Dados B3 › Transparency
Transparency
Transparency — how you check the numbers yourself
Data version 2026-11-09-ancora-do-fre · page generated 2026-11-22 13:05 UTC. If these do not match /saude, you are reading a cached copy.
What is checked, before every publication — 1448 automated tests, collected live
83 are DATA invariants (does the number tie out?) and 148 are product tests — access gate, navigation, language, AI connector. Calling all 1448 'invariants' would be inflating it: the split is checkable in the list itself. Not a hand-picked highlight reel: every test below was found by introspecting the actual test files just now, the same way pytest --collect-only would. Grouped by what it protects against, not by which file it lives in.
Level 1 — Integrity: does the data tie out? (75)
CT201— The e-mail ceiling is a single one: the newsletter spends part of the day and the weekly watch e-mail sends only what is left, and with the day full the magazine invitation and the Sunday Desk are barred and nothing goes out.CT202— The reverse case: the weekly watch e-mail uses the whole day and the newsletter stays pending without being marked as done, and on the next day the ceiling resets and the queue goes out.CT203— The monthly e-mail ceiling counts every routine together and resets at the turn of the month: with two e-mails already spent in November and a monthly ceiling of three, only one goes out on Monday the 30th and the rest leaves on the 1st of December.CT204— The e-mail ledger day is the Sao Paulo day: 11:30 pm on November 30 and 12:30 am on December 1 in Sao Paulo fall on different days and months, and what was counted before midnight does not count toward December.CT205— The e-mail ledger counts every message the provider accepted, from any routine and including transactional ones, and never counts a refused one, a network failure or a message kept in the file because no provider is set; the recipient never reaches the log, and transactional e-mails make the weekly watch step back on the same day.CT206— When the e-mail counter cannot be read the ceiling closes: the weekly watch e-mail sends nothing and marks nothing as done, the next round recovers, unreadable settings and a closed connection also close it, and a failure to write the ledger does not undo an e-mail that already went out.CT207— The owner campaigns and the checkout reminder stop when the day is full and resume when there is room: with one e-mail left, each sends one and leaves the rest for the next run without registering it.CT208— Every batch e-mail routine asks the single e-mail ceiling before sending and none redoes the ceiling arithmetic on its own, and the ledger is written at the only exit point to the provider.CT209— The e-mail ceiling settings have a daily and a monthly ceiling that both fit inside the free provider quota, leaving room for transactional e-mails that are never barred.CT210— A real batch that holds an open transaction on the follow-up database while sending records every send in the cash ledger without waiting or losing any: three accepted, three counted, and the day total matches what the provider accepted.CT211— The e-mail usage of the day and of the month comes from one aggregated query per table, at most 14 SELECTs in total against about 250 before, and gives the same total as adding day by day, with skipped states left out and the days of another month left out.CT212— The daily backup copies the e-mail cash ledger together with the other small databases on the disk, through the SQLite backup API, and the copy holds the ledger content.CT01— The My portfolio configuration holds the numbers the owner decided: free 5 tickers and 1 rule, Pro 200 tickers and 50 rules, 30-day session, 30-minute link, and the weekly and daily e-mails switched OFF. An unknown plan gets the free limits.CT02— The 7 new tables live in the follow-up file, never in the public database; creating them again changes nothing; the e-mail exists only in the account and pending-login tables, and the news table has no column about a person.CT03— The My portfolio reader understands one ticker per line with quantity and average price in Brazilian format: WEGE3 30 25,40 is 30 shares at 25.40, with thousands dot, R$, Excel semicolons and the odd-lot code.CT04— A line with a malformed code, unreadable number, zero, negative or huge quantity or price, or an extra field is refused whole, cut to 40 characters; blank lines and comments are ignored; above the line cap the rest is refused with a notice.CT05— The same ticker on two lines adds the quantities and the average price becomes the weighted average; if one line has no price the average price is left empty; a code-only line does not erase what came before.CT06— Free keeps up to 5 tickers and Pro up to 200: a new ticker beyond the limit is refused and reported, and a ticker the account already has can still be updated.CT07— Someone with 8 tickers and 3 rules on Pro who went back to free sees 5 tickers and 1 rule in force and the rest paused; nothing is deleted, new tickers or rules are refused, and renewing brings everything back.CT08— Free keeps 1 alert rule and Pro 50; a rule with an indicator outside the list, an unknown type, a threshold without direction or number, an infinite number or a thesis with a value is refused as invalid.CT09— Deleting the account removes every row of that person from the 6 tables with personal data and leaves the other account and the news table, which belongs to no one, untouched.CT10— The delete-my-data flow (single-use link plus button) also deletes the My portfolio account with everything in it, through the single delete function; before, the site knows it holds data for the e-mail; after, it does not; the other account stays.CT11— Cleanup deletes expired sessions, login requests older than 2 days, sends older than 12 months and accounts cancelled more than 12 months ago in full; an old active account, a recently cancelled one and a valid session stay; the site's daily retention calls this cleanup.CT12— The admin count has numbers only (accounts, accounts with tickers, tickers, rules, weekly on, valid sessions, login requests), and no account operation writes the e-mail to the log.CT13— An e-mail with capitals and spaces is the same account as the lower-case e-mail; creating again returns the same account and clears the cancellation; an invalid e-mail creates no account.CT14— Two accounts with the same ticker each keep their own quantity and average price; changing or removing in one does not touch the other, and the rule limit is counted per account.CT15— In a follow-up file without the new tables, delete and cleanup create nothing and do not fail, and the delete-on-request flow still removes the old requests of the e-mail as before.CT16— The configuration and texts of the My portfolio module use no word that sounds like investment advice (the site's forbidden list), and the stored price is called average price.CT17— An account with no access for more than 12 months (the same yardstick as the 12-month retention) is deleted in full, leaving no row of that person in any table; with access one day before the limit, or exactly at the limit, it stays.CT18— A newly created account is born with a last access, and an old account with no access record counts from its creation; logging in again or validating the session renews the access, written at most once a day.CT19— Someone with an active Pro plan (billing in progress) is not deleted for inactivity: the cleanup takes the plano_pro_ativo question and protects whoever it says is Pro; without the question, an account linked to Stripe or to a Pro key stays.CT20— Two accounts with the same ticker, one idle for more than 12 months and one in use: the cleanup deletes everything of the idle one and leaves the other exactly as it was, row by row.CT21— An account with an empty last access uses its creation date; with empty or unreadable access and creation, or an empty cancellation, the cleanup deletes nothing (dirty data never causes a deletion), and a truly expired account still goes.CT22— If checking the plan of one account fails, it stays (when in doubt, no deletion); if deleting one account fails midway, only that one is rolled back whole; the other expired accounts and expired sessions go just the same, and the log carries no e-mail.CT23— The news table has exactly the columns of the alert engine's register, and the rules table has every rule field the engine reads and all the memory it returns; rules read back pass the engine's own validation.CT24— End to end, offline: the news register and the rules' memory go to the database and come back each round; the first load is silent, the week's new filing alerts once, the threshold alerts on crossing and stays silent afterwards, the thesis keeps the periods seen, and one account's memory does not leak into the other.CT301— The whole sign-in path works: the form asks for the e-mail, the request sends ONE confidential e-mail with the link, opening the link shows the button, the button creates the account and the session and leads to the account page, which shows who signed in and records the access.CT302— The link works once: the second click on the button gives the link-no-longer-valid page (410), with no cookie and no new session, and opening the link again does too.CT303— An expired, used, deleted (it-was-not-me) or made-up link gives the SAME page (text and 410 status), when opening and when confirming, and nothing signs in.CT304— The link requested for one e-mail only signs in to that e-mail's account, even when the other account exists, and each cookie shows only its own account.CT305— Opening the link (GET), any number of times, neither spends it nor sets a cookie, and opening the it-was-not-me link (GET) deletes nothing: only the button (POST) acts.CT306— Every POST (ask for the link, confirm, it was not me, sign out) without an origin, with another site's origin or with a null origin is refused (403) and does nothing; from the site itself (Origin, Sec-Fetch-Site or Referer) it goes through.CT307— Asking for the link for an e-mail that has an account and for one that does not gives the SAME answer (status, text and headers), and nothing about the account is looked up before it: the work runs after the answer.CT308— The login cookie is born only on the confirm POST, with HttpOnly, Secure, SameSite=Lax, Path=/ and the session's validity; the rest of the flow sets none, and signing out only deletes it (zero validity, empty value).CT309— Home, a stock page, the API page, /carteira, the privacy policy, delete my data, the sitemap and the data API answer without a cookie, with or without a login cookie in the request.CT310— The link e-mail and every sign-in and sign-out screen, in Portuguese and English, avoid the forbidden words of My portfolio and show no script.CT311— The it-was-not-me button deletes that e-mail's pending link requests, and only them: the account, the portfolio and the open session stay, and the deleted link no longer signs in.CT312— An expired session, one of a cancelled account or one ended by signing out does not open the account page (which only shows the way to sign in), and signing out deletes the session from the database.CT313— The database keeps only the SHA-256 of the link and of the cookie: neither appears in clear in the file; and without an e-mail provider the pending queue gets the notice without the text (the link is not stored on disk).CT314— Along the whole path (ask, open, confirm, account page, it was not me, sign out, invalid link, per-address limit), no server log record carries the person's e-mail, the link or the cookie.CT315— Every sign-in, sign-out and account page goes out with Cache-Control private, no-store and noindex in the header and the HTML, with no canonical or hreflang; the prefixes are in the API's no-public-cache and no-search-signal lists, and none of their addresses is in the sitemap.CT316— Link requests are throttled: per address (the same resend rule as the key and the portal, on its own counter) the excess sends nothing and the screen does not change; per connection per day the excess gets 429; a robot gets the sent screen and nothing goes out; an invalid e-mail gets 400 and nothing goes out.CT317— The privacy policy, in Portuguese and English, says only people who sign in get the login cookie, what it is for, how long it lasts (the configured value), how to sign out and that nothing else sets a cookie, without loosening the rest (still no tracker, no ads); delete my data mentions the account and the result names each of its tables.CT318— The per-address limit of the sign-in link uses the same window and cap as the key and the portal, but on its own counter: using up one does not use up the other, and key, portal and delete still add up together.CT319— The link is spent only after the account is created or reactivated: if that fails (nothing returned or an error), the person sees try again, with no cookie, and the same link still signs in later; and a cancelled account is reactivated on sign-in.CT320— The final word on the link and on the session belongs to compare_digest from the hmac module (constant-time comparison) between the stored and the computed hash: if it says no, neither the link nor the cookie is valid.CT321— Signing in again on a device ends only the session that device carried; another device's session goes on, and the privacy policy says so in Portuguese and English.CT411— Whoever signed in saves tickers through the text box (code, with optional quantity and average price) and the page shows each one with quantity, average price, last close, value, weight and the change over the average price, plus the calculation warning for a code the database does not recognise.CT412— Two accounts never see each other: the page, the downloaded CSV, editing, removing, uploading a CSV, clearing the average price and deleting everything in one account only touch that account's portfolio; a ticker that only the other account holds gets exactly the same response (status and text) as a ticker that exists nowhere.CT413— The free plan keeps 5 tickers and Pro keeps 200: a new ticker beyond the limit stays out and the screen says which ones; a ticker already saved can still be updated; an account that had more tickers (expired Pro) sees the ones above the limit as paused, left out of the numbers, with nothing deleted.CT414— Editing saves exactly the quantity and the average price (an empty field clears the number); an unreadable number, zero or an ambiguous price like "25.400" gives 400 and nothing changes; removing takes the ticker out; clearing the average price and deleting everything only act with the confirmation box ticked, and the account stays.CT415— Pasting the /carteira address (full, with domain, or just the path) brings in the tickers with their quantity; a ticker already saved gets its quantity replaced and keeps the saved average price; an address of another page or with no ticker gives 400 and nothing comes in.CT416— The downloaded CSV (semicolons, decimal comma, header) uploaded into another account recreates the same portfolio; the reader skips the header and the Excel byte-order mark, and refuses a line with a price and no quantity and an empty file or one larger than 64 KB.CT417— Every write route of My portfolio refuses a request without the site's own origin (403) and a request without a session (401), changing nothing; every response goes out with Cache-Control private, no-store and noindex, with no new cookie, no script and none of the forbidden words, in Portuguese and in English; the e-mail does not go to the log.CT418— The BDR (AAPL34) enters the portfolio calculation with the last close, value, weight, the group by programme type, the link to the /bdrs page, the 12-month yield and the dividend to receive; a code that is not a REIT, ETF, stock or BDR still stays out with the warning; the public /carteira still computes up to 30 positions and My portfolio passes the plan limit.CT419— Every POST to /minha-carteira with a body above the cap (CSV: 96 KB; others: 48 KB), with an invalid length or sent in chunks without a length is refused before the server reads the body, and nothing changes in the portfolio; a normal request goes through.CT401— On the own domain, behind Cloudflare, the visitor IP still comes from the header Cloudflare rewrites, in the usual order: CF-Connecting-IP, then True-Client-IP, then the first X-Forwarded-For item, then the socket.CT402— On the old host the CF-Connecting-IP, True-Client-IP and the front of X-Forwarded-For count for nothing: changing the value on every request always gives the same IP and the same per-connection hash, the one from the last X-Forwarded-For item that the Render proxy appended.CT403— On the old host, with no X-Forwarded-For, the IP is the socket one, even with CF-Connecting-IP and True-Client-IP in the request; with no socket, empty.CT404— The old host is recognised exactly the way the redirect recognises it, without port, trimmed and lower-cased: upper case or a port in the Host does not bring back trust in forged headers.CT405— The per-connection limit of the sign-in page is not bypassed on the old host by changing CF-Connecting-IP, or the front of X-Forwarded-For, on every request: the request after the ceiling gets 429 and the POST there is not redirected; on the own domain the same header still counts as Cloudflare delivers it.CT406— The per-connection limit of the open showcase is not bypassed on the old host either: with the limit on, changing CF-Connecting-IP on every request does not renew the quota, which is counted by the last X-Forwarded-For item.CT450— The IP diagnostic exists only on the old host and only when the request carries X-Diag-IP: without the header, or on another host, the function returns nothing and the request goes on as always.CT451— On the old host, with X-Diag-IP, the diagnostic shows CF-Connecting-IP, True-Client-IP, every X-Forwarded-For item, the socket and the IP the limit used, and lets through only address characters.CT452— End to end, /saude on the old host with X-Diag-IP returns the X-Diag-IP header and Cache-Control no-store; without the request, or on the own domain, the response does not carry the header.
Level 2 — Semantics: does the account mean what the calculation assumes? (8)
CT601— A threshold rule fires when the number CROSSES the threshold (it was on one side at the previous check and is on the other now); a number already past the threshold when the rule was created does not fire, and staying past it on the following days does not fire either. Works for "above" and "below".CT602— After one notice, the same rule stays silent for 7 days even if the number goes back and crosses again; from day 7 on, a new crossing fires again. Another rule is not affected by the first one's silence.CT603— A missing number (None, a stock without dividend yield, a bank without net debt/EBITDA, a value with a caveat) does not fire and does not erase the remembered side: when the number comes back past the threshold it counts as a crossing only if it was on the near side before.CT604— A REIT accepts only 12-month yield and P/B, and the REIT P/B is the one on the fund card (monthly report), not price over NAV; a stock accepts the eight indicators; a REIT rule reads only REIT values and a stock rule only stock values, even with the same ticker; an invalid rule is refused without bringing down the others.CT605— The first load of each family in the news register (statements, dividends, material facts, restatements) marks everything as old and fires nothing; after that only what is new fires, once; a family that appears later also loads silently; news dated more than 14 days ago loads silently; the same statements filed again with another date become a re-filing.CT606— The default notice (new statements, re-filing, restatement, dividend announced and material fact) goes only to the tickers in the person's portfolio; the text carries no forbidden term nor the subject written by the company (the link to the original goes instead); the news register row has no e-mail, account or portfolio; the module reads no database, sends nothing and has no route.CT607— The thesis notice compares each marked indicator with the immediately previous period and fires when the watcher band changes (margin, debt, ROIC by the verdict bands; ROE by its sign; REIT by the REIT watcher cuts); the first time the rule sees the indicator it only takes note; the same period never fires twice; a number with a caveat, an indicator that does not apply to banks and a period with a gap do not fire.CT608— For stocks, a share-count caveat mark in the nightly snapshot (acoes_salto_sem_evento, market_cap_de_unit_nao_ajustado) silences 12-month P/E, P/B and EV/EBITDA and no threshold notice goes out; the dividend yield, which is dividends per share over price, stays; the site note marks and the Reference Form count without treasury shares pass; a mark the site does not know silences all four numbers; with no mark the notice goes out.
Level 3 — Economic: is the result possible in the real world? (0)
Other product tests (0)
Access gate, navigation and the MCP connector — they test whether the product works, not whether a number is right, so they don't fit the 3 levels above by design. Still counted, still listed:
Reconstruct it yourself — net margin across 5 companies, 4 sectors
We don't ask you to trust the numbers. Here's the net margin of five companies in four different sectors, each rebuilt straight from its annual report (DFP) at the CVM — net income ÷ revenue, matching the published figure exactly. For ANY indicator, of any company, the whole chain down to the line in CVM's file is at /linhagem:
| Company · sector | Net income (CVM acct) | Revenue (CVM acct) | Margin =÷ |
|---|---|---|---|
| WEGE3 · industrial | R$ 6.78 bn DRE:3.11 | R$ 40.80 bn DRE:3.01 | 16.6% |
| VALE3 · mining | R$ 11.81 bn DRE:3.11 | R$ 213.59 bn DRE:3.01 | 5.5% |
| PETR4 · oil & gas | R$ 110.61 bn DRE:3.11 | R$ 497.55 bn DRE:3.01 | 22.2% |
| ITUB4 · bank | R$ 45.85 bn DRE:3.09 | R$ 387.12 bn DRE:3.01 | 11.8% |
| BBAS3 · bank | R$ 16.78 bn DRE:3.11 | R$ 319.46 bn DRE:3.01 | 5.3% |
Download any of these DFPs from the CVM, take the income and revenue accounts, divide — you get the same number. Banks use interest income as revenue (what makes sense for a bank), so the reconciliation is sector-aware; the others use sales revenue. Figures are the latest fiscal year.
Why we may differ from another site (and it's not an error)
A difference between two sites usually isn't one being wrong — it's a method choice. We disclose ours:
- Average vs ending capital: ROE and ROIC use average equity/capital (this year + last ÷ 2), not the ending balance — so they're not a naive single-year division.
- Controlling vs consolidated: we state which one each item uses.
- TTM vs annual: quarterly multiples use trailing-twelve-months earnings.
- IFRS 16, goodwill, cash, exceptional tax: each handled explicitly and flagged when it distorts.
