Plain-English descriptions

Every record, for AI decisions and for audit evidence, