, INFOTECH, SEVITEC, PULSO) in Búsqueda, la diaria, El Observador and Montevideo Portal. (2) The contract-by-contract verification of the accumulated extensions against the original amount: it is the only thing that can say whether any of these extensions exceeded the ceiling of art. 74, and that figure has to be requested because it is not published. (3) The tax office's answer: there may be a documented administrative explanation — for instance, that extension resolutions are signed in a single year-end act — that reduces the finding to a one-line note, and nobody has asked yet. (4) The years 2025 and 2026: the series is comparable up to 2024 and must be shown that way, because of the 52-day publication delay and the missing channel in the 2026 feed.
- Who has to answer
- The Dirección General Impositiva, the tax office: why 93.5% of what it resolves on the last day of the fiscal year are extensions of earlier purchases, and why the share of the annual value resolved on that day went from 7.3% to 46.6% between 2019 and 2024. And the state procurement agency, which runs the portal where those resolutions are published: why a resolution dated 31 December can take six months — and in fourteen cases more than a year, with a maximum of 1,120 days — to become public, when the same body's other resolutions are published the same day or the next.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-dgi-31dic.verify.ts
17 / 32
The supplier identifier changed shape in 2026 and our site counted 6,039 companies too many
- What we measured
- Our site publishes 43,000 state suppliers. If the identifier's punctuation is ignored, the distinct ones number 36,961: there are 6,039 records too many, 14.0%. The cause is measured and it sits on the source-data side: in 2026 the procurement agency's OCDS feed stopped writing the slash in the supplier identifier. In 2022 and 2023 the form «R/210002980010» was 100.0% of the award-supplier lines; in 2025 the transition begins and it falls to 97.2%; and in 2026, out of 60,732 lines, 13 keep the slash. The result is that 5,987 tax IDs now have two records: the historical «R/…» and the 2026 «R…». The detached record holds 79,990 million pesos; discounting 159 rows whose 2026 average per contract exceeds twenty times their historical average — suspected of the lump-sum amount artefact we have already documented — 39,422 million remain across 5,828 records, and that is the conservative figure we use. Of the 6,527 records with 2026 activity, 5,813 — 89% — duplicate a company that already had one. In the site's supplier ranking, 49 of the first 100 rows are half of a split tax ID: TEYMA URUGUAY appears with 22,022 million when across its two records it is 24,205, and its supplier page, which queries the exact identifier, leaves out the 7 contracts that ended up on the other side. The change was made uniformly: it starts gradually in 2025, in 2026 it is practically total, the tax ID stays complete and the company name is the same in both records — «TEYMA URUGUAY S.A.» and «TEYMA URUGUAY S A» — so anyone who normalises the identifier can rejoin them. On our side the defect is that the rollup groups by the identifier's literal string, which is reasonable as long as the issuer does not change it, and nobody announced that it had been changed. Two other patterns coexist in the corpus and are worth distinguishing from this one, because they are not the same problem: 234 tax IDs with two different names, mostly verifiable corporate changes (BULL URUGUAY became ATOS URUGUAY, SODEXO SVC became PLUXEE, NOLVER became ADIUM); and the fuel traders that appear several times — TRAFIGURA with 77,605 and 8,298 million, VITOL with 51,406 and 17,092 — whose second identifier is an internal sequence number the agency used between 2019 and 2020 before moving to real registry numbers. Three caveats on the figures. The gross 79,990 million are contaminated by the known lump-sum amount artefact, which is why we publish 39,422. In the foreign examples (TRAFIGURA, VITOL) we could not control for the known null-currency error: their amounts may be counted as pesos even though they are quoted in dollars, which affects the magnitude but not the fact that the record is split. And 2026 is distorted by a missing feed channel, so no figure from that year is compared against earlier years.
- Which law it would engage
No rule forbids changing an identifier's format, and it is worth saying up front: there is no wrongdoing here. What there is is a distance from the standard Uruguay committed to publishing. The OCDS provides a `parties[].identifier` block with `scheme` (for instance UY-RUT) and `id` precisely so that a company remains the same company across the years; the Uruguayan feed carries it in none of the corpus's 2,184,332 records, and publishes only the free-text identifier string, which then changed shape without back-filling the historical series. On top of that runs the duty of active transparency: the information the State publishes on its own initiative must be complete and up to date, and an identifier that changes mid-series makes the published information not comparable with itself. The other half of the failure is ours: the site groups by the literal string and that is why it shows 43,000 suppliers where there are 36,961.
Ley 18.381 de Derecho de Acceso a la Información Pública, art. 5 (active transparency: public bodies must publish up-to-date information on their sites). Open Contracting Data Standard 1.1, fields `parties.identifier.scheme` and `parties.identifier.id`, adopted by Uruguay as a commitment of the 3rd Open Government Plan (2016-2018). Decreto 155/013, which creates the state suppliers' registry and identifies each supplier by its document number.
- What is missing before asserting it
- The agency's confirmation that the 2026 format change is deliberate and not an export defect, and whether it plans to back-fill the series or publish the `parties.identifier` block. It is also missing whether the break coincides with the agency's own renaming or with a migration of the procurement system: we looked and found no public announcement of the identifier change. On the figures: the gross 79,990 million are contaminated by the lump-sum amount artefact, which is why the publishable figure is the conservative 39,422 million; we could not control for the null-currency error in the foreign examples (TRAFIGURA, VITOL); and 2026 is distorted by a missing feed channel. And on our side the fix is missing, which is the part that depends only on us: normalising the identifier to digits before grouping, both in the analytics rollup and on the supplier page, which today queries the exact identifier. The story is published together with the correction and not before.
- Who has to answer
- The state procurement agency, which publishes the OCDS feed and runs both the procurement system and the suppliers' registry: why the supplier identifier changed in 2026, whether there was any notice, whether it will back-fill the historical series, and whether it will publish `parties[].identifier` with the UY-RUT scheme as required by the standard Uruguay adopted. And ourselves, who counted 43,000 suppliers where there are 36,961, who show TEYMA URUGUAY with 22,022 million when it is 24,205, and who leave companies without their 2026 activity on their own page. Half of this finding is a defect of ours and we did not see it until we went to count.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-duplicados.verify.ts
18 / 32
The State writes down why it is not tendering on the web record and leaves the field empty in the open data
- What we measured
- Between 2020 and 14 August 2026, 3,732 exception purchases were awarded — the art. 33 TOCAF route for not tendering — for 27,048 million pesos: 5.6% of the money whose purchase method can be identified, and more than the 14,616 million of all ordinary direct purchasing, with 3,732 contracts against 126,844. That total is always shown alongside another: without the largest record (a purchase by the state bank BROU from ATEL, 6,037 million, 55,000 hours at US$ 2,740.93 an hour, a price we did not verify and which may be a peso amount labelled as dollars) the total falls to 21,010 million. Most of that route is not discretionary, and this comes before any other reading. We read 609 official records, one by one: a census of the 120 awards of 20 million or more, plus some 70 drawn at random per year. The most frequent ground, 31.7% of the records, is item 26: purchases the health ministry makes in compliance with court decisions, of medicines and devices not on the therapeutic formulary. That explains the route's largest user, the national health directorate, and the four suppliers that concentrate their billing there, all of which sell medical devices to that same body. The second ground, 19.9%, is item 33: the schools authority maintaining teaching premises. And the one that moves the most money in the census, 42.4%, is item 2: the tender was held and no bids came in. What is not in the open data is the ground. The OCDS field that exists for exactly this, `tender.procurementMethodRationale`, is empty in all 2,184,332 records of the corpus, and so are `tender.legalBasis`, `tender.rationale` and `tender.procurementMethodJustification`. Each purchase's HTML record does print the item number and its legal text: in the 609 we opened it was always published, 584 citing an item of art. 33 and 25 citing another law. There is no way to count the country's urgency purchases without opening pages one at a time. Urgency appears where it is expected and in the proportion that can be measured: only 30 of those 609 records, 4.9%, invoke urgency (art. 33, letter D, item 10), and only 5 of the 120 large awards, for 320 million, 1.7% of that money. The health emergency did not come through here either: of the 204 tender documents from 2020 that name covid, coronavirus, pandemic or health emergency, 172 are direct purchases and only 12 are exception purchases, and 2020 is the year with the lowest exception spending of the whole series, at 918 million. Nor does the word «emergencia» or «urgencia» in the tender text identify the procedure: it appears in 1,466 of 269,788 tender documents since 2020 and describes the thing being bought — emergency medical cover, an emergency door, a resuscitation trolley — not the route. One open fact remains, and it is the same size as the rest. In 2025 the state telecoms company ANTEL concentrated ninety-nine exception awards worth 6,456 million — 23.9% of all exception money in the period — with no awardee either in the open data or on its public record: those purchases reach the feed as a tender with an amount and an empty supplier list, and the five records we opened show neither the participating-suppliers block nor the awarded-items block, only the amount. At least four of them invoke item 5, «obras de arte, científicas o históricas»; among them a three-year VMware licensing purchase, US$ 11,344,906, record 105286/2024. That item has a tail reading «o deban confiarse a empresas o personas especializadas o de probada competencia», so a licence renewal can be framed there even though the heading speaks of works of art.
- Which law it would engage
What is at stake is not an illegality in the purchases but the duty to state reasons and to publish them. The TOCAF requires that the ground for exception be justified «en forma detallada» and be «un aspecto sustancial en la motivación del acto»; for urgency it further requires prior certification by the finance ministry as to the ground being met and as to market prices. That justification exists — each purchase's HTML record prints it — but it does not travel to the open data, even though the OCDS standard Uruguay chose to publish has the exact field for it and it is empty in all 2,184,332 records. We verified no rule specifically obliging the ground to be carried into the OCDS feed, so this is a data-policy gap and not a violation. The 99 ANTEL awards worth 6,456 million with no published awardee are a different matter: there it is fair to ask under what rule the winner's name appears neither in the open data nor on the record.
TOCAF (Decreto 150/012), art. 33, letter D, item 10, verbatim: «Cuando medien probadas razones de urgencia no previsibles o no sea posible la licitación, concurso de precios o remate público, o su realización resienta seriamente el servicio, extremos cuya invocación deberá fundamentarse en forma detallada, constituyendo un aspecto sustancial en la motivación del acto que dispone el procedimiento de excepción» (impo.com.uy/bases/tocaf-tcr/150-2012/33). Item 5, the one invoked by ANTEL: «Para adquirir, ejecutar, restaurar, transportar, montar obras de arte, científicas o históricas, cuando no sea posible el concurso de méritos o antecedentes o deban confiarse a empresas o personas especializadas o de probada competencia». Item 26, the most frequent: «Las compras que realice el Ministerio de Salud Pública, en cumplimiento de decisiones jurisdiccionales, de medicamentos o dispositivos terapéuticos no incluidos en el Formulario Terapéutico de Medicamentos ni en los programas integrales…». The requirement of prior finance-ministry certification for item 10 is published by the procurement agency in its own manual and attributed to Decreto 90/000, art. 11: we took it from the agency's documentation and did not verify the original text. Open Contracting Data Standard 1.1, field `tender.procurementMethodRationale`.
- What is missing before asserting it
- Five things. (1) The count of grounds is a sample and not a census: we read 609 of the roughly 3,750 exception awards with an amount, and to publish a firm national percentage one must read the remaining 3,150, which is about an hour of reading at one request per second and where extraction works 100% of the time. The script accompanying this card redraws 60 records every time it runs, so that the shape of the distribution can be checked without redoing the census. (2) The largest record (BROU-ATEL, 6,037 million, 22% of the total) has 55,000 hours at US$ 2,740.93 an hour; the official record reproduces that multiplication, but the price looks like a peso amount labelled as dollars — 2,741 pesos an hour is about US$ 62 — and it has to be confirmed against the minutes before using the 27,048 million total, which is why the total without that record, 21,010 million, is shown as well. (3) The PDF minutes of ANTEL's 99 purchases from 2025 still have to be opened to learn who collected the 6,456 million, and it must be asked whether the absence of the awardee is a publication rule for bodies operating in competitive markets or an error. (4) The finance ministry's certifications for the 30 purchases with an urgency ground still have to be requested under the access-to-information law: without them nothing can be said about whether urgency was properly invoked. (5) The known null-currency error does not appear in this set — every row carries a declared currency — but part of the amount is in dollars and depends on the exchange rate applied.
- Who has to answer
- The state procurement agency, which publishes the OCDS feed and which prints the ground on the HTML record: why the `tender.procurementMethodRationale` field of the standard it chose goes empty in 2,184,332 records, and whether it intends to fill it. ANTEL: why its 99 exception purchases of 2025, worth 6,456 million, show no awardee either in the open data or on the public record, and on what criterion it framed the VMware licensing purchase of US$ 11,344,906, record 105286/2024, under item 5 — «obras de arte, científicas o históricas». And the Ministry of Economy and Finance: how many of the item-10 urgency purchases arrived with the prior certification the State itself requires.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-excepcion-motivo.verify.ts
19 / 32
Florida recorded 72,580 excavator hours as a direct purchase: 55.7% of everything it bought in 2017
- What we measured
- The Intendencia de Florida, the departmental government, recorded Direct Purchase 213/2017 for 72,580.00 excavator hours at US$ 94.9999 an hour: US$ 6,895,092.74 before tax and US$ 8,412,013.15 as «Monto Total de la Compra» according to the State's own record. The awardee, CYCLATER S.A. (tax ID 217270870013, registered in Cardal, Florida department), appears exactly once as an awardee across the 2.18 million records of the 2002-2026 corpus, and was the only participating supplier. That contract equals 55.7% of everything the departmental government reported buying in 2017 — US$ 12,380,053 across 49 records, each converted at the central bank rate of its own month — and on its own it is 1.26 times all the rest of the year put together, which adds up to US$ 5,484,960. The declared object, verbatim and with its misspellings, is «Contratacion de Horas de Excavadora por deterioro ocacionado producto de las serveras inclemencias del tiempo». The purchase is dated 19/01/2017 and was published on 17/11/2017, ten months later. At 2,000 hours per machine-year, 72,580 hours are some 36 machine-years: consuming them within 2017 would take about 36 excavators working full time all year in a department of some 67,000 inhabitants. There are three administrative explanations and none is ruled out. The first, and the strongest, is a data-entry error: this corpus contains a documented family of records where the quantity or the unit price is entered wrongly and the total inflates by itself, and the site has a page devoted to it. The unit price, US$ 95 an hour, is an ordinary market rate for an excavator; the only thing out of scale is the quantity, and that is why our own anomaly detector did not flag this record: it compares unit prices against their baseline and does not look at quantities. The second, if the amount were real, is the one the record itself states: a storm destroyed rural roads, the departmental government needed machines immediately, and the TOCAF allows contracting without tendering in unforeseen and urgent situations; that would also explain the single participating supplier and the company never appearing again. The third is that 72,580 is a contractual ceiling rather than what was actually consumed. Any of the three defuses the case. This record came out of a sweep of suppliers that appear in a single year of the corpus, and the sweep itself found no pattern: of the 215,076 million pesos billed by the 6,475 single-year suppliers since 2015, the vast majority is explained by two structural facts — the special-purpose vehicles of the road and prison public-private partnerships, and ANCAP's crude and fuel purchases from international traders — the largest of all is an inflated-amount artefact, and only a minimal fraction comes from the direct-award route, almost all of it property purchases, where the exception is the only lawful path because a specific property has a single owner. What survives is this case, not a rule. Two caveats. Every figure in this card is in dollars or converted at the central bank rate of each record's own month: the peso amount in our own database for this release, 276.1 million, is wrong because the converter applied the exchange rate of 10 August 2025 to 2017 dollars, and it must not be cited. And on direct machinery contracting in Florida there is a precedent that is not this case and did not come from our measurement: El Observador reported in 2023 that the departmental government made repeated direct purchases of a backhoe from a company linked to the mayor's secretary, with several observations from the audit court. It is another company, other amounts and another administration.