Do the tests bite? Yes — real cases the suite has caught
A test that never fails could mean perfect data — or a weak test. These were born from real errors that slipped through, and now fail — only shown for tests that actually exist in the collected suite above:
- I-B16 · banks' P/E = 0: the CVM reports share count sometimes in units, sometimes in thousands (varying by company and year); market cap came out 1000× too small and P/E was zero. I-B16 (price × shares recomputed) caught it — we fixed 661 annual and 2,027 quarterly multiples.
- I-B24 · the vanishing profit: the profit/equity account varies by bank (Itaú 3.09, BB 3.11); if the label search fails, the number vanishes silently and ROE is born wrong. I-B24 makes that hole fail.
- H-B23 · the 800% dividend yield: old B3 dividends carry a price not adjusted for splits; without H-B23 the yield would look absurdly real. The flag keeps the record and warns.
- I-B20 · the missing COTAHIST 2023: on the first deploy one year of prices failed to download silently; the API shipped with ~50k fewer prices and the suite passed, because nothing checked coverage. We added I-B20 — a year without prices now fails the build loudly.
- I-B33 · Oil & Gas at −89.6% in one week: COTAHIST isn't split-adjusted; a 10:1 split read as a −90% weekly return and stayed in the cumulative return forever, with nothing failing (a blind amplitude filter didn't fix it either — it just flipped which direction was wrong). I-B33 (no sector week beyond ±35%, the ruler is the exchange itself) stops the build.
- I-B31 · 364 of 865 sector-index weeks frozen: extending the series back to 2010, a sector with no company carrying a market cap that week turned into a factor of 1.0 — a flat line that the cumulative return read as 'market stood still', and the 'since 2010' return came out fictional (+527%). I-B31 fails on any week repeated to the cent.
- I-E05 · EMBRAER's 2010 P/E priced with 2025 data: the point-in-time price lookup fell forward with no ceiling whenever the current ticker had no price at balance-sheet time (ticker change, share-class migration); the multiple came out priced 15 years into the future, in 22% of annual and quarterly rows, silently. The 45-day window (and I-E05) blocks it: no session inside the deadline, no multiple computed.
- I-E04 · AZUL at R$ 1.67 quadrillion market cap: the balance sheet carried 54.7 trillion shares from the judicial recovery issuance, but the price used was already on the post-reverse-split (1:150,000) basis — two ends measured on different bases, market cap wrong by orders of magnitude. I-E04 (market cap never above Brazil's GDP) fails the build on that absurdity.
How far back each block of data goes
No asterisks here: this is coverage year by year, counted right now. Indicators (ROE, ROIC, margins, growth) come from the filings alone and cover the whole series. Multiples (P/E, P/BV, EV/EBITDA) need a price and a share count — and the share count comes from CVM's reference form, which the further back you go the fewer companies filed in a usable format. The gap between the two columns is a source limit, not an ingestion hole. If you look for an old P/E and do not find it, it is because nobody has it — not because we hid it. The same numbers as JSON: /cobertura.
| Year | Companies with indicators | Companies with multiples |
|---|---|---|
| 2025 | 441 | 290 |
| 2024 | 443 | 296 |
| 2023 | 441 | 298 |
| 2022 | 428 | 295 |
| 2021 | 422 | 296 |
| 2020 | 410 | 274 |
| 2019 | 371 | 166 |
| 2018 | 326 | 167 |
| 2017 | 317 | 165 |
| 2016 | 311 | 154 |
| 2015 | 303 | 144 |
| 2014 | 301 | 140 |
| 2013 | 292 | 131 |
| 2012 | 291 | 112 |
| 2011 | 284 | 98 |
| 2010 | 277 | 34 |
Live coverage
457 companies · 886,277 price points · 8,650 recorded restatements · last refresh 2026-11-20 01:39:51. Full live counts at /saude.
Sources: CVM (open data, ODbL) and B3 (COTAHIST). Not affiliated with B3 or the CVM. Not investment advice.
Numbers on this page are live. Data version 2026-11-09-ancora-do-fre · page generated 2026-11-22 13:05 UTC. If this does not match /saude, you are reading a cached copy.