You pick the seed. We don't. The draw is deterministic: the same seed always returns the same cases, so your result is reproducible by anyone else — and we cannot re-draw until a convenient sample comes up.
Open, no key and no sign-up. An audit that requires registration is not an audit.
184731927416531902500 ready-made seeds, if your tool cannot compose the URL
These three seeds were picked by an external reviewer, not by us — deliberately. Swap in any integer:
/auditoria/random/YOUR_SEED/25
https://dadosb3.com/auditoria/random/184731/25 https://dadosb3.com/auditoria/random?seed=184731&n=25
Same seed, same sample — an invariant enforces it. The path form exists because automated browsing tools often refuse query strings.
The draw was POSITIONAL and is now keyed by a hash of the case. A sample drawn before that date will not reproduce from its seed: different cases will come back. This is not the base being swapped underneath you nor a choice of ours — every response declares esquema_amostragem. The change was made because under the old scheme any update to the base undid the whole sample, so a published audit result stopped being checkable by anyone else.
Published value, flag, formula, the declared CVM accounts, the account that company actually used in that financial year, and the link to the CVM file so you can redo the arithmetic yourself.
A small sample does not measure an error rate. For that, run several seeds and add them up — publishing the cases that matched too. A rate without a denominator is not a rate.
audit history (including the errors found) · /auditoria/universo · /desafio (cases we chose, with the trap declared) · /transparencia · /metodologia