Data and checks

A national figure published as "0.2%" is any true value between 0.15 and 0.25. A rebuilt 0.23% agrees with it as closely as the published figure allows.

Casen publishes three expansion factors, one per level. Communes summed with the communal factor do not reach the national total, and are not meant to: 11 communes are outside the sample entirely.

SII publishes its register twice: once per commune, once per commune and branch of activity. Firms and workers agree to the unit between the two. Sales do not — between 0.5% and 5.7% of the UF a commune declares appears under no branch, every year since 2005. The gap is declared here and the branch charts rank by workers, which is the column that reconciles.

Eight of the ENE's 197 quarter labels do not follow the one before: one is a typo (a year in the eleventh century) and seven are a change of convention — since 2019 a quarter crossing into a new year is labelled with the year it starts in. Each was corrected in the year only, and only where the month it names matches the month the sequence requires. Read literally they would put sixteen regional rates on the wrong date.

The agricultural census counts a holding once in every commune it has land in, and 17% of them hold more than one plot. So its communes add up to 180,574 holdings where Chile has 175,556 — and the overcount behaves exactly as that explains: self-consumption plots, which are under two hectares and never straddle anything, are thirteen units out, while the company-held ones are 22% over. The hectares have no such problem and add up to the published national figure, so PuroChile sums hectares and never sums holdings.

SERVEL's count of election night is what each polling table wrote on its record, and it flags 2,656 tables of the first round whose parts do not add up to their total. The scrutiny boards and the Tribunal Calificador resolve exactly those, so the qualified result moves every total a little — and it is the qualified result PuroChile draws. The distance is published as a declared difference, race by race. So is one more: SERVEL's participation base counts 5,023 more voters than the Tribunal counted ballots in the first round, and neither authority publishes how the two are reconciled.

Casen 2024 expands to projections built on the 2017 census, about 9% above what the 2024 census counted. Census counts and Casen shares are never multiplied together here.

It publishes AI use by US occupation (SOC), not by economic sector. What the sector pages show is that index weighted by the occupational mix Casen observes in each branch in Chile, with both classifications collapsed to six groups coarse enough to agree. It feeds no indicator, no total is summed from it and it is reconciled against nothing: there is no published Chilean figure for AI use by sector to check it against.

Anthropic Economic Index, junio 2026 · Anthropic (CC BY) · Sector pages, "Artificial intelligence" section

No Chilean authority publishes it. All sixty files of the SII release are aggregates — tax secrecy — and the financial regulator publishes no machine-readable activity code for the issuers it registers. So the sector pages show how concentrated a sector is, from brackets of annual sales that are published, and name no firm. A list of "the main players" here would be a list PuroChile made up.

Every export and import declaration names the customs agent who filed it, and Aduanas's open databases leave that field out — as they leave out the importer and the exporter. So the trade pages show the customs office, the transport, the country and the region of origin, and no agent or firm. Nor do they say where an import is used: a declaration names the office that cleared it, and a region profile shows imports as what its offices cleared, never as what the region consumed.

ANID's register names only the principal institution of each project, not its partners, and no Chilean source publishes co-authorship. So the knowledge network joins subjects, not people or institutions, and the affinity between two universities is how alike their mix of Fondecyt study groups is — similarity of agenda, never collaboration. Nor are publications or the money awarded counted: the register carries amounts in nominal pesos of forty different years, and adding them would be a sum without a unit.

Neither source knows. SII files a job at the company's tax address and Casen finds it at the worker's home, and a mine is at neither. Rather than pick one, the sector pages publish both counts side by side and never subtract them: they are different measurements of different things, and the distance between them is largest exactly where work is run from a head office.

Because in 2026 the border between them moved. The CFT intake falls by 25,214 and the IP intake rises by 32,452, and the fall is matched programme by programme — Técnico en Mecánica Automotriz goes from 3,928 to 282 at the CFTs while going from 2,055 to 6,098 at the IPs. That is one institution absorbing another, not students changing what they study: CFT INACAP has no row at all in 2026, and of the 26,579 people who started there in 2025, three in four are enrolled at IP INACAP the following year. The register has done this once before, with CFT La Araucana absorbed by IP La Araucana in 2015. Read apart, a list of what collapsed at the CFTs would be a list of paperwork; added together the transfer cancels. Postítulo, the fourth level the register holds and larger than postgrado, is left out of the tracks entirely: it is classified into ten broad fields rather than programmes, which is not the same object as the other lists.

Not their home address — Mineduc publishes no residence in either module. What it does publish is where they were enrolled in secondary school, and that is the approximation PuroChile uses: four years of an education happen in that comuna. It is labelled as an approximation wherever it appears, and it is not a claim about catchment — a commune whose schools draw pupils from next door counts them here. The two modules are joined on the MRUN Mineduc publishes in both, a link the ministry made rather than one PuroChile inferred from names.