- Which law it would engage
TOCAF (Decreto 150/012), art. 33: contracting is done by public tender unless the amount falls below the thresholds of the simplified procedure or an express ground for exception applies. For 2017 the ordinary direct-purchase ceiling was $410,000 and the abbreviated-tender ceiling $8,199,000. The US$ 6,895,092.74 before tax equal some $201.6 million at the November 2017 rate (29.231 pesos to the dollar): on the order of 490 times the direct-purchase ceiling and 25 times the abbreviated-tender one. Even if departmental governments were covered by a doubled ceiling, the difference would still be two orders of magnitude. What would be engaged, if it were confirmed that the amount executed is the amount recorded, is the procedure: a purchase of that size required a public tender, and if it relied on an emergency ground it had to be processed and labelled as an exception, with a reasoned resolution and preventive review by the audit court. The State's record labels it plainly «Compra Directa 213/2017», and that same site labels purchases «Compra por Excepción» when a body uses the exception route. None of this is an established infringement, and a supplier appearing in a single year breaks no rule.
TOCAF, Decreto 150/012, art. 33 (contracting procedures and grounds for exception), with the 2017 ceilings published by the state procurement agency: ordinary direct purchase $410,000, abbreviated tender $8,199,000. Sources: the table «Montos límites de las adquisiciones estatales» the agency publishes on gub.uy, and the January 2017 TOCAF published by the agency itself.
- What is missing before asserting it
- Three things, and none of them is in the feed. (1) How much was actually paid: the «Monto Total de la Compra» the State publishes is computed by multiplying quantity by unit price, so if the 72,580 hours were a data-entry error the US$ 8.4 million total would be one too, and the official figure is not independent confirmation; the departmental government's accounts or the payment file are needed. (2) The resolution that grounded the procedure: whether there was a declared climate emergency in Florida in January 2017 and whether the spending went through the audit court; we looked for a storm in Florida on that date and did not find it documented in indexed press, and the Florida disaster that is documented is the one of March 2024. (3) Whether the 72,580 hours were a contractual maximum or a quantity actually consumed. There is also a limitation of our own: the peso amount in our database for this release is computed with a 2025 exchange rate applied to 2017 dollars and cannot be cited.
- Who has to answer
- The Intendencia de Florida — the mayor in 2017 was Carlos Enciso, of the Partido Nacional, in office between 2010 and 2019, a matter of public record and not of our measurement: why a contract of that size was processed as a direct purchase, what emergency grounded it, how many hours were executed and how much was actually paid. The Junta Departamental de Florida, which holds political oversight of departmental spending. The Tribunal de Cuentas de la República, the audit court: whether the spending was reviewed and whether it was formally objected to. And CYCLATER S.A. (tax ID 217270870013, Cardal, Florida): with what fleet it supplied 72,580 excavator hours.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-florida-excavadora.verify.ts
20 / 32
Why did the Bella Unión auxiliary hospital resolve twelve direct purchases from the same provider on the same day?
- What we measured
- First of all, what this angle does not show. The largest cases by money in our own measurement collapse on opening the official record: the top ones are oncology medicines that the national health directorate buys dose by dose and patient by patient — 2 vials of pembrolizumab, 56 tablets of axitinib — and the State's page declares them «Compra por Excepción» (art. 33, letter D), a figure that has no ceiling, so there is no ceiling to dodge; another is the professional stamp duty, which we have already published; and the social-security institute's wheelchairs are, in the two records we opened, «Concurso de Precios», a competitive procedure. Publishing that raw ranking would be telling a different story. The measurement: since 1 January 2024 there are 578 combinations of public body, supplier and article with six or more separate purchases of the same article within 15 days, all of them below that year's direct-purchase ceiling; 156 of those combinations reach ten or more. Those windows hold $302 million awarded, and $2,732 million adding up every purchase of those 578 pairs since 2024. Three cases survive verification against the official record. The first is the state health provider ASSE. The Bella Unión auxiliary hospital awarded, on 9 July 2026, twelve direct purchases to the same provider, the Sociedad Médico Quirúrgica de Salto, with consecutive numbers between 10813 and 10830/2026. They add up to $7,122,915 including tax; nine of the twelve fall between $640,000 and $654,000, and three are exactly $654,000, which is the legal maximum for a direct purchase in 2026. The explanation the record itself suggests is in plain sight: the hospital has no trauma service, it operates on its patients in a private clinic in Salto, and each record is a different procedure — open reduction of a bone fracture, major osteosynthesis with external mini-fixators — on a different patient, so it enters as a separate purchase. The price on the same day varies between $353,146 and $654,000, so it cannot be asserted that the amount was set by the ceiling rather than by the complexity of each operation. The second is DINACIA, the civil aviation authority. Between 2019 and 2024 it contracted the cleaning of its premises with one annual purchase — the last, $8,193,349 including tax, an extension of Public Tender 10/2022 — and since 28 August 2025 it buys it month by month and site by site: 22 direct purchases, several of them for exactly $263,000 including tax, which is the 2025 ceiling. The usual administrative explanation is that the annual contract ran out and the service cannot be cut off while the new tender is being processed. The third is the child-welfare institute INAU. Its Paysandú departmental office awarded, on 22 December 2025, five direct purchases to the same supplier for building maintenance, $1,060,420 in total, four of them between 96.1% and 99.8% of the ceiling. There is a methodological by-product worth having on its own: this check is impossible with the open data. The award release does not carry `tender` — zero of all those reviewed — and in a random sample of 300 award ocids only 71 have a sibling release with the method, so the type of procedure is not in the feed; and the feed publishes amounts excluding tax while the ceiling is measured including tax, so purchase 1281409 appears with $215,574 in the feed and $263,000.00 on the official record. To know whether a purchase sits at the ceiling one has to open the record one at a time: we opened 42. Moreover, the burst is measured on the resolution date and not on the release date: in a random sample of 60,000 releases only 33.1% share a day, and measured on the release date large false positives appear that are bulk loads of the feed. Nothing measured shows that anyone split a purchase. That judgement belongs to the spending officer and to the audit court, and it requires the files, which are not published. On the same buyer there is prior coverage that did not come from our measurement: El País reported on 29 June 2025 that ASSE pays $280,000 for each specialised medical transfer by private ambulance from Bella Unión to Montevideo — another supplier and another angle, the price rather than the splitting — and that pair also appears in our ranking. LaRed21 reported on 8 June 2026 on direct contracting at the Salto departmental government.
- Which law it would engage
TOCAF (Decreto 150/012). Art. 33, letter C sets the ceiling for the ordinary direct purchase, which the statistics institute updates each year by inflation (art. 156). Art. 32 establishes that splitting a purchase artificially so that the operation fits within certain limits is a serious offence. The rule would only be engaged if it were confirmed that the contracted object was a single one and that it was split to avoid the competitive procedure its amount required: twelve direct purchases on the same day for $7.1 million from a single provider would amount, taken as a single contract, to an abbreviated tender ($13,705,000 ceiling in 2026, with a price competition mandatory above $1,371,000); DINACIA's $263,000 monthly cleaning per site amounts to several million a year, which is what used to be contracted by public tender. Nothing measured proves that condition on its own: the TOCAF itself requires the splitting to be artificial, and that judgement belongs to the spending officer and the audit court, not to the data. The exception purchase, by contrast, has no ceiling, which is why the national health directorate's cases fall outside this analysis.
TOCAF, art. 32: «La comprobación de que se fraccionare el gasto artificialmente para que la operación encuadre en determinados límites será considerada falta grave a efectos de las sanciones que correspondan». Ceilings in force according to the statistics institute, «Montos Límites de las Adquisiciones Estatales», art. 33 letter C: 2024, $251,000 (other public bodies) and $940,000 (departmental governments); 2025, $263,000 and $987,000; 2026, $654,000 and $1,024,000 (Ley 20.446 of 16/12/2025). Extended direct purchase (art. 44): $1,254,000 in 2024, $1,317,000 in 2025, $1,371,000 in 2026.
- What is missing before asserting it
- The files, in all three cases. For Bella Unión: the detail of the invoices the twelve purchases regularise — how much they came to, how many operations — and whether they correspond to twelve distinct surgical procedures or to instalments of a single debt; the official record names neither patients nor surgery dates. For DINACIA: whether a tender was under way between August and December 2025 and at what stage, and whether the monthly purchases correspond to different sites; the official record says only «LIMPIEZA INTEGRAL DE LOCALES» and the amounts repeat identically month after month without identifying the place. For INAU: what works each of the five same-day purchases covered. In all three it remains unknown whether the audit court formally objected to this spending: the objection lives in the file and in the court's PDF, not in the feed, so we can never write that the court objected to something on the basis of the data. And the prior-coverage check still has to be closed: the search budget ran out halfway through, and the check was done via Google News RSS in Uruguayan Spanish without sweeping the archives of Búsqueda, la diaria or Brecha; before publishing it must be redone with the exact names: «Sociedad Médico Quirúrgica de Salto», «DINACIA limpieza», «INAU Paysandú mantenimiento».
- Who has to answer
- ASSE, the management of the Bella Unión auxiliary hospital and its purchasing department: for the twelve direct purchases of 9 July 2026 and for the accumulated volume with that same provider since 2024. DINACIA, of the Ministry of National Defence: for the shift from an annual public tender to 22 monthly direct purchases between August and December 2025. INAU, Paysandú departmental office: for the five purchases of 22 December 2025. And the Tribunal de Cuentas de la República, the audit court: whether any of this spending was formally objected to. The request for comment from all three bodies goes inside the story.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-directas-mismo-dia.verify.ts
21 / 32
383 public works doubled through an extension and appear in the open data as a new contract
- What we measured
- Extending a contract up to 100% is lawful and is published, and that comes before everything else. Art. 74 of the TOCAF authorises it with the awardee's agreement and under the same agreed conditions, without requiring a fresh tender; in road works it is usually the cheapest route, because the company is already mobilised and prices are locked in. The extension appears on the purchase's public record, with its PDF resolution, its date and its amount, and the extension's record links back to the original. And the money is not mis-added: the feed counts both purchases, which is what was spent. What does not exist is the link. Of 2,184,332 releases, none has the `contracts` stage, where the OCDS standard stores modifications, and none has `awards.amendments`. The only amendment tag that exists, `awardUpdate` (6,871 releases), is a data-correction channel: of the 3,434 awards it corrects, 74.5% do not raise the amount — 1,498 lower it and 1,062 leave it unchanged — and an extension by definition raises it. And none of the 405,646 tender releases carries a purchase method that says extension. The extensions are there, in another shape. We narrowed to public works between 2015 and 2025 (the SICE catalogue family «CONSTRUCCIONES, MEJORAS Y REPARACIONES EXTRAORDINARIAS», 549 codes, 13,935 awards) and looked for distinct purchases by the same body to the same tax ID for exactly the same amount down to the cent: 528 groups and 624 repeated purchases. We verified the 624 one by one against the government's public record: 383 — 61% — carry the «Ver Compra Original» banner, meaning they are extensions. The figure that orders everything is 383 out of 383: every confirmed extension reaches the feed without the procedure — without a title, without a method and without any reference to the original purchase — whereas of the 528 «first» purchases of each group, 336 do carry it, mostly abbreviated and public tenders. A contract that doubled without competing again is recorded in the open data as a loose award, indistinguishable from a direct purchase, and any competition statistic built on that feed — including our own on single bidders and apparent competition — counts it wrongly. The witness case can be seen without a database, by opening two pages. Purchase 529774 is Public Tender 595/2016 of the Montevideo city government, awarded to GRINOR S.A. for $69,726,649.99 including tax, and on its own record the «Ampliación/Renovación de contrato» block reads «Ampliación Nro. 1 — 07/08/2017 — $69,726,649.99»: exactly 100%. Purchase 610468 carries the same heading «Licitación Pública 595/2016» and the same amount down to the cent, but at the top it says «Ver Compra Original» instead of «Ver Detalle del Llamado». The contract went from $69.7 million to $139.4 million. In the feed, the original returns two releases — a tender one, with method and title, and an award one — and the extension returns a single one, with no method, no title and no field linking it to the original. The button exists on the web and does not exist in the data. One question remains open and is published as a question. In four works the government's own record shows two extensions on the same original contract: the Colonia departmental government with EBAFOX S.A. (purchase 453562, extensions 474310 and 474325), the Salto departmental government with JOSE CUJO S.A. (1003514, extensions 1289863 and 1289877), the Florida departmental government with INSUR S.A.S. (1015718, extensions 1190516 and 1198929) and the Salto departmental government with BATISTA OLIVERA HONORIO BENITO (1135421, extensions 1253629 and 1253698). All eight extensions are verified one by one as separate purchases carrying the «Ver Compra Original» banner. If the two accumulate on the same object, the result is 200%; if the second re-registers the first, it is not. That is settled by the resolution, not by the database. Two figures that do not come from our measurement and are worth citing: Uruguay's entry in the Open Contracting Partnership data registry documents the gap in its own catalogue — «Count of contracts: 0», «Information on extensions or renewals made to an award is not included» — and the use of the instrument has already been reported case by case in earlier publications, among them CSI Ingenieros charging the state power company US$ 2.8 million for extensions that exceeded 130% of the original contract, and the public-ethics board finding spending fragmentation and irregular extensions at the Artigas departmental government. What had not been published is that the mechanism cannot be audited at scale because it is not in the data. Limit of the measurement: the fingerprint only sees extensions of exactly 100%, because they are the only ones that double the amount down to the cent; partial ones, of 20% or 50%, are invisible in the corpus, so the 383 are a floor.
- Which law it would engage
Two things are worth separating. The data gap breaks no rule: art. 50 of the TOCAF requires the award act to be publicised on the state procurement website «así como las ampliaciones», and that is complied with, because the extension is on the purchase's HTML record with its PDF resolution. What does not exist is the machine-readable channel: the OCDS export, the only format that allows oversight at scale, does not carry that information, and the Open Contracting Partnership's own registry documents this for Uruguay. The rule that would be engaged is art. 74 of the TOCAF, which caps increases at 100%: in four works the government's record shows two extensions on the same original contract, and if those two accumulate on the same object the cap has been left behind; if not, it has not. It is a question for the public body, not a verdict.
TOCAF (Decreto 150/012), art. 74, verbatim: «Las prestaciones objeto de contratos podrán aumentarse o disminuirse, respetando sus condiciones y modalidades y con adecuación de los plazos respectivos, hasta un máximo del 20% (veinte por ciento) o del 10% (diez por ciento) de su valor original en uno y otro caso y siempre que el monto definitivo no sobrepase el límite máximo de aprobación para el cual está facultada la respectiva autoridad. Cuando exceda ese límite deberá recabarse la aprobación previa de la autoridad competente. También podrán aumentarse o disminuirse en las proporciones que sea de interés para la Administración y que excedan de las antes indicadas, con acuerdo del adjudicatario y en las mismas condiciones preestablecidas, con la limitación de que en ningún caso los aumentos podrán exceder el 100% (cien por ciento) del objeto del contrato». TOCAF, art. 50 (text under Ley 20.212): publication on the state procurement website is mandatory for the call to all competitive procedures and for the «acto de adjudicación, declaración de desierta o de rechazo de ofertas» in procedures exceeding 20% of the direct-purchase limit, «incluidos los realizados por mecanismos de excepción, así como las ampliaciones y los actos de reiteración de gastos observados por el Tribunal de Cuentas»; public bodies «contarán para ello con un plazo de diez días luego de producido el acto que se informa». Sources: impo.com.uy/bases/tocaf2012/150-2012/74 and /50; Uruguay's entry at data.open-contracting.org/en/publication/43.
- What is missing before asserting it
- Three things. (1) The real numerator: the fingerprint only detects extensions of exactly 100%, partial ones are invisible in the corpus, and therefore the 383 confirmed are a floor and not a total; the total can only come from a sampled control over works awards that are in no pair, and that is an estimate with uncertainty. (2) The four works with two extensions on the same original contract: the PDF resolutions of each still have to be read to know whether the second accumulates on the same object or re-registers the first, and without that PDF no breach can be asserted. (3) The procurement agency's answer: if the extension is published on the HTML record and has its own «Ver Compra Original» button, it remains unknown why the OCDS export does not mark it — whether that is a decision, a system limitation or an outstanding task — and that answer changes the tone of the story. It also remains unmeasured whether the pattern holds outside public works: we only measured the construction family, and the mechanism is general to any contract.
- Who has to answer
- The state procurement agency, which produces and publishes the OCDS feed: why the extension appears on the web record, with its own link to the original purchase, and disappears in the open-data export, and whether there is a plan to incorporate the standard's `contracts` stage. The bodies with the most confirmed works extensions — the departmental governments of Colonia, Salto, Montevideo, Florida, San José and Tacuarembó, and the national roads directorate: why extending to the cap was better than a fresh tender. And specifically the departmental governments of Colonia, Salto (two cases) and Florida, for the four works with two extensions on the same original contract.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-ampliacion-invisible.verify.ts
22 / 32
More than half the money awarded is published with no article code, and two thirds of that is ANCAP buying crude
- What we measured
- The OCDS feed publishes awarded items under two classification schemes. One, `x_catalogo_arce`, carries the article code from the SICE catalogue (for instance 6194, SERVICIO DE SANITARIA). The other, `x_ODG` (object of expenditure), always carries the code «0» — that is, none — and only free text. Between 2016 and 2025 the State awarded $1,275,483 million; of that, $697,006 million — 54.6%, across 214,230 records — carry no article code on any of their lines. Two thirds of that block is ANCAP, the state oil company: $467,433 million, 67.1%, almost all of it crude and fuel purchases. And there the explanation is complete: a cargo of «CRUDO BONNY LIGHT» or «CRUDO WTI MIDLAND» is not a catalogue article, it is paid against the day's Brent or WTI and not against a history of Uruguayan purchases, so coding it would enable no comparison. The same, to a lesser degree, holds for the power company's diesel and for the turnkey works of the water and ports authorities, where the «article» is an entire project. Discounting ANCAP, $229,573 million remain without a code, 18.0% of everything awarded in the decade, and that is the figure that does not have that explanation. The route is institutional and not a data-entry oversight: no record mixes the two schemes — either everything is coded or nothing is — and only 9 public bodies use `x_ODG`: ANCAP (99% of its own spending), the water authority OSE (99%), the ports authority ANP (100%), the Canelones departmental government (100%), the Montevideo city government (89%), the power company UTE (56%), the state insurer BSE (50%), the state bank BROU (14%) and the central bank BCU (2%). They are autonomous entities, decentralised services and two departmental governments, with their own systems connecting to the feed through a different pipeline from the central administration's procurement system. The description is published all the same and is often specific («SULFATO DE ALUMINIO LIQUIDO», «CRUDO QUA IBOE», «ÁCIDO SULFÚRICO»). The proportion was not always like this: it went from 0.3% of the money in 2014 and 1.5% in 2015 to 33.8% in 2016, 48.0% in 2019 and 60.0% in 2020, and stayed near that level up to 2025 (53.1%). Where the explanation stops holding is in the labels that are not commodities: the state bank's «NO CATALOGADO» appears on 14,424 lines, and «RUBROS GLOBALES», «RUBROS UNITARIOS», «LL SS» and «IMPREVISTOS» from the water authority, the Montevideo city government and the ports authority take up entire purchases under a heading that does not say what was bought. The free text does not aggregate either: between 2020 and 2025 there are 156,910 distinct descriptions for 348,466 lines under `x_ODG`, and 72.0% appear only once, against 22.3% on the catalogued side. We do not publish an amount per label because the amount lives on the record and a single purchase carries several of those labels at once: attributing the total to each would inflate them all. The consequence is measurable and is not an insinuation: the price-baseline collection holds 40,829 entries and not one for the code «0». A line without a code cannot be compared against any reference price, by construction. Half of the State's money is incomparable, not because of us and not because of anyone. Two more things, and one is about us. The hypothesis we started with — that generic articles of the «other» or «miscellaneous» type were absorbing large amounts inside the catalogue — was ruled out: they are 20 codes and $232 million, 0.04% of coded spending. The catalogue is specific; what does not pass through it is half the money. And our own products page is built on a rollup that excludes that block, calling it a «junk sentinel», without telling the reader it leaves out more than half of the spending. That warning goes on the products page the same day as this card. One last limitation: 2026 falls outside the whole series because that year the feed carries a null classification scheme on 100% of the lines, so it is not comparable.
- Which law it would engage
No rule is clearly broken, and it is worth saying that before anything else. The only provision engaged is art. 331, item 4 of Ley 19.889, which creates the state procurement agency and charges it with «desarrollar, publicar y coordinar con las diversas entidades estatales, la efectiva aplicación de un catálogo único para la adquisición de bienes». Two important limits: the obligation falls on the agency — to coordinate effective application — not on each public body as a sanctionable duty to code every line; and the text speaks of «bienes», goods, not of services or works, which is where much of the uncoded block is concentrated. The agency itself maintains on its site that since that law came into force the catalogue of goods, services and works applies to all state public administrations, but that is an administrative interpretation and not the legal text. What the data measures, then, is not a breach: it is the degree of effective application of a catalogue the law charges the agency with making effective, and which over half the money is not.
Ley N° 19.889 of 9/7/2020 (Ley de Urgente Consideración), «Compras Públicas» section, chapter IV «Creación de la Agencia Reguladora de Compras Estatales», art. 331 item 4: «Desarrollar, publicar y coordinar con las diversas entidades estatales, la efectiva aplicación de un catálogo único para la adquisición de bienes» (impo.com.uy/bases/leyes/19889-2020/331). The agency's interpretation on its «Reestructura del catálogo» page, on gub.uy: «A partir de la vigencia de la Ley 19.889, el catálogo de Bienes, Servicios y Obras es de aplicación para todas las Administraciones Públicas Estatales». General framework: TOCAF and the agency's powers, arts. 329 to 339 of the same law.
- What is missing before asserting it
- Four things, all answerable in writing. (1) What `x_ODG` is: whether it is a foreseen and accepted publication mode, a technical limitation of each body's integration, or a granted exception; the data does not distinguish between the three, and they are three different stories. (2) Whether the code exists internally and is lost on export, which is the hinge question: if ANCAP, the water authority or the Montevideo city government classify internally and the code does not travel to the OCDS, it is a publication problem; if it never existed, it is a procurement-management problem; from the database it cannot be known, because we only see what the feed brings. (3) What the non-commodity labels cover: «RUBROS GLOBALES (I, II, III)» and «RUBROS UNITARIOS» from the water authority, «LL SS», and the state bank's «NO CATALOGADO» on 14,424 lines; there no international-market explanation is possible and each one's tender document is needed. (4) The 2016 break: the jump from 1.5% to 33.8% of the money between 2015 and 2016 looks like new bodies joining the feed through a different route, or a change in the exporter, and it has to be confirmed against the agency's integration history before being read as a deterioration. A limitation of our own: 2026 falls outside the whole series because that year the feed carries a null classification scheme on 100% of the lines, so it is not comparable. And what depends only on us is still missing: saying on the products page what share of the spending is not there.
- Who has to answer
- The state procurement agency, first of all, because the law charges it with the effective application of the single catalogue and it runs both the procurement system and the OCDS publication: why 54.6% of the money awarded between 2016 and 2025 is published under the `x_ODG` scheme with no article code, whether that is foreseen, and whether there is a plan to map it. The four bodies with the highest share of their own spending — ANCAP (99%, $467,433 million), the water authority OSE (99%, $89,838 million), the ports authority ANP (100%, $19,391 million) and the Canelones departmental government (100%, $20,679 million) — each of them: whether they classify their purchases internally with the catalogue and the code does not reach the open data, or whether they simply do not use it. The water authority, on what «RUBROS GLOBALES (I, II, III)» comprises; the state bank, on what «NO CATALOGADO» means across its 14,424 lines; and the power company, on why 56% of its spending goes without a code when the other 44% does use the catalogue, being the same body and the same system. And ourselves: our products page is built on a rollup that discards that block, calling it a «junk sentinel», and never told the reader they were looking at less than half of the State's spending. The warning goes on that page, and this card is not published before it.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-sin-codigo.verify.ts
23 / 32
If the tender was published on the national register, why was the outcome not? 279 departmental tenders with no recorded ending
- What we measured
- We took the abbreviated and public tenders of the departmental governments published between 1 July 2020 and 30 June 2025 — 137,273 releases swept, grouped by ocid — and looked for whether an award, an award update, an award cancellation or a tender cancellation ever appeared under the same ocid. The most recent tender in the window closed bidding in July 2025: thirteen months before this measurement. Across 19 departmental governments and 1,480 tenders, 685 have a published outcome: 46.3%. Tenders, with outcome, and share: Cerro Largo 43 / 0 / 0% · Lavalleja 11 / 0 / 0% · Soriano 60 / 0 / 0% · Rocha 62 / 1 / 1.6% · Maldonado 107 / 3 / 2.8% · Durazno 247 / 58 / 23.5% · Tacuarembó 91 / 63 / 69.2% · Treinta y Tres 29 / 23 / 79.3% · Salto 5 / 4 / 80% · Artigas 66 / 53 / 80.3% · Flores 90 / 74 / 82.2% · Río Negro 24 / 21 / 87.5% · Florida 9 / 8 / 88.9% · Rivera 10 / 9 / 90% · Colonia 229 / 217 / 94.8% · San José 61 / 58 / 95.1% · Paysandú 47 / 45 / 95.7%. The first five add up to 283 tenders and 279 have no published outcome at all. Lavalleja is 11 cases: the zero is out of eleven, and it is read with that number next to it. Montevideo and Canelones fall outside the measurement, and that exclusion is part of the finding: their purchases use ocids with an «i» prefix and the State's own OCDS endpoint returns 404 for them, so the automatic check does not run. Without those two, the average of the remaining 17 is 53.5% (637 of 1,191). We checked against the official source: on three outcome-less tenders of those departmental governments, /ocds/record returns 200 and a single tender block, never an award; on three controls from Colonia and Paysandú that do have an award, the official record carries it. None of the 1,292 outcome-less tenders in the period carries an amount, so we cannot say how much money is behind them. What was being bought is visible: repaving a riverfront avenue (Soriano), 7,000 m3 of gravel (Lavalleja), a telescopic handler (Rocha), three years of security services (Maldonado), synthetic turf for a municipal pitch (Cerro Largo). The competing explanation is strong and goes here, not in a footnote: the register emits a new record when there is an award, and it has nowhere to note that nothing happened. If a tender is declared void, set aside, or falls through because the budget line never arrived, the purchase stays as it was. Montevideo proves it: we opened three official HTML records of its tenders that the feed shows with no outcome (i322321, i288892, i447021) and all three carry a published resolution, «Declarada sin efecto», with number, date and PDF — and the feed emits not one release for it. It is possible that a good share of these 279 tenders are tenders that died without an award, and a failed tender is not a hidden contract. What answers that explanation is a fact: on the nine records we reviewed from the five departmental governments there is no resolution block of any kind — not awarded, not void, not set aside. The second explanation also holds: the departmental government files the tender because prior publicity is mandatory and is how suppliers find out, and then processes it in its own file without returning to the national system. And the third: Colonia, with 229 tenders in the same period, closes 94.8%, and the five above are departmental governments whose tender offices are one or two people. Two of the five changed: Cerro Largo began publishing awards in November 2025 and Maldonado in July 2026, in both cases after the change of departmental government. Durazno went the other way: it published 126 awards between 2021 and 2024 and none since July 2024.
- Which law it would engage
TOCAF (Decreto 150/012), article 50, as in force after Ley 19.355 and Ley 20.212: it requires state public administrations to publish on the state procurement website the award act, the declaration that a tender was void, or the rejection of offers, within ten days of the act, for every procedure whose amount exceeds 20% of its direct-purchase ceiling. Departmental governments fall within the TOCAF's scope under its article 2. The inference we make, and we give it as an inference: an abbreviated or public tender is, by definition of the ceiling regime, above the direct-purchase ceiling and therefore above 20% of it. The theoretical exception exists, because nothing prevents voluntarily using a more competitive procedure than the amount requires, and we could not verify it case by case, because the amount is precisely the missing datum.
TOCAF art. 50: state public administrations shall publish on the state procurement website «al acto de adjudicación, declaración de desierta o de rechazo de ofertas» of all their procurement procedures above «20% (veinte por ciento) del límite de su procedimiento de compra directa», including those carried out through exception mechanisms, extensions and the acts reiterating expenditure observed by the Court of Auditors, having «un plazo de diez días luego de producido el acto que se informa». Text at impo.com.uy/bases/tocaf2012/150-2012/50. Competent bodies: the state procurement agency (runs the portal) and the Court of Auditors (external control).
- What is missing before asserting it
- We could not measure Montevideo or Canelones. The State's own OCDS endpoint returns 404 for their purchases — 13 of 13 tried, including one that does have an award in our database — and the three official HTML records of Montevideo's city government we reviewed carry a published resolution even though the feed brings no release: for them, «no award» does not mean «did not publish». Measuring them requires reading the HTML records one by one. The HTML sample is 9 out of the 279 outcome-less tenders of the five departmental governments, 3%. The /ocds/record check covers others of those 279 with the same result, and for Soriano, Rocha and Lavalleja the finding does not depend on the sample: no award by those bodies exists in the corpus since 2013, and in Soriano's case, ever. We cannot say how much money is behind it: 0 of the 1,292 outcome-less tenders carry an amount, and any peso figure would be invented; this card carries none. We did not verify the amount threshold case by case: we inferred it from the type of procedure. We do not know whether the resolution exists on another medium: Rocha's own portal publishes only current tenders and refers the rest to comprasestatales, and Soriano's advertises itself as a list of purchases and tenders without showing results; we did not check the Official Gazette, departmental council minutes or case files, and the TOCAF requires publication on the state procurement site but the resolution may exist and be published elsewhere. We ruled out the outcome being published under a different ocid — which is what happens when the agency gives a new number to an extension: Soriano never published an award on the national register in the whole 2002-2026 corpus, Rocha published two, both in 2012, and Lavalleja's last were in 2013, so no sibling is possible; for Cerro Largo and Maldonado we ran the sibling search by shared article codes within 18 months and it returned 2 and 7 candidates, all with a single generic code in common and a different subject. And there is something we did not understand and therefore do not give as a finding: on the three Montevideo records the date of the «set aside» resolution is the same as or earlier than the publication date of the tender.
- Who has to answer
- The departmental governments of Soriano, Rocha and Lavalleja, still at zero: how the supplier who did not win, and the resident, found out how each of these tenders ended; whether the award act, the declaration of void, or the rejection was issued; and where it was published, if not on comprasestatales.gub.uy. Maldonado and Cerro Largo, which began publishing awards in July 2026 and November 2025: what changed, and whether they will fill in the 2020-2025 files retroactively. Durazno, which published 126 awards between 2021 and 2024 and none since July 2024 while still publishing tenders: what was interrupted. The state procurement agency: why a «declared void» or «set aside» resolution appears on the HTML record and generates no release in the OCDS feed, so that the open data shows the tender still open; why /ocds/record returns 404 for purchases with an «i» identifier; and whether it monitors compliance with article 50's ten-day deadline. The Court of Auditors and the transparency board: whether there is any precedent of an observation for breach of article 50 on the publicity of departmental awards.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-intendencias-sin-resultado.verify.ts
24 / 32
The same firm charged the public health service 875 pesos in Pando and 2,290 in Rocha for the same scan
- What we measured
- We worked on the 1,638 award lines of catalogue code 41238, BREAST ULTRASOUND, between 2007 and 2026. All 1,638 are in pesos, so the null-currency trap does not apply. We excluded 6 lines with a unit price of 20,000 of today's pesos or more — whole agreements entered into the unit-price field — and 29 lines with a unit price of 250 or less, which are professional stamp duties filed under this code: the stamp's legal value is 170 in 2026 and 160 in 2025. That leaves 1,603 lines. In 2026 there are 126 clean lines across 12 buying units: the median for the same scan runs from 875 pesos at the Pando auxiliary centre to 3,087 at the Florida departmental centre, 3.5 times. The spread holds with the same seller. GASTROCLINICA S.R.L. (tax id 213506820011) sold code 41238 to seven units of the public health service in 2026: 875 in Pando, between 2,290 and 3,105 in Rocha, 2,290.7 at the San Carlos hospital, 2,460 at the Colonia primary-care network, between 2,405 and 3,105 at the San José primary-care network, 3,105 in Florida and 3,105 at the Cerro hospital. Same firm, same code, same year, same currency. It is not only the breast scan: for adult abdominal ultrasound (code 104040) the 2026 median runs from 831 in Pando to 2,460 at the Colonia network and 4,097 in Rocha; for soft tissue (104048), from 875 in Pando to 2,290 in Rocha and 2,460 in Colonia; for testicular (42807), from 796 in Pando to 2,405 in Rocha. The price moves within a single unit too: Pando paid 2,524 in 2024, 2,345 in 2025 and 875 in 2026, in nominal terms and for the same scan, 65% less in two years; Rocha went the other way, from 750 in 2021 and between 950 and 1,248 from 2022 to 2025 to 2,290 in 2026, on changing provider. The shape of the purchase: of the 180 sibling tender documents from 2024 to 2026, 168 are direct purchases, 93%; and of the 109 ocids with a 41238 line in 2026, only 13 have any tender release at all. We checked both ends on the State's own page. Pando, purchase 1312418: direct purchase, GASTROCLINICA S.R.L. tax id 213506820011, 63 breast ultrasounds at 875 pesos before tax. Rocha, purchase 1310750: direct purchase, the same firm and the same tax id, 18 breast ultrasounds at 2,290 pesos before tax, with a single participating supplier. Neither record mentions a radiologist's report. The competing explanation goes here: volume and bundling, in a system decentralised by design. The 875 in Pando is one line inside a contract that in 2026 bought thousands of scans across the whole ultrasound menu from GASTROCLINICA; the 2,290 in Rocha comes out of dozens of direct purchases of about three scans each. A supplier quoting differently for guaranteed volume is ordinary economics. On top of that, the health service's purchasing is decentralised by executing unit: there is no single price, there are twelve departmental markets with different costs, distances and availability of sonographers, and Rocha's previous provider was an individual professional, without a company's cost structure. Two facts limit that explanation. The Colonia primary-care network bought more scans from the same firm than Rocha did, and through an abbreviated tender — that is, through the competitive procedure — and pays 2,460, more than either. And Pando itself was paying 2,524 in 2024 and 2,345 in 2025 before dropping to 875 in 2026, with no change in the scan.
- Which law it would engage
The general rule for State purchasing is the competitive procedure; the direct purchase is the exception. Under this code, 93% of the sibling tender documents from 2024 to 2026 are direct purchases, and the Rocha purchase we verified lists a single participating supplier. The second rule engaged is the one that forbids splitting expenditure so that each operation fits under the direct-purchase ceiling: Rocha bought more than a hundred breast ultrasounds in dozens of separate purchases, from the same supplier, over the year. A necessary clarification: we did not measure whether any unit exceeded the direct-purchase ceiling. The ceilings are updated every year and we did not check them against each executing unit's aggregate. The pattern of repeated purchases from the same supplier is a question, not an accusation, and the word «fraccionamiento» is not used here as an assertion.
TOCAF (Decreto 150/012), art. 33: «Las contrataciones se realizarán mediante licitación pública u otro procedimiento competitivo expresamente previsto...». TOCAF, art. 32: «La comprobación de que se fraccionare el gasto artificialmente para que la operación encuadre en determinados límites será considerada falta grave...». Consulted in the TOCAF published by the state procurement agency (comprasestatales.gub.uy/ManualesDeUsuarios/manual-procedimiento-compras/TOCAF.html) and on IMPO (impo.com.uy/bases/tocaf2012/150-2012/33). Competent bodies: the public health service as the spending authority, the Court of Auditors as external control.
- What is missing before asserting it
- What each price includes. The feed does not describe the service: we do not know whether Rocha's 2,290 includes the radiologist's report, the sonographer's travel, on-call or emergency cover, and Pando's 875 does not. The only documentary clue points against the report explaining the expensive end — the one buyer whose tender documents state verbatim that the report is included pays 1,916, below Rocha, Flores, Florida and San Carlos — but it is a single buyer, and it is settled with the tender documents and purchase orders, not with the feed. Whether other prices were sought in Rocha: purchase 1310750 appears as a direct purchase with a single participant, and in 96 of the 109 ocids of 2026 there is no tender release at all, so we can assert that no public record was left of other prices being sought, not that they were not sought. The direct-purchase ceiling per executing unit, which we did not check. Whether GASTROCLINICA delivers the service with a different structure in each department — its own equipment in one place, subcontracting in another — which is the explanation the feed cannot measure. 2026 is incomplete: it is January to August, and this year is missing a feed channel; the 2026 levels hold (126 clean lines, 12 units, both ends verified on the State's site), the annual total of scans does not. And every amount is before tax, because the OCDS feed does not carry the tax-inclusive total: both ends of the comparison are on the same basis, so the ratio between them is unaffected, but no figure here is a final price. A method warning for anyone repeating this: the same tax id lives in the feed under two identifiers, «R213506820011» and «R/213506820011»; filtering by only one loses 82 or 149 of the supplier's 231 lines depending on which you pick, and grouping by name never works.
- Who has to answer
- The public health service's procurement directorate, and the directorates of the executing units involved: the Rocha departmental centre, the Pando auxiliary centre, the Florida departmental centre, the Cerro hospital and the Colonia primary-care network. The questions: why the same firm charges 875 in Pando and 2,290 in Rocha for code 41238 in the same year; what each price includes and whether the radiologist's report is in or out in each case; why Rocha bought more than a hundred scans in dozens of separate direct purchases instead of an annual contract like Pando's; and whether the service has or intends to have a central reference price for outsourced imaging. GASTROCLINICA S.R.L. (tax id 213506820011), to explain the price difference between departments.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-ecografia-precio.verify.ts
25 / 32
Almost half of public tenders come back observed by the Court of Auditors, and the dominant reason is not lack of funds
- What we measured
- The figures on this card do NOT come from our database: they come from the 14 Annual Reports of the Court of Auditors published on tcr.gub.uy between 2011 and 2024, which we downloaded and whose tables we extracted. All 14 have a text layer and none is a scan. Parsing produced 101 tables and 100 reconcile against their own printed TOTALS (82 exactly, 18 with differences of up to 3 thousand pesos from source rounding); the only one that does not reconcile is a source error, in the 2015 report. The denominator is what nobody used: every year the Court publishes how many public tenders it processed and how many came out observed. Observed over processed: 2011 213/823 = 25.9% · 2012 245/782 = 31.3% · 2013 191/428 = 44.6% · 2014 199/459 = 43.4% · 2015 247/534 = 46.3% · 2016 244/462 = 52.8% · 2017 202/544 = 37.1% · 2018 204/597 = 34.2% · 2019 196/647 = 30.3% · 2020 150/448 = 33.5% · 2021 215/529 = 40.6% · 2022 238/563 = 42.3% · 2023 279/606 = 46.0% · 2024 259/573 = 45.2%. Direct contracts under the exception of article 33 item 3, in 2024: 116 of 329, 35.3%. That denominator is almost the whole universe, not a sample: our corpus has 548 tenders with the Public Tender method in 2024 and the Court says it processed 573. In other years the gap is wider — 2022: 738 in the corpus against the Court's 563 — so the cross-check serves to say «same order of magnitude», not for a fine equivalence. The reason is the finding. The «MOTIVO DE LA OBSERVACIÓN» tables for the central administration and article 220 bodies are the only ones with the reason broken down. Across the 226,428 items of expenditure observed and reiterated between 2015 and 2024: 157,689 (69.6%) under article 33 of the TOCAF and following — the procurement regime, how it was bought; 3,844 (1.7%) under article 15, which is lack of available budget; and 35,461 (15.7%) under «Other», not broken down. In 2024 alone: 67.4% under article 33 against 2.8% under article 15. In none of the ten years does lack of funds exceed 3.4%. Who: the public health service is almost all of the central administration, 92.9% (31,056 of 33,422) in 2022, 95.0% (28,682 of 30,182) in 2023 and 76.2% (18,993 of 24,923) in 2024; among the state enterprises, in 2024, the oil company 16,536, the power company 5,934 and the water authority 2,329. The full 2024 financial year adds 24,923 from the central administration and article 220 bodies, plus 25,711 from autonomous entities and decentralised services, plus 3,101 from financial institutions — 53,735 in the central State — and 17,477 from departmental governments, 693 from departmental councils and 2,564 from municipalities: 74,469 items of expenditure observed that the spending authority reiterated. The competing explanation goes here, and it is the key to reading everything above: reiterating breaks nothing. Article 211 letter B of the Constitution instructs the Court to certify legality and to observe; article 114 of the TOCAF gives the spending authority the right to insist on reasoned grounds, the expenditure is paid under its exclusive responsibility, and the Court reports it to the General Assembly. An observed and reiterated expenditure is a documented disagreement between two branches, resolved through the foreseen procedure and published. Moreover, an article 33 observation is often formal: it is enough to have extended a contract above the permitted percentage, renewed it without the prior act, or bought directly what should have been tendered because the previous tender was void, and the table does not distinguish a procedural defect from a substantive problem. A single decision generates many items: a health agreement paid month by month produces one observed payment order per month and per provider, so the health service's 18,993 are not 18,993 decisions. And a hospital cannot stop buying medicines while a tender runs, which is the situation article 114 contemplates. On the denominator: the share of observed tenders going from 25.9% to 45.2% may reflect the auditor sharpening its criteria or devoting more resources; the series measures the activity of the controller as much as the conduct of the controlled, and we have no way to separate the two. Press precedent, not our measurement: la diaria published in May 2026 a piece headlined «ASSE: el 70% del gasto en convenios con el Círculo Católico fue observado por el Tribunal de Cuentas entre 2020 y 2024», behind a paywall, which uses Court observations for one specific relationship between a public body and a provider.
- Which law it would engage
Article 114 of the TOCAF (Decreto 150/012) and article 211 letter B of the Constitution, which together build the observation-and-reiteration mechanism. And, as the dominant reason for observations, article 33 of the TOCAF; as the near-absent reason, article 15. None of this describes a breach: it describes the normal working of a prior control, and what the card measures is where it concentrates.
Constitución, art. 211 lit. B): the Court of Auditors must «intervenir preventivamente en los gastos y los pagos, conforme a las normas reguladoras que establecerá la ley y al solo efecto de certificar su legalidad, haciendo, en su caso, las observaciones correspondientes». TOCAF art. 114: the spending authority may insist «en forma fundada, expresando de manera detallada los motivos», and the payment is made «bajo la exclusiva responsabilidad de dicho ordenador»; if the Court maintains the observation, it «dará noticia circunstanciada a la Asamblea General o a quien haga sus veces, o a la Junta Departamental respectiva». TOCAF art. 33: «Las contrataciones se realizarán mediante licitación pública u otro procedimiento competitivo expresamente previsto, de acuerdo a lo que mejor se adecue a su objeto, a los principios generales de la contratación administrativa…». TOCAF art. 15: «No podrán comprometerse gastos de funcionamiento o de inversiones sin que exista crédito disponible, salvo en los siguientes casos: 1) Cumplimiento de sentencias judiciales, laudos arbitrales…». Source of the data: Annual Reports of the Court of Auditors, index at tcr.gub.uy/memoria.php, financial years 2011 to 2024.
- What is missing before asserting it
- The reason is broken down only for the central administration and article 220 bodies. For autonomous entities and financial institutions the Court publishes counts and amounts per body: of the oil, power, water and telecoms companies we know how much was reiterated, not why, and the state enterprises' 25,711 items in 2024 — more than the whole central administration — are left without a reason. «Other» is 15.7% of the decade, 35,461 items, and 27.3% of 2024, unopened: if article 33 were in there, the 69.6% would be a floor; if something else were, we do not know. The departmental-government grids carry all eight reasons, but in columns whose headers the extractor returns out of order, so we took only the TOTALS rows, which are unambiguous, and we assert nothing about which reason dominates in departmental governments. The taxonomy changed: the row «Art. 211 lit. B) y Art. 33 del TOCAF» appears only from the 2019 report, and before that those cases could fall under «211 lit. B) y otra norma»; the conclusion about article 15 holds across the ten years, but article 33's year-to-year swing is not comparable across 2019. There is no money total: the Court's observed-amount column assigns the water authority 73,045 million pesos in 2024, and in our corpus it awarded 40,370 million that year; that column mixes multi-year contracts with one-off payments, it serves to rank bodies and not to add a national total. Each financial year keeps being processed in later reports — the 2023 year appears with 30,182 items in the 2023 report and with another 39,172 in the 2024 one — so any per-year figure is a floor and the most recent year is the most incomplete. The card does not say which specific expenditure was observed: the report gives aggregates and the file with a name and a subject is in each Court resolution, in PDF. And we could not tie the series to any identifier in our database: the Court counts items of expenditure, that is, intervened payment orders, and we count awards; they are not the same unit and there is no key to join them. Four parsing traps, all verified, any of which produces a false number: the word «excepto» inverts the meaning of two labels and, searching «art. 15» over the whole string, that reason goes from 35 to 5,641 cases in 2018; the header splits across two lines and, without rejoining it, two consecutive tables collapse into one; when a label overflows, a blank row leaves the label dangling right before TOTALS, which ends up consumed as data; and footnotes stuck to a figure break the row and delete the largest body in that table.
- Who has to answer
- The Court of Auditors ([email protected]): why article 33 concentrates seven of every ten observations and what is inside «Other». The public health service, which is between 76% and 95% of everything observed and reiterated in the central administration: why the volume of reiterations has held since 2017 and what changed in 2024, when it drops from 28,682 to 18,993. The oil and power companies, which in 2024 rise to 16,536 and 5,934 reiterated items, against 4,887 and 613 the year before. The General Assembly, which under article 114 receives a detailed report of every reiteration: what it does with them; Parliament's Court of Auditors Observations Committee publishes the register. And the state procurement agency, on the gap between the public tenders in the feed and those the Court says it processed.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-tcr-memoria.verify.ts
26 / 32
The social development ministry is the only one where commercial companies do not receive most of the money, and 44 of its 50 largest purchases of 2026 went through the no-tender route
- What we measured
- We measured budget line 15 in full — the five executing units of the Ministry of Social Development — between 2021 and 2026, and the other thirteen ministries with the same rule. We attribute the money per award: each one gets the share of the normalised amount proportional to its raw total, and only then is it split among its suppliers; in this line no award has more than one supplier, 0 of 8,063, so there is no ambiguous split. In the period the line awarded 23,634,950,532 pesos to 1,354 distinct awardees. 42.4% of that money went to entities whose registered name carries an explicit non-profit legal form (cooperative, foundation, association, welfare society, congregation, parish, rural development society). The second of the fourteen ministries is agriculture, at 10.0%. The other twelve sit at 8.3% or below: education and culture 8.3%, defence 6.8%, the presidency 4.2%, interior 3.8%, health 2.4%, labour 2.0%, industry 1.5%, tourism 1.0%, foreign affairs 0.6%, transport 0.3%, economy 0.1%, housing 0.1%. The mirror of the same fact: only 19.1% of the social development ministry's money went to suppliers with a commercial-company name (S.A., S.R.L., Ltda., SAS, EIRL), the lowest of the fourteen by fifteen points — the second lowest is labour, at 34.0% — while economy reaches 91.2% and housing 87.3%. The 42.4% is a floor, because the rule classifies by name. The ministry's largest supplier, OTRAS MANOS (tax id 216136900010, 1,844,215,526 pesos, 7.8% of the line), is a civil organisation whose name carries no legal-form word and is therefore not counted; it is among 600 suppliers, holding 38.2% of the money, in the same situation: CENTRO DE INVESTIGACIÓN, INFORMACIÓN Y ESCUCHA (722 million), FACTOR SOLIDARIDAD (543), CLAEH (471), CENTRO PARA EL DESARROLLO DE INTERVENCIONES Y ESTUDIOS SOCIALES (445), CENTRO DE EDUCACIÓN POPULAR (392). We verified the mechanism by opening, one by one, the 50 official records of the line's largest awarded purchases of 2026, which add up to 3,374,385,225 pesos, 75.2% of what the line awarded that year — that count comes from reading documents, not from our measurement. Of those 50, 44 are exception purchases invoking verbatim «Art. 33, 30: Contratación de bienes y servicios del MIDES con pequeñas empresas, asociaciones u organizaciones civiles», for 2,783,428,065 pesos, 82.5% of that money; inside the Social Development Directorate it rises to 93.3%. Five are public tenders and one is an exception purchase under a different item. In 42 of those 44 the «participating suppliers» block lists a single organisation, which is the awardee: in an exception purchase there is no call for bids, so that does not mean only one applied. The comparison that opens the question is from the same ministry and the same service, by two routes. Public Tender 5/2024 (purchase 1135850), a call open to entities of any legal form to run up to three 24-hour shelters and one mobile team in Montevideo: eight organisations applied and it was awarded to Fundación A Ganar for 86,842,297 pesos. Public Tender 13/2024 (purchase 1160325), technical teams for the gender-violence services network: ten organisations applied and it was awarded to Fundación Voz de la Mujer for 90,736,740 pesos. Exception Purchase 29/2026 (purchase 1323802), article 33 item 30, running 24-hour shelters for homeless people: a single organisation appears, OTRAS MANOS, for 399,597,517 pesos, and it is the ministry's largest contract of 2026. Exception Purchase 260/2025 (purchase 1332102), same item, up to 700 places in overnight and 24-hour contingency shelters: a single organisation, OTRAS MANOS, 56,750,501 pesos. This is not a practice of this government nor of the previous one: the share of the ministry's money going to non-profit entities by name is 44.6% in 2015, 40.7% in 2019, 50.8% in 2021, 42.5% in 2024, 46.9% in 2025 and 39.9% in 2026; it spans three governments of two parties and predates the current wording of item 30, which is from December 2023. Three things we measured that are NOT a signal, published so they are not read as one: the ministry's ten largest suppliers hold 35.7% of the money, less than health (56.6%), labour (63.4%) or housing (91.1%); of the suppliers that ever sold to line 15, 15.5% sold to no other line, and the median across the 60 lines with more than 100 suppliers is 17.4%, so having exclusive suppliers is ordinary; and 91 awardees are present in all six years, out of 1,354 distinct ones, holding 56.3% of the money, barely above education (53.2%), health (51.1%) and defence (50.0%). The competing explanation was written by Parliament, and it goes here. A 24-hour shelter is not a purchase of paper: there are people inside, and re-tendering every year means changing the team, the key workers and the relationship with people living on the street. Item 30 exists so that continuity of care does not depend on a tender calendar, and it names the social development and agriculture ministries, which are exactly the top two of the table. The entities able to sustain an overnight shelter with educators and social workers are few and are, by design, non-profit: the State decided to deliver social policy through civil society rather than hire civil servants, and that explains why in this ministry commercial companies do not dominate the spending. Every one of these purchases is published with its number, its subject, its resolution in PDF and its awardee on the State's site, and all pass through the Court of Auditors' prior control. Press datum, not our measurement: Fundación A Ganar, awardee of Public Tender 5/2024, has prior coverage over other matters — La Red 21 on a direct agreement of December 2024, El Telégrafo on its total contracting with the State, Caras y Caretas on its political ties; that coverage is about that organisation, not about the mechanism we measure here.
- Which law it would engage
TOCAF (Decreto 150/012), article 33, letter D), item 30, as worded by Ley 20.245 of 29 December 2023: it authorises direct contracting, with no call for bids, of goods and services by the social development or agriculture ministries with cooperatives, associations or civil organisations, in all cases non-profit, under agreements to deliver their sectoral policies. They are the only two ministries named in the item, and they are the top two of the non-profit purchasing table (42.4% and 10.0%; the third drops to 8.3%). Nothing we measured is a breach: the route exists because Parliament created it, and what the card shows is how much it is used and against which alternative it compares.
«La contratación de bienes y servicios que realicen el Ministerio de Desarrollo Social o el Ministerio de Ganadería, Agricultura y Pesca, con cooperativas, asociaciones u organizaciones civiles, en todos los casos sin fines de lucro, en el marco de convenios o acuerdos específicos para el cumplimiento de planes que se relacionen en forma directa con la ejecución de las políticas sectoriales de dichos Ministerios.» — TOCAF, art. 33 lit. D) num. 30, text given by Ley Nº 20.245 of 29/12/2023 (impo.com.uy/bases/tocaf-tcr/150-2012/33). Each purchase record labels it: «Art. 33, 30: Contratación de bienes y servicios del MIDES con pequeñas empresas, asociaciones u organizaciones civiles». Source warning: the old TOCAF version still published at impo.com.uy/bases/tocaf2012/150-2012/33 shows a different text under item 30, about vocational training; the «tocaf-tcr» base is the updated one to cite. Oversight bodies: the state procurement agency and the Court of Auditors.
- What is missing before asserting it
- We could not measure whether the exception costs more than the tender. The subjects are not comparable: Public Tender 5/2024 contracts up to three shelters and a mobile team, Exception Purchase 260/2025 up to 700 places, and 29/2026 shelter management with no published unit. There is no common unit price, so any price comparison would be invented, and this card does not make one. Competition cannot be measured from the precomputed collections: of the 4,173 ocids of line 15 since 2021, the competition collection holds 0, the bidders-per-tender collection 13 and the bidders-per-minutes collection 2; the participant counts we use come from reading 50 State records by hand. The procurement method does not live in the feed either: only 24.4% of line 15's ocids since 2021 have procurementMethodDetails on any sibling release, which is why the records had to be read. The 82.5% is measured on the 50 largest purchases of 2026, which are 75.2% of the year's money: it is a census of the large tranche, not a sample of the whole year, and it cannot be extrapolated to the count of small purchases. 2026 is incomplete: the line's awards run to August and that year's feed is missing a channel. Name-based classification is not a legal register: there is no legal-form field in the database — the supplier registry carries corporate name, address and status, not company type — the rule was applied identically to all fourteen ministries and underestimates the social development ministry more than the rest, so the gap is a floor. And we did not check whether the organisations winning by exception compete against each other in other tenders outside these 50 cases. A trap for anyone repeating the measurement: there is also a buyer.id 24-15 named «Ministerio de Desarrollo Social» with 7 releases, which is line 24 and not 15, and stays out.
- Who has to answer
- The Ministry of Social Development, its General Secretariat and its Social Development Directorate — unit 15-2 concentrates 17,280,329,455 of the line's 23,634,950,532 pesos: why the ordinary route for running 24-hour shelters is the exception purchase and not the open call, when the open call they ran in 2024 attracted eight and ten organisations. The state procurement agency: whether it keeps a record of the use of item 30 by ministry and by amount. The Court of Auditors: whether it observed any of these exception purchases; the ruling is not in the feed, only in its resolutions. And the largest awardee organisations — OTRAS MANOS, Fundación Plenario de Mujeres del Uruguay, Fundación A Ganar, CIIE: whether they would rather compete in a tender.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-mides-cooperativas.verify.ts
27 / 32
Half the foreign money is in the name of companies that appear twice, and our own ranking counts them twice
- What we measured
- First, the scope: almost all the money in this finding is nine companies and a single buyer, the state oil company ANCAP. The large percentage does not describe a problem spread across the State. We measured on supplier_patterns, the collection that feeds our own supplier ranking: 43,000 records and 1,648.63 billion pesos. Suppliers without a Uruguayan tax id, which the official record labels «Empresa Extranjera» and the feed identifies with an X or E prefix, are 1,558 identifiers and 417.42 billion, 25.3% of all the money in the corpus. Before counting duplicates we applied the fix for the other known defect, the 2026 format change that dropped the slash from the identifier: we folded identifiers that are identical once punctuation is stripped, and 122 collapse there. Even after that, 160 names remain split across 344 distinct identifiers, holding 226.79 billion pesos, 54.3% of the foreign money and 13.8% of the whole corpus. No format normalisation rescues them, because what changes is not the number's punctuation but the number. Nine companies are 97% of that money — 220.15 billion, 52.7% of the foreign total — and all nine sell crude or fuel to ANCAP: TRAFIGURA PTE LTD 85.90 billion (X/CHE201176 plus X/SGPGST19-9601595-D); VITOL S.A. 68.50 (X/CHE201056 plus X/CHECHE-116.346.864); GUNVOR S.A. 18.65; SHELL WESTERN SUPPLY AND TRADING 17.07 (X/BHS106835834 plus X/BRB201351); SOCAR TRADING 10.15; SAHARA ENERGY 7.66; PETROBRAS GLOBAL TRADING 5.12; YPF 4.02; GEOGAS TRADING 3.10. The pattern has a shape: 28 identifiers in the corpus are of the form X/{COUNTRY}2xxxxx, a six-digit number between 200052 and 205379 handed out consecutively across nine different country prefixes, adding up to 59.49 billion. Not one of those 28 numbers appears as an identification in the supplier registry the agency itself publishes (June 2026 snapshot, 116,894 records, 2,533 non-Uruguayan): the shortest foreign identification the registry holds is eight digits. The other identifier of each pair, by contrast, matches the registry digit for digit, and we checked it record by record: VITOL S.A. with 116346864 (Switzerland, active), TRAFIGURA PTE LTD with 199601595 (Singapore, active), GUNVOR with 112675659, SOCAR with 113990112, GEOGAS with 106480892, YPF with 30546689979 (Argentina) and SHELL WESTERN with 106835834 (Bahamas). In three cases with large amounts, the country letters of the short number do not match the registry: TRAFIGURA appears as CHE (Switzerland) and the registry says Singapore; SAHARA ENERGY as CHE and the registry says United Kingdom; SHELL WESTERN as BRB (Barbados) and the registry says Bahamas. It is not an artefact of our ingestion: the official record of the ANCAP award of 06/09/2019 reads verbatim «VITOL S.A. (Empresa Extranjera CHE201056)», and the one of 19/02/2021, same firm and same body, «Empresa Extranjera CHECHE-116.346.864». We also closed the null-currency trap on these cases: the 269 items awarded to those identifiers declare dollars explicitly, zero nulls, so none of these amounts is a dollar counted as a peso. The effect on what we publish is direct: in our ranking of 43,000 suppliers, TRAFIGURA sits at number 1 and again at 38; VITOL at 2 and again at 18; GUNVOR at 15 and again at 1,311. The result does not depend on how we group: we tried three name normalisations — exact normalised, without corporate suffixes, and truncated to 22 characters to tolerate the feed's truncated names — and they give 47.2%, 47.2% and 54.3%. The competing explanation goes here, and it is strong. A foreign company registers with a Uruguayan tax id, with its foreign tax code, or with a number the agency itself assigns when it has not yet evidenced its own. The identifier printed on an award is the one in force the day it was awarded, and rewriting an old file to give it a newer number would be altering a historical document: the correct thing is to leave it as it is. The data supports that reading almost point by point. In eight of the nine pairs the two identifiers never overlap, and the short number always comes first: VITOL uses CHE201056 until December 2020 and CHECHE-116.346.864 from February 2021, without a single month of overlap. The awards are published, with the buyer, the date, the amount and the company name written the same under both identifiers. The only thing that breaks is the key by which they are joined, and that key is ours. Even the country letters admit a benign reading: Trafigura's trading desk operates out of Geneva even though the company is Singaporean, and a form asking for a country accepts both answers. Two things that explanation does not cover: SOCAR used the short number in February 2023 and the registered number in 2024, so it is not a backlog closed in the past; and 220 billion pesos, 13% of everything we measure, are split in our own ranking today. One standard-level fact closes the context: the block OCDS defines so that a company stays the same entity across the years, parties[].identifier, is empty in all 2,184,332 records of the corpus.
- Which law it would engage
Being registered in the State Supplier Register is a requirement for any supplier wishing to contract with the State, and not being registered is grounds for incapacity to contract. The register is run by the state procurement agency, and it is where the number identifying a foreign company must be findable. The nine companies are registered and active today: what the published data does not allow is verifying, for the awards carrying the short number, which registration they correspond to.
TOCAF, article 46 (Ley 18.834 of 4/11/2011, art. 46; item 1 amended by Ley 19.355 of 19/12/2015 and general wording given by art. 36 of Ley 19.996 of 3/11/2021), which lists among the grounds of incapacity to contract «Estar suspendido o eliminado del Registro Único de Proveedores del Estado (RUPE)» and «No estar inscripto en el RUPE de acuerdo con lo que establezca la reglamentación». The exceptions in force are in Decreto 202/024 of 10/07/2024, art. 2 item 2. Text at impo.com.uy/bases/tocaf2012/150-2012/46. The register as open data: catalogodatos.gub.uy/dataset/arce-registro-unico-de-proveedores-del-estado-rupe-2026. The standard field: Open Contracting Data Standard 1.1, parties[].identifier.
- What is missing before asserting it
- We do not know whether the short number was a valid identification the agency issued and later withdrew. The open register is a snapshot of the present, June 2026, not a historical series: we cannot see which number VITOL was registered under in 2019. Nor do we know whether that number exists in the register and simply is not published in the open data: we asked the dataset, not the body. Grouping by name does not prove two identifiers are the same legal person; in the nine cases the buyer (ANCAP in all), the currency (dollars across the 269 items), the product and non-overlapping periods in eight of the nine pairs also coincide, but a corporate restructuring inside a trading group would look the same. We do not assert that any of these companies lacked registration: all nine are in the register today, active. We cannot know whether ANCAP performed the registry check on each purchase, because that lives in the file and not in the feed. Of the 160 split names the measurement counts, nine hold 97% of the money and the rest are small cases we did not review one by one: the publishable figure is the money, not the count. And the OCDS field that would settle this at the root, parties[].identifier, is empty in all 2,184,332 records, so nothing in the publication solves it for us.
- Who has to answer
- The state procurement agency, which issues the identifier, publishes the feed and runs the register: whether the six-digit number is one of its own, why it does not reach the register's open data, why in three cases the country letters do not match its own register, and whether it will populate parties[].identifier, the standard field that closes the problem. ANCAP, the buyer in all nine: which registered supplier each award corresponds to. And ourselves, the other half of the problem: nobody forced us to group by the literal identifier string, and until we fold these identifiers our ranking shows TRAFIGURA at number 1 and again at 38, and VITOL at 2 and again at 18. Publishing the finding without fixing the page is stopping halfway.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-extranjeros-partidos.verify.ts
28 / 32
Montevideo's city government did not vanish from 2026: its purchases travel through a channel the open data publishes once a year
- What we measured
- In our corpus, Montevideo's city government has 29,079 releases in 2024, 19,584 in 2025 and 82 in 2026. Canelones, 1,669 / 1,554 / 0. The National Ports Authority, 947 in 2025 and 0 in 2026. The other 17 departmental governments have a normal 2026: Paysandú 937, Colonia 449, Florida 330, Maldonado 238. It is not a departmental-government problem: it is three bodies. Those bodies publish through a second agency channel whose identifiers carry an «i» prefix (ocid ocds-yfs5dr-i483583), and that channel does not appear in the OCDS RSS. We downloaded and parsed the feed: ocds/rss/2026/07 carries 14,556 releases and in it there are zero identifiers with an «i» prefix and zero award cancellations — while in the database cancellations are 3,417 in 2024 and 3,146 in 2025. Nor are they in the per-release API: GET /ocds/release/adjudicacion-i483583 returns 404 even though it is a 2025 record we already hold, and the control GET /ocds/release/adjudicacion-1307316 returns 200. The only place it lives is the annual archive: on catalogodatos.gub.uy, the agency's historical procurement dataset has 24 resources and the last one is 2025 — ocds-2023.zip created 15/01/2024, ocds-2024.zip 27/02/2025, ocds-2025.zip 19/01/2026; ocds-2026.zip does not exist. How much it weighs: measuring by the ocid prefix and with the normalised amount between 0 and 50 billion, the «i» channel is 59.0% of the spending recorded in 2023, 59.5% in 2024 and 55.4% in 2025. In 2026 it is 0%. Seen through the RSS arrival marker, which only exists from 2025: of the 178.1 billion pesos recorded in 2025, 116.2 billion (65.2%) arrived in releases the RSS never published, and within that tranche the oil company contributes 66.8 billion, the telecoms company 11.6, Montevideo's city government 9.8, the state casinos board 7.2, the water authority 6.1, Canelones 2.0 and the ports authority 1.9. That 65.2% cannot be flipped into «2026 is missing 65% of the money»: annual totals are lumpy — a single award by the power company moves a whole month — and 2026 already adds up to 100.0 billion pesos through RSS alone. The figure that holds is the channel's: between 55% and 59% of the spending recorded in each of the three closed years travelled through a channel that 2026 does not have. It is not by body, it is by purchase: the power company runs 94.6% through RSS, Montevideo's city government 0.6%, and Canelones and the ports authority 0.0%; with at least 100 releases in 2025, only those three bodies have under 5% presence in the RSS. And the data exists today on the State's site: the same query to the agency portal for budget line 98 unit 1, which is Montevideo's city government, returns 4,085 publications for all of 2025 and 3,258 between 1 January and 14 August 2026, with detail pages hanging off the same identifier space; there is also a portal RSS filterable by body which returned 106 items from the city government over the last seven days, 106 of 106 with an «i» identifier, among them Abbreviated Tender A193590/2026 published on 13/08/2026. The portal's units and ours are not comparable to each other: the 4,085-to-3,258 pair works as the portal's internal evidence that the city government keeps publishing, and the 19,584-to-82 pair works as evidence of what the open data shows. The competing explanation goes here, and it is strong. The agency publishes two things because they are two things: a real-time OCDS feed of purchases processed inside the central procurement system, and a consolidated historical archive that also includes the purchases certain bodies report rather than process. Those bodies have contracting autonomy and their own systems; the agency receives their data as a periodic report, not as a live transaction, so there is nothing to emit in real time. The annual archive was published every year: January 2024 for 2023, February 2025 for 2024, January 2026 for 2025. And meanwhile no purchase goes unpublished: tender A193590/2026 has been on the site since the day before this measurement. What is missing is not public information, it is a bulk-download format for a year that has not ended yet. What stands after conceding that: the cost falls on whoever looks at the current year. You can see one tender by the city government; you cannot see its 2026 spending pattern, or compare it, or detect an anomaly, until 2027. And the data is already structured, already has a stable identifier and already has its own per-body RSS on the same site. This includes us, with a date and a line of code. Our spending-trend page marks 2026 as partial, but for the wrong reason: src/jobs/refresh-spending-trend.ts:811 sets «partial: year >= currentYear» and the text says it is a year in progress, not that it is missing an entire channel. On top of that, our site's 2025 line was incomplete until 16 July 2026 — the 2025 «i» channel releases all entered in a single run that day, and the archive had been published since 19 January — and the 2024 ones entered on 7 August 2025, with the archive available since 27 February. Importing the annual archive is manual (scripts/sync-year-zip.ts), it is not on the cron, and nothing warns when the new archive appears.
- Which law it would engage
Ley 18.381 (Right of Access to Public Information), article 5, permanent and proactive disclosure on websites, including tenders. That obligation is being met: Montevideo's city government purchases are on comprasestatales.gub.uy today. The gap is not one of publicity but of open data: the agency's commitment to publish in the OCDS standard was born in the 3rd Open Government Plan (2016-2018) and was reinforced in the 5th Plan (2021-2024) with the commitment to strengthen the openness, visualisation and download of public procurement data. What fails is that channel's cadence, not the duty to publish.
Ley 18.381, art. 5: «Los sujetos obligados deberán prever la adecuada organización, sistematización y disponibilidad de la información en su poder, asegurando un amplio y fácil acceso a los interesados. Los organismos públicos, sean o no estatales, deberán difundir en forma permanente, a través de sus sitios web u otros medios que el órgano de control determine, la siguiente información mínima: (…) Concesiones, licitaciones, permisos o autorizaciones otorgadas, especificando los titulares o beneficiarios de éstos.» Documentary complement, not our measurement: in the OCP Data Registry (data.open-contracting.org) the publication «Uruguay: ARCE (Historical)» is listed with update frequency «Annually» and range «Jan 1970 - Dec 2024» (last retrieved 12/01/2026); the «ARCE (API)» publication is listed as «Real time», range «Nov 2017 - Aug 2026».
- What is missing before asserting it
- What is missing is for the «i» channel — the purchases that state enterprises and departmental governments report to the agency instead of processing inside the central system — to be published as open data at the same speed at which it is already published in HTML on the portal itself. Concretely: that releases with an «i» identifier enter the monthly OCDS RSS, or at least that GET /ocds/release/adjudicacion-iNNNNNN stop returning 404, today even for 2025 records the agency has already published in its annual archive; that the consolidated archive stop being annual, with a monthly or quarterly cut that makes the current year analysable; and that award cancellations enter the live feed, because they are 3,417 in 2024 and 3,146 in 2025 and zero in the RSS of any month. On our side, what is missing is labelling the current year for what it actually lacks — a whole channel, not the fact that the year has not ended — automating the annual archive import with an alert when it appears, and evaluating ingestion of the portal's per-body RSS so as not to depend on January. A limitation of the measurement: we only began marking RSS arrival in 2025, so 2023 and 2024 come out 100% unmarked by construction, and the channel's weight in those years has to be read from the ocid prefix, which is what we do. And the agency's side of the story is missing: we did not ask before measuring.
- Who has to answer
- The state procurement agency, which runs both channels and decides the consolidated archive's cadence. Three questions: why reported purchases do not enter the OCDS RSS or the per-release API, given they already have a stable identifier and are already published on the portal the same day; whether a cut more frequent than annual is planned; and when ocds-2026.zip will be published. And it answers to us. We publish a spending-trend page that shows 2026 with a year-in-progress label, when what is measurable is that 2026 lacks the channel through which between 55% and 59% of the spending recorded in each of the three previous closed years travelled. And we published a closed 2025 that was missing the entire channel until 16 July 2026, and a 2024 that was missing it until 7 August 2025.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-im-canal-anual.verify.ts
29 / 32
Our anomaly detector looked at a 12,265-million purchase by the power company and did not flag it: it measures unit prices and never looks at quantities
- What we measured
- The entry case is already published on the card about the power company's drivers and is not repeated here: release ocds-yfs5dr-1315467 (UTE, ESTILO S.R.L., 21/05/2026) has a single item, «SERVICIO DE CHOFER», classification 32917, quantity 27,000,000 units, unit price 454.29 pesos and an amount of 12,265,830,000 pesos, while the minutes say 27,000,000 plus VAT. What we measure here is why our tool did not flag it: the anomalies collection returns zero documents for that release. We audited the code in its 14 August 2026 version. The detector scores a single number: the unit price. Quantity appears in two places in the job and in neither can it raise a flag — in the function that discards the line-total artefact, where it only serves to suppress a flag, and in the metadata of the already-emitted document, where it is decoration. There is no rule, in the job's 1,125 lines or the statistical module's 656, that compares a quantity against anything. We re-ran the production scorer, without touching the database, against the live baseline for classification 32917 in pesos and unit «unidad»: n=12, p25 59,683.11, p50 223,095.49, p95 3,800,439.46, minimum 454.29. It returns null, because the rule is that if the price does not exceed the p95 there is no spike. The 454.29 pesos did not merely pass the filter: they are the lowest unit price in the whole category, and that baseline's minimum is this very purchase. The detector looked at it, took it into its own price reference and classified it as cheap. The problem was never the price. How many more share the shape: within the detector's own window, the last 24 months, lines with a quantity of 100,000 or more and quantity times unit price of 100 million pesos or more are 46 lines across 42 records, for 513,859,359,636 pesos. Flagged: zero. Restricted to records below the 50-billion validity ceiling, which is what does enter our published totals, they are 43 lines and 50,465,193,437 pesos. The coverage comparison: the anomalies collection holds 6,307 documents, all of the price-spike type, and adding up their line value — detected price times quantity — everything the detector flagged in its history is worth 10,846,892,139 pesos; a single line it did not flag, the power company's, is worth 12,265,830,000. The largest quantity appearing in any of the 6,307 flags is 500,000, and only 3 of 6,307 exceed 100,000 units. There is a subset that needs no statistics: filtering units that are calendar periods (MENSUAL, ANUAL) bought 100,000 times or more leaves 7 lines for 9,059,627,358 pesos, all 7 from the power company, and none is flagged. It includes 9,600,000 «MENSUAL» units of security services at 329 pesos, which is 800,000 years, and 4,146,624 «ANUAL» units of premises cleaning at 358.42, which is 4.1 million years. The missing anomaly type is already declared everywhere except where it is needed: «outlier_quantity» is a valid value in the Mongo schema (shared/models/anomaly.ts), in the shared types, in the public API enum, in the MCP server schema, and in the frontend it has a colour, an icon, a label and its own number formatter that deliberately does not use currency format. Documents of that type in the database: 0 of 6,307. What it drags downstream: the data-error page filters by the AI verdict's category, which only exists on an already-flagged anomaly, and one of the five per-body signals — unexplained prices — is computed over flagged anomalies; a data-entry error that does not produce a price spike cannot reach either surface. The competing explanation goes in whole, because almost everything the detector does is well built. First, it never promised this: it emits a single type, price spike, and its documentation says what it replaced and what it measures; a spike detector that does not flag a correct price is answering the question it was asked, and 454.29 pesos an hour for a driver and 395 for an administrative worker are ordinary rates, so it got both right. Second, a naive quantity detector would be a source of false positives: the State really does buy 200,000,000 litres of fuel oil, 18,521,228 kilos of sulphuric acid, 1,245,000 tablets and 4,568,020 school meals, and a rule like «a quantity above one million is an anomaly» would flag the oil company supplying a thermal plant and the primary-schools authority feeding schools. Third, the exact rule that let this case through is a measured correction: the upper-tail floor was added because it removed 24% of the unexplained-price set, and everything it removed sat below the 95th percentile, that is, by definition was not overpriced; removing it to catch this purchase would reimport that noise. Fourth, the record was never hidden: it appears on the site with its amount, in the totals, in the power company's ranking and on the supplier's page. And fifth: the power company's minutes say «Valor hora/persona 454,29» and award up to exhausting the awarded amount, while the form asks for quantity and unit price, and a contract awarded by amount has no quantity.
- Which law it would engage
No rule demands anything of a private transparency site, and this card imputes no legal breach to anyone. What is at stake are three layers. The first is ours: when we publish an anomalies page and a per-body signal called unexplained prices, the reader understands there is something watching the spending; what there is, is something that watches unit prices upwards, and that distinction was written nowhere visible. The second is the source data's, and there a citable technical standard exists: the feed declares that it follows OCDS 1.1, and the standard defines quantity as the number of units to be provided and unit.value as the monetary value of a single unit; a quantity field carrying pesos is data non-conforming with the schema the feed itself says it implements, and that matters because checking conformance needs no statistics — 9,600,000 «MENSUAL» units is invalid reading only the field definition. The third we already invoked on the drivers card and it is repeated only as context: the publicity duty of TOCAF article 50 and the ceilings of articles 33 and 44. We do not run that second check either, and we had the data: we already store each tender's procurement method and we already use the ceilings table on another published card.
Open Contracting Data Standard 1.1, release schema, Item object: Item.quantity, «The number of units to be provided»; Item.unit, «A description of the unit in which the supplies, services or works are provided (e.g. hours, kilograms) and the unit-price»; Item.unit.value, «The monetary value of a single unit» (standard.open-contracting.org). Uruguay publishes its procurement feed as OCDS through the state procurement agency. TOCAF, Decreto 150/012: art. 50 (duty to publish the award act), arts. 33 and 44 (per-procedure ceilings); ceilings in force per the statistics institute, «Montos Límites de las Adquisiciones Estatales, Enero–Diciembre 2026» (30/12/2025, Ley 20.446 of 16/12/2025): abbreviated tender 13,705,000 pesos and extended abbreviated tender under art. 44, 68,525,000 pesos. Audited code, in this repository and as of the measurement date: src/jobs/detect-anomalies.ts (1,125 lines), src/jobs/anomaly-stats.ts (656), shared/models/anomaly.ts, shared/types/database.ts, app/composables/useAnomalies.ts, app/server/utils/openapi.ts, packages/mcp/src/tools.ts.
- What is missing before asserting it
- The quantity detector still has to be written. The type already exists in the schema, in the public API, in the MCP server and in the frontend with its own colour, icon and formatter; the job that emits documents is what is missing. The minimal version needs no statistics: if the unit is a calendar period and the quantity exceeds what a contract can last, it is invalid reading only the field definition, and today that is 7 lines for 9,059,627,358 pesos. Cross-checking the published amount against the procedure's legal ceiling is missing: we already have the procurement method per tender and we already use the statistics institute's ceilings table on another card. Persisting what today is only printed to the console is missing: the function that suppresses flags for the line-total artefact reports how many it suppresses and does not store it, so we cannot say how many anomalies are suppressed that way without running the job, which writes; we declare that as not measured. We did not verify as impossible the remaining lines in the set of quantity above a million with a low unit price: inside there are real purchases, such as 200,000,000 litres of fuel oil from the oil company to the power company and 4,568,020 primary-school meals, which is why the figure we publish as the finding is 7 lines and 9,059,627,358 pesos, and as the audited unflagged population, 46 lines and 513,859,359,636. Nor do we publish the whole-history figure, dominated by 2003-2005 records that exceed the validity ceiling and fall outside our totals. And asking the agency whether the procurement system validates the quantity field against anything, and whether the OCDS feed passes any schema-conformance check before publication, is still missing: that goes through a freedom-of-information request, not through observation.
- Who has to answer
- Ourselves, first. The detector is ours, the decision to score only unit prices and only upwards is ours, and the 46 lines it let through in its own window it let through under rules we wrote and documented. What follows is writing the missing detector, saying on the anomalies page itself which question it answers and which it does not, and publishing this card before writing it, not after. The state procurement agency, owner of the procurement system and responsible for the feed, for the source layer: if the quantity field accepts pesos, no consumer of the feed can tell a quantity from an amount without downloading the resolution in PDF; we do not measure third-party tools and assert nothing about them, what we say is that the field is the same for everyone. The power company, only as the body that concentrates 12 of the 13 records of 2026 with this shape; the substantive complaint about that data entry is already made on the drivers card and is not repeated. And one contextual fact that is a reproach to nobody: in the case that triggered this, the Court of Auditors' delegated accountant observed the expenditure at 27,000,000 pesos, while the public site and our own detector were working on 12,265,830,000.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-anomalias-cantidad.verify.ts
30 / 32
What the State buys small is published the same day; what it buys big, months later
- What we measured
- We measured the distance between the date of the award resolution and the date that award was published on the state procurement site, across the 480,178 awards published between 1/1/2022 and 13/8/2026. First we validated that the release date field is the publication date, rather than assuming it: in the 221,595 records that carry the attached minutes, that date matches the publication date of the award document exactly in 221,547 cases, 99.98%; we also checked two records against the State's site and they match to the minute. Of the 480,178, 4,502 carry no award and fall out: we measured 475,676. The resolution date comes as a date in 420,045 and as text in 55,631, all from 2025 and all parseable; 18 records carry impossible years, which ruins any average, which is why we work with medians and thresholds and never with means; 265 records (0.06%) have publication earlier than the resolution. The median is 1 day, the 75th percentile is 7, the 90th is 35 and the 95th is 88. Across the whole universe, 79.57% is published within ten days (378,512): 97,164 fall outside, one in five. With fourteen calendar days, the ceiling of ten working days without holidays, it rises to 82.99%. 23,240 awards (4.89%) took more than 90 days and 5,155 (1.08%) more than a year. Article 50 only requires publication above 20% of the direct-purchase ceiling: restricted to that band, using 130,800 pesos — 20% of the 2026 ceiling applied flat over the whole period — across 145,985 awards compliance falls to 68.95%, with 45,332, almost three in ten, out of time, and it falls year by year: 70.2% in 2022, 72.8% in 2023, 68.7% in 2024, 67.8% in 2025 and 60.3% in 2026. The finding is the gradient: the more formal and the larger the purchase, the later the public finds out. Crossing each award with its sibling tender document by ocid (126,594 cases, 26.6% of the total): direct purchase, 89,705 awards, 88.1% within ten days, median 0 days, p90 14, over 90 days 2.1%; exception purchase, 2,791, 68.6%, median 3, p90 40, 3.1%; price competition, 19,521, 59.9%, median 7, p90 46, 3.9%; abbreviated tender, 12,216, 30.3%, median 27, p90 137, 16.6%; public tender, 1,403, 15.7%, median 56, p90 208, 32.4%. The same by amount: under 100,000 pesos, 84.6% in time and a median of 1 day; between 100,000 and a million, 76.5% and 2 days; between one and ten million, 45.3% and 14 days; between ten and a hundred million, 29.8% and 33 days, with a p90 of 153. The band above a hundred million gives 43.4% and we do not use it: it is contaminated by the lump-sum artefact we already documented. Two cases verified against the State's official record, not against our database. The water authority OSE, Public Tender 17311/2021, «Contratación de Servicio de Seguridad y Vigilancia en el Departamento de Rivera», awarded to SECURITAS URUGUAY S.A. for 212,557,261.48 pesos: the record says «Fecha de la Resolución: 27/07/2021» and «Publicación: 21/04/2025 23:02hs», 1,364 days. The state casinos board (finance ministry), exception purchase under article 33 item 3, awarded to CIA RIOPLATENSE DE HOTELES S.A., resolution of 24/03/2023, published on 17/12/2025 at 14:25, 999 days; for this second one we cite no amount, because the record carries the lump-sum artefact pattern. There are bulk uploads and they must be read as such: on 19/03/2024 the state insurer published 3,261 awards in one day, 2,372 of them more than 90 days after the resolution, and that same day Montevideo's city government published 2,646, the oldest 753 days back, plus 2,331 on 15/07/2024 and 1,262 on 01/10/2024. A good share of the 2024 drop in the total is those two bodies catching up. Recent behaviour by body, 2025-2026, only the 135 with 200 awards or more: Veterinary Faculty 6.3% (287), Science Faculty 7.6% (370), Armed Forces Health Service 8.1% (904), Northern Littoral University Centre 8.6% (278), Florida departmental government 11.6% (553), the technical education council 12.9% (1,220), Agronomy Faculty 17.2% (221), the water authority 18.4% (1,962), the secondary education council 21.2% (481), Colonia departmental government 22.1% (734). At the other end: the Soriano (648) and Lavalleja (427) police headquarters at 100%, San Carlos hospital 99.8% (2,641), Maldonado police headquarters 99.8% (825). The competing explanation is strong, it is measured, and it goes here. Article 50's clock starts «luego de producido el acto», and in a formal procedure the act does not end the day it is signed: the resolution goes to the Court of Auditors for prior intervention; if the Court observes it, the body reiterates; only then are bidders notified. A body may read «the act» as the final, intervened act and publish at the end of that chain. The data supports it: of the 221,595 releases with attached minutes, the 5,134 that carry an expenditure-reiteration document — that is, those the Court observed — meet the ten days 26.7% of the time, against 70.7% of the 216,461 that do not, and exceed ninety days 14.9% against 7.7%. It also explains why the gradient follows the procedure and the amount: the larger the purchase, the longer the chain of prior controls. On top of that, the article says ten days without clarifying whether they are working days, the 2024 spikes have the shape of an administrative catch-up and not of concealment, and a small body can fall behind because of a leave or a system change. Where that explanation does not reach: article 50 itself requires publishing, besides the award act, the extensions and the acts reiterating expenditure observed by the Court of Auditors, so passing through the Court adds a publication, it does not suspend the first; cases with reiteration are 2.3% of the universe with minutes, so they explain the shape of the tail and not the bulk of the volume; and if the problem were only how the deadline is counted, we would expect publications at 20, 40 or 60 days, not also 5,155 published more than a year later and a 212-million-peso public tender published three years and nine months after its resolution.
- Which law it would engage
TOCAF (Decreto 150/012), article 50: it requires publishing the award act on the state procurement site within ten days of the act, for every procedure whose amount exceeds 20% of the body's direct-purchase ceiling, including exception mechanisms, extensions and the acts reiterating expenditure observed by the Court of Auditors. Two clarifications that bear on what can be asserted. The article says «diez días» and does not clarify whether they are working or calendar days, while articles 51 and 52 do say «días hábiles» when they mean working days: which is why we give both figures, 79.57% at ten calendar days and 82.99% at fourteen. And below 20% of the ceiling there is no obligation to publish at all, so the direct purchase's 88.1% compliance is partly self-selection: what is easy to publish and what nobody requires gets published fast. The number the rule addresses is the other one, 68.95%.
TOCAF, art. 50, second paragraph: «Todas las administraciones públicas estatales deberán dar publicidad en el sitio web de compras y contrataciones estatales, al acto de adjudicación, declaración de desierta o de rechazo de ofertas, a todos sus procedimientos de contratación de monto superior al 20% (veinte por ciento) del límite de su procedimiento de compra directa, incluidos los realizados por mecanismos de excepción, así como las ampliaciones y los actos de reiteración de gastos observados por el Tribunal de Cuentas, en la forma que disponga la reglamentación. Estos organismos contarán para ello con un plazo de diez días luego de producido el acto que se informa». Verified at impo.com.uy/bases/tocaf2012/150-2012/50 and reproduced in the state procurement agency's procedures manual. Drafting contrast: arts. 51 and 52 of the same text do specify «días hábiles». Direct-purchase ceiling in force January-December 2026: 654,000 pesos (departmental governments 1,024,000), published by the agency; 20% is 130,800.
- What is missing before asserting it
- Survivorship bias, which is the main thing. We can only measure what was published: an award that was never published is not in the corpus and by definition never appears as late, so the 31% non-compliance in the mandatory band is a floor. Closing it requires cross-checking against another source of the spending and seeing how many resolutions never reached the portal; we did not do that. We cannot distinguish «resolved and published the same day» from «the system copied the date»: 178,782 records (37.6%) have a zero difference, and some bodies are at 99% or 100% zero difference; it may be genuine — in a direct purchase the act and its recording are almost the same act — or it may be autofill, and either way the error runs in one direction only, it inflates compliance. The cut by procedure holds over 26.6% of the universe: only 126,594 of 475,676 awards have a sibling tender document by ocid, what is lost is mostly extensions and exception purchases to which the agency gives a new identifier, and the gradient is monotonic and large but the exact percentages come from that subset. Working days versus calendar days is not settled by data: we give both figures and do not choose. The legal threshold is approximate: we use 130,800 pesos, 20% of the 2026 ceiling, flat over 2022-2026, when the ceiling is indexed each year and is different for departmental governments, so the subset we measure is inside the obligation but does not exhaust it. The cut by amount is affected by two known corpus traps, awards with a null currency counted as pesos and the lump-sum artefact in the upper band, which we report and do not use. 2026 is distorted — a feed channel is missing and Montevideo's city government has a handful of awards in the whole year against fifteen thousand in 2025 — so 2026's 60.3% is not comparable and does not go in the headline. And we asked nobody for their side: not the agency, not the Court of Auditors, not the bodies.
- Who has to answer
- The state procurement agency, which runs the site and defines the publicity regime: whether it measures and publishes compliance with article 50's deadline; whether there is any control over a publication made 1,364 days after the resolution, which the system accepted without flagging; and whether it will publish the per-body compliance metric, which it can compute from its own data. Also, its position on whether the deadline counts working or calendar days, and from which act. The bodies of the 2025-2026 period: the university, for its Veterinary Faculty (6.3%), Science Faculty (7.6%), Northern Littoral Centre (8.6%) and Agronomy Faculty (17.2%); the Armed Forces Health Service (8.1% over 904); the technical education council (12.9% over 1,220); the Florida departmental government (11.6%); the water authority (18.4% over 1,962, and the Securitas case of 1,364 days); the power company (23.8% over 1,633, with 69 published more than a year late); the state casinos board (and the Rioplatense case of 999 days). The state insurer and Montevideo's city government, over the 2024 episode: what happened for 5,907 awards to be published on the same 19 March. The Court of Auditors: whether delay in publicising the award act is a matter for observation and whether it ever observed one. We have the body and the buying unit, not the person: there is no proper name in this finding.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-plazos-publicacion.verify.ts
31 / 32
The mandatory supplier register holds up: 99.79% of what was awarded since 2024 went to active suppliers, and crossing without aligning dates invents 75 times more non-compliance
- What we measured
- This is a negative result and it is published as such: the mandatory supplier register is being complied with. What the card contributes is the method and the trap. Universe: 285,925 award releases with a supplier since 1/1/2024, yielding 372,772 award-by-supplier pairs — 164,921 from 2024, 144,640 from 2025 and 63,211 from 2026 — across 14,418 distinct private tax ids plus 80 identifiers that are public bodies acting as suppliers (926 pairs, 0.25%), which stay out of the denominator because they are not subject to the register. The register: we downloaded the eleven monthly snapshots the agency publishes on the national open-data catalogue — January, April, July and October of 2024 and 2025; January, April and June of 2026. The register goes from 100,618 suppliers in January 2024 to 116,894 in June 2026, and publishes four statuses: in the June 2026 snapshot, ACTIVE 57,759, IN REGISTRATION 39,659, TAX-AUTHORITY DEREGISTERED 19,431 and VOLUNTARILY DEREGISTERED 45. Cross-check aligned in time, by tax id digits, looking for each award's immediately preceding and immediately following snapshot: across 371,846 private-supplier pairs, 371,049 (99.79%) are ACTIVE in at least one of the two; 206 pairs (0.055%) are not active in either; 591 pairs (0.159%) have no record in either. Opening those cases: 31 awards from 18 tax ids deregistered in both snapshots, 175 from 103 tax ids IN REGISTRATION — which the rule contemplates and which is not non-compliance — and 591 from 155 tax ids with no record. Applying the amount exemption the rule itself grants — contracts below 35% of the direct-purchase ceiling do not require registration, and 43.4% of the pairs sit above that threshold — what remains is 14 awards from 9 deregistered companies, 92 awards from 32 companies with no record, and 33 from 20 IN REGISTRATION. The trap: crossing the June 2026 register against awards from 2024 onwards, without aligning dates, produces 2,331 awards to 365 deregistered companies. Aligned in time they are 31: a factor of 75. The artefact's signature is in the shape of the curve: 343 tax ids deregistered by the tax authority among 2024's awardees, 135 in 2025 and 7 in 2026 — the curve of «they closed afterwards». Moreover the status almost never reverses: of the 16,602 deregistered in January 2024, 16,328 remain the same in June 2026 and only 205 (1.2%) returned to active, which is why the preceding and following snapshots suffice as proof. We checked the two largest cases against the government record, and one collapsed. Purchase 1099770 (Colonia departmental government, abbreviated tender, kerb and gutter work in Nueva Palmira) was the largest case by amount, with a deregistered status: the official record awards it to HECTOR JAVIER BELEN DUPUY SAS, tax id 060126020012, which is active in the register, while the feed stored the sole trader's tax id, 060056190011, which is indeed deregistered; they are two records under the same surname and the feed carries the wrong one. The other holds: purchase 1060457 (public health service, National Rheumatology Institute, abbreviated tender 99/2023, cleaning services) was awarded to COOPERATIVA DE TRABAJO COMPROMISO SOCIAL, tax id 215146320019, and that tax id has no record in any of the eleven snapshots, with 16 awards since 2024. A firm side fact: 1,410 awards since 2024, for 168,325,943,373 pesos, went to suppliers without a Uruguayan tax id, 34.8% of what was awarded in the period, and the rule exempts them from the register; they are almost all fuel traders. More than a third of the money awarded goes to suppliers about which the mandatory register says nothing, by legal design. And what the open register does not publish is what matters most: articles 5 and 6 of the decree place final sanctions and bans on contracting inside the register, but the monthly file carries only the four administrative statuses, so open data cannot answer whether anyone awarded to a sanctioned or banned supplier. The explanation that competes with any scandal reading goes here, and it wins. The register is a snapshot of the present and awards are the past: a company that sold to the State in 2024 and closed its tax id in 2025 shows today as deregistered without anyone having done anything, and that explains 2,300 of the 2,331 awards the unaligned cross flags. The rule exempts small contracts from registration, and a good share of the absentees are suppliers of purchases worth a few thousand pesos. «IN REGISTRATION» is not non-compliance: article 12 places validation of the registration after the bid opening in the contracting body's hands, and tender documents give two working days after notification to become active, a window a monthly register cannot see. A sole trader who becomes a company has two tax ids and two records, the old one stays deregistered and the feed may store the old one, which is exactly what happened with the case that collapsed on us; the same happens with estates. And public bodies buy from each other: the interior ministry, the health service's Rivera branch or the national internal audit office appear as suppliers with an in-registration record or none, and those are transfers between public offices. Search-engine datum, not verified first-hand: a 2026 piece on uruguayaldia.com.uy asks how the register keeps companies under attachment and in debt listed as active, citing the Walfur S.A. case; it is a different angle — the solvency of an active supplier, not registration — and the URL returned 404 when we tried to open it.
- Which law it would engage
Decreto 202/024, of 10 July 2024, published on 23 July, which implements article 523 of Ley 15.903 (article 76 of the TOCAF) and repeals Decreto 155/013 in its article 29. It is the register's rule in force, and it must be cited: most tender documents and press pieces still cite 155/013, which no longer applies. Three articles decide how the cross-check reads. Article 2 makes registration a requirement to contract, but exempts contracts below 35% of the direct-purchase ceiling, foreign suppliers not domiciled in the country under certain grounds, and whatever the agency expressly exempts. Article 12, last paragraph, places on the contracting body the validation and approval of the registration after the bid opening for those in the process of registering, which is what makes the IN REGISTRATION status legitimate at the moment of award. And articles 5 and 6 establish that the register contains final sanctions and bans on contracting, none of which is in the open data.
Decreto N° 202/024 of 10/07/2024, art. 2: «La inscripción en el RUPE constituye un requisito para contratar con las Administraciones Públicas Estatales, aplicable a todos los procedimientos de contratación que realicen, con independencia de su fuente de financiamiento. Quedan exceptuados de este requisito las contrataciones menores al 35% (treinta y cinco por ciento) del tope de la compra directa establecido en el literal C del artículo 33 del TOCAF sin considerar la excepción dispuesta para los Gobiernos Departamentales…». Art. 12, last paragraph: «Efectuada la apertura de ofertas, el organismo contratante tendrá a su cargo la validación y aprobación de la inscripción en el registro de aquellos interesados que se encuentren en el proceso de inscripción o actualización de información». Art. 29: repeals Decreto N° 155/013 of 21 May 2013. Source: impo.com.uy/bases/decretos/202-2024. Register: catalogodatos.gub.uy/dataset/arce-registro-unico-de-proveedores-del-estado-rupe-2026 and the 2024 and 2025 datasets. Ceilings used: 263,000 pesos in 2025 and 654,000 in 2026 (raised by article 28 of Ley 20.446); for 2024 we used the 2025 ceiling, which is an upper bound and therefore a stricter threshold, flagging fewer cases and not more.