The profile of each higher-education institution follows its first-year entrants by the masked id Mineduc publishes in every register for that purpose: back to the school they left (the comuna of the school, not of the home), forward into the next year's enrolment (still there, at another institution, or nowhere: SIES's first-year retention), into the register of titles (who graduated, where, and how long it took against the formal length), and, for postgraduates, back to their undergraduate title. Each register starts in a given year — school leavers in 2006, titles in 2007 — so the people not found are published as such: not finding a title is not the same as not having one. Admission preferences come from DEMRE and cover only the universities in the centralised system. Enrolment counts enrolments, not people, as SIES does; every share and every mean on those pages is taken after adding counts, never by averaging rates.

Where it is taught is the last year's first-year entrants by institution and campus, with the comuna of the campus — not of the students — the timetables, the formal length and the list fee, a plain mean over the entrants with a fee in pesos (fees in UF are left out and counted). A place with fewer than three entrants is listed without its count. Retention follows each cohort into the next year's enrolment four ways: still in the programme at the same institution, at the same institution in another one, at another institution, or nowhere. Graduation follows the same cohorts into the register of titles: a title in the same generic programme, anywhere, or failing that a title in something else; a recent cohort has a short curve, not a low one. Employment and income are not in the registers PuroChile reads: they are what SIES publishes on mifuturo.cl, crossed with tax returns, per institution and generic programme, and they are shown as published — one row per institution, never averaged, the income as the band SIES gives and not as a number. To check that PuroChile reads the register as SIES does, its own first-year retention is computed with SIES's definition for every pair mifuturo publishes and the two are compared; how many land within three points is in the table below, under its own source. The trend of a programme compares three-year averages, never two single years, and is drawn beside its whole series. Each programme also has a page at each institution where it reached ten entrants in some year and is taught today or published on mifuturo: the same counts read with the institution as part of the key — its entrants, its own cohorts into the next year and into the register of titles, the real length of the titles it gave — next to the programme across Chile, counted the same way. The institutions of a programme add up to the programme in entrants, in every cohort's four outcomes, in every cohort's titles and in every title's length, and each of those sums is a check in the table below. What is followed person by person belongs to the institution as a whole and is not split by campus. Institutions' logos come from Wikimedia Commons, only under licences that allow anyone to reuse them, are served from this site, and each carries its credit where it appears.

First-year dropout is the last of those four outcomes read over a window, the last three cohorts followed (2023–2025): everyone enrolled at no institution at all the year after they started, over everyone who started — the three cohorts added together, never an average of three yearly rates. A window under 30 entrants carries no rate, and a cohort withheld because its institution under-reported the following year is out of it and named on the page. Beside it, from the same sums, the share still in the programme and the share that changed programme or institution, which is not dropping out. Some of those enrolled nowhere come back a year or two later, which one year's register cannot see: it is dropout after the first year, not final dropout, and every page says so. The institutions of a programme add up to the programme in that window, entrants and each outcome; and each institution's whole undergraduate intake, added up programme by programme, matches the retention its own profile follows by institution code — the same register counted twice, apart. Both are checks in the table below.

The 2021 agricultural census maps 313 units for 346 communes, and sixteen of those units are two or three communes published together. Splitting them back by any rule would be inventing the figure, so nothing is split: those communes carry no land figure of their own. Every pooled unit sits inside one province, so the province, the region and the country are complete — 8.8 of the 45.7 million hectares reach the site only above the commune.

Iquique and Alto Hospicio · Calama and Ollagüe · Mejillones and Sierra Gorda · Tocopilla and María Elena · Copiapó and Caldera · Chañaral and Diego de Almagro · Valparaíso, Viña del Mar and Concón · El Quisco and El Tabo · Chiguayante and Hualqui · Coronel and Lota · Talcahuano and Hualpén · Cisnes and Guaitecas · Tortel and O'Higgins · A group of Santiago comunas (1) · A group of Santiago comunas (2) · A group of Santiago comunas (3)

The cancer split by site of the body is the finest figure PuroChile publishes — fifteen ICD-10 sites in every commune, where the median commune has about five hospitalisations in each — and the only one with cell-level disclosure control. A site with fewer than 5 hospitalisations in a territory is not published on its own: it is pooled with at least one other into "reserved sites", so the tiles still sum to the true total and what is unknown is how the pool divides. The reserve propagates through the commune-province-region tree in both directions, because a cell hidden in one commune alone would be its region's figure less its neighbours'. The counts stay complete in the data; what is pooled is what is drawn.