- What is missing before asserting it
- The open register publishes neither sanctions nor bans, which is what articles 5 and 6 of the decree require to be recorded. Open data cannot answer whether anyone awarded to a sanctioned or banned supplier: that is a freedom-of-information request. The feed does not guarantee that the supplier's tax id is the awardee's, and it is verified: in purchase 1099770 the feed stores the sole trader's tax id and the government record awards to their company, which is another tax id and another record with another status; until that is measured at scale, no name on this list is published without opening its record, which is why this card names a single case, the one we verified. The amount per supplier is missing: the normalised amount belongs to the release and when there are several awardees attributing the total to each inflates everything, so the amounts cited are upper bounds. Temporal resolution is missing: the snapshots we use are quarterly and, when the award falls in a snapshot month, the before and after collapse into one point; with the twelve files of each year, which are published, the window drops to a month. We did not query the live register: a supplier's public page on rupe.gub.uy would say whether the tax ids without a record are registered and the file simply does not carry them, or whether they are not, and without that «absent from the published register» is not the same as «not registered». We did not control for the null-currency trap: it does not affect the headline, which is a count of awards and not a sum of money, but it can move the amounts cited. And we identified public bodies acting as suppliers by the identifier prefix, which recognises 80: that is a floor, and some of the 155 tax ids with no register record may be public offices the prefix did not mark.
- Who has to answer
- The state procurement agency, responsible for the register under article 4 of the decree, on two things: why the monthly open-data extract does not include the sanctions and bans the decree itself requires to be recorded, nor suspension statuses; and whether the 32 tax ids that appear awarded above the obligation threshold with no record in any of the eleven snapshots are registered in the live register. The buying bodies on the short list, one by one and only after verifying each record: the public health service / National Rheumatology Institute (Cooperativa de Trabajo Compromiso Social), the state bank, the state casinos board, the Paysandú departmental government, the Colonia departmental government, Montevideo's city government, the child welfare institute, the water authority and the state insurer. The question is the one article 3 asks: verifying the registration is their responsibility, so how did they do it on that purchase. The Court of Auditors, whether any of these awards was observed for lack of registration: in the 2,464 resolutions we hold there is none that mentions the register, but we only have the metadata, not the text of the ruling. And ourselves, on method: we publish this negative result because the naive cross-check — the one anyone would do, and the one we did first — returns 2,331 cases where there are 31, and that difference is the card.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-rupe-activo.verify.ts
32 / 32
A legal price cap for a single product: the reference prices the procurement agency almost never published, and that the Budget deleted on 1 January 2026
- What we measured
- For five and a half years two statutory provisions said the price to be paid could not exceed «los precios publicados por la Agencia Reguladora de Compras Estatales». We measured what was published. On the document side: the agency's complete announcements archive holds 57 entries, and exactly 2 are reference prices — both for the same product, fresh milk: finance ministry Resolution 104/023, in force from May 2023, and 162/024, in force from June 2024. A third entry indexed by search engines returns HTTP 403 today, «access denied». The other 55 are framework agreements, extensions, annual procurement plans, and notices about the procurement and supplier systems: none carries a price. The price lives inside an .xlsx that is an order form: in the June 2024 one, the «Precio ajustado a la fecha» column reads $42.34 per litre «delivered by the supplier» and $37.26 «collected at the plant». There is no CSV, no API, no historical series, no HTML table. And the agency does not calculate the price: the attached PDF is finance ministry Resolution 162/024 published in the Official Gazette, which sets the maximum sale prices of «leche pasteurizada envasada en bolsitas de polietileno, con un tenor graso no inferior a 2,6%» across the country — over the counter $43.00, home delivery $43.30, to the retailer $42.34, at the plant loading bay $37.26; it is the general price control on regulated milk, and the agency relays it. The announcement itself cites art. 8 of Decreto 130/008: «el precio de la leche será actualizado en forma semestral por el Poder Ejecutivo». Between May 2023 and today about seven half-yearly adjustments fall due; the agency published two announcements. There is also a loose file on the agency's server, «Resol Precio Leche Fresca 19.3.2025.pdf», which we did not find linked from any section: we downloaded it and it has 0 font objects and 2 images — it is a scan with no text layer, not machine-readable. In open data, the national catalogue's API returns 10 agency datasets — the supplier register 2020-2026, historical procurement data, site publications and the article catalogue — and zero on prices; the agency's own site search returns 0 results for «precios de referencia». On the corpus side, the universe the cap claimed to cover: the single catalogue holds 90,368 articles, of which 80,788 are goods and 74,770 are current. The agency published a price for one product: 0.001% of the catalogue. Spending on goods between 2021 and 2025 — the universe of LUC article 318, which reached every catalogue article including direct purchases and petty cash — was 215,682,798,431 pesos across 1,001,565 award lines (2021: 38,049,215,741; 2022: 49,667,671,946; 2023: 53,571,285,958; 2024: 38,747,871,021; 2025: 35,646,753,766), and it is a floor, because it only counts lines with a catalogue code and a plausible amount. The universe of TOCAF art. 33 D) item 16, food: the corresponding sub-family holds 1,823 articles, 852 of which were actually bought between 2021 and 2025 for 17,766,912,252 pesos across 96,172 lines. And nobody invokes the cap: searching for exact phrases over the text index of the corpus's 2,184,332 records, «precio máximo de adquisición» returns 0, «precio testigo» 0 and «precios máximos» 0; «precios de referencia» returns 13, «precio de referencia» 6 and «precio máximo» 7, and we reviewed all 26 one by one — 11 are the water authority's stock phrase («participar en el Concurso de Precios de referencia»), 7 are indicative quotation requests from one hospital, 2 are 2003-2004 regularisations and the 7 «precio máximo» hits are the tender document's own caps: none invokes a price published by the agency. The only verifiable case shows both things at once. Fresh pasteurised milk for human consumption, catalogue code 2988, in pesos per litre and without a single null-currency line: 2023 n=27, range 26.72 to 180, median 44.30; 2024 n=31, range 33.90 to 125, median 45.90; 2025 n=75, range 35.25 to 90, median 48.38; 2026 n=50, range 38.52 to 180, median 44.47. The June 2024 reference price appears verbatim in the corpus: there are 12 lines at exactly $42.34, all from the national dairy producers' cooperative selling to Rocha, Florida, Bella Unión and the Florida departmental government. That is, the instrument exists, is used, and can be seen working. In the same period the same code was bought 44 times at 60 pesos a litre or more, up to 180, above even the maximum home-delivery retail price ($43.30). And here goes the warning that kills the easy headline: the finance ministry resolution caps only polyethylene-bag milk with 2.6% fat or more, and the OCDS record does not state the presentation. Therefore it CANNOT be asserted that those purchases breached a cap: what can be asserted is that nobody can know with what is published. The competing explanation is strong and goes above everything. Article 318 was never operative because it was never implemented by decree, and implementing it was the Executive's job, not the agency's: the article itself says the Executive «reglamentará la aplicación de este instrumento», and without a decree the agency had no legal instrument to publish anything, so the body's silence may be, literally, somebody else's silence. The definition was also technically fragile, and milk proves it: «the lowest purchase price in force for each catalogue article» assumes a code identifies a comparable product, and code 2988 covers regulated bagged milk, bottles, different fat contents, home delivery and plant collection, with legitimate prices from $37.26 to over $60; publishing the lowest price as a cap would have blocked perfectly correct purchases and pushed bodies into justifying almost every purchase in writing, petty cash included, so the repeal may be a sensible correction of a badly designed rule and not a step backwards. Item 16 is not obvious negligence either: that cap applied to a very narrow ground, and for the products in that basket there is no administered price the agency could copy — only milk has one, because regulated milk has been price-controlled since Ley 10.940 — and publishing prices for tomatoes or squash would have required standing up an agricultural price observatory that is not the agency's remit. And something that cuts against us: the information exists, only not as a cap; every awarded unit price is in the feed and on the procurement site, and a diligent buyer can see today what others paid for the same code — our own site is the proof — so the State could argue that price transparency is already solved by publishing the awards. What none of those explanations covers: why, if the instrument was unworkable, it was left five and a half years in the Code with an express obligation on every purchasing authority; and why the one price that was published is published inside an Excel file, twice, with a broken link and a text-less scan. Prior coverage, to be clear about what is ours: the milk price adjustment is covered as a consumer story by Montevideo Portal and El Observador, without mentioning that the price is the legal cap on State purchasing; and the repeal appears as a bullet point in two law-firm bulletins of December 2025 and January 2026, with no measurement at all. We found no Uruguayan news piece saying the agency never published those prices, nor one crossing the cap against actual spending.
- Which law it would engage
There are two legal caps tied to «precios publicados por la ARCE», both in force between July 2020 and 31 December 2025, and both repealed by the same Budget law. The first: TOCAF art. 33, letter D), item 16 — direct purchase of nationally produced food and fresh provisions offered by family producers, worker cooperatives and organisations qualified under Ley 19.292 — whose last paragraph set a price cap referred to what the agency publishes, as worded by art. 35 of Ley 19.996, and which was repealed by art. 29 of Ley 20.446 with effect from 1 January 2026. The second, the more ambitious: art. 318 of Ley 19.889 (LUC), «Precio máximo de adquisición», which defined the maximum price as the lowest purchase price in force for each article in the single catalogue of goods, required the purchasing authority to place the agency-published value in the file on every purchase — including small direct purchases and petty cash — and to justify any difference in writing before awarding; repealed by art. 24 of Ley 20.446, same effective date. Nothing we measured is anyone's breach: what the card shows is that a control written into law left no verifiable trace, and that it was repealed without a single official measurement of its application.
TOCAF art. 33 lit. D) item 16, last paragraph (wording of art. 35 of Ley 19.996; identical in art. 314 of the LUC): «En cualquier caso, los precios a pagar no podrán superar los precios publicados por la Agencia Reguladora de Compras Estatales para ese producto». Ley 19.889 (LUC), art. 318: «Se define como “precio máximo de adquisición” al menor precio de compra vigente a un momento dado, para cada artículo contenido en el catálogo único de bienes adquiridos por el Estado. En todo trámite de compra que refiera a un artículo contenido en dicho catálogo único, el ordenador respectivo deberá incorporar en las actuaciones el valor del precio máximo de adquisición publicado por la Agencia Reguladora de Compras Estatales… Quedan comprendidos por este requisito todas las operaciones de compra, aun las dispuestas como compras directas por monto menor y las que se realicen con cargo a fondos fijos… El Poder Ejecutivo, con el asesoramiento de la Agencia Reguladora de Compras Estatales y del Tribunal de Cuentas, reglamentará la aplicación de este instrumento». Ley 20.446, art. 24: «Derógase el artículo 318 de la Ley N° 19.889, de 9 de julio de 2020.»; art. 29: «Derógase el último inciso del numeral 16) del literal D) del artículo 482 de la Ley N° 15.903…»; art. 3: «Esta ley regirá a partir del 1º de enero de 2026…». Source warning: a law-firm bulletin attributes the repeal of art. 318 to art. 25; on impo it is art. 24 — art. 25 repeals the last paragraph of art. 40 of Ley 20.075 and art. 26 repeals art. 330 of the LUC. Complementary rule for the one published case: Decreto 130/008, art. 8, and finance ministry resolutions 104/023 and 162/024 issued under Ley 10.940 and arts. 31, 33 and 44 of Ley 18.242.
- What is missing before asserting it
- The series does not exist in any format: there is no dataset, no page, no API holding the list of «precios publicados por la ARCE» that two statutory provisions used as a cap, and the only price published in the body's entire history is fluid milk's, which comes inside an order spreadsheet. Article 318's «maximum acquisition price» was never published for any article: zero announcements, zero datasets, zero mentions across 2.18 million files; it is an obligation written for 90,368 articles and met for none. The implementing decree is missing: art. 318 instructed the Executive to implement it with advice from the agency and the Court of Auditors, and impo's updated page for that article records only the repeal and its effective date, with no «implemented by» note; we found no implementing decree, and if there is one, let them show it. The record is missing: even where the cap existed and was used, the OCDS file has no field saying «applicable reference price» or «product presentation», which is why a purchase of code 2988 at 90 pesos a litre and another at 42.34 are indistinguishable to anyone trying to check — it cannot be known whether the first is unregulated milk or regulated milk bought above the maximum. Accessibility of the little there is, is missing: one of the three milk-price announcements returns HTTP 403, and the March 2025 PDF is a scan with no text layer that is not linked from any section we could find. And the reckoning nobody did is missing: Parliament repealed both instruments in December 2025 without a single public report of how many times they were applied, how much they saved, or why they did not work. Three reservations of our own, carrying the same weight as the numbers. First: no purchase measured is in breach; the milk cap covers only polyethylene-bag milk with 2.6% fat or more and the OCDS record does not state the presentation, so purchases at 60-90 pesos a litre are an open question and never a violation. Second: the 17,766 million in food is the sector's upper bound and not the amount actually reached by item 16, because that ground is narrow and the feed does not allow isolating it. Third: the «zero» mentions is measured over the OCDS text fields — title, description, items, names — not over tender documents in PDF, so a tender document may invoke the cap without our seeing it.
- Who has to answer
- The state procurement agency: did it ever publish a «maximum acquisition price» under LUC art. 318? For how many articles of the single catalogue? Where? And why does the only reference price it did publish travel inside an order .xlsx rather than as open data, when the body already publishes ten datasets on the national catalogue? The agency again: between May 2023 and today about seven half-yearly milk-price adjustments fell due under Decreto 130/008; why did only two reach an announcement, and why does one of them return «access denied» today. The Ministry of Economy and Finance, which sets the milk price and signed the repeal into the Budget: what assessment of art. 318 was made before repealing it, and whether any application report exists. The Executive: whether the implementing decree art. 318 itself required was ever issued and, if not, why for five and a half years. The Court of Auditors, which art. 318 named as adviser on the implementing decree: whether it ever observed an award for lack of the maximum-price record that article required to be placed in the file. And the bodies that buy food from family producers and cooperatives under item 16 — the food institute, the education authority, the health service, the departmental governments: against which published price did they check the cap between 2021 and 2025.
Measured on 2026-08-14 · Re-measure it with: npx tsx tests/unit/hallazgo-arce-referencia.verify.ts
Tell it where you can be protected
We publish what is in the open data. What isn't — the file, the message, what was said in the meeting — goes to Uruguay Leaks, the anonymous inbox run by la diaria, PODER and DATA Uruguay. The message is written for you: copy it, add what you know and paste it into their form.
Do not send it from your work computer or work email. Uruguay Leaks protects the submission, not the machine you send it from.
Uruguay Leaks is a project by la diaria, PODER and DATA Uruguay. We are not affiliated with them and receive no copy of what you send.
Where to check
The law and the record
Every finding is re-measured with the command at its foot, and every figure links to the investigation where it is developed with its official records.
The law and who applies it
How to read this investigation
This is an analysis of public procurement data, which is public record. It documents verifiable facts — amounts, dates, suppliers, procurement method — and flags patterns worth scrutiny, keeping proven facts apart from open questions.
An exception purchase, a high price or supplier concentration is not, on its own, proof of wrongdoing. Many contracts may have valid administrative justification. The goal is to enable citizen oversight, not to issue a verdict.
Companies and people are named only as state suppliers, as they appear in the official system. Anyone named may add their response or correct a figure, and it will be incorporated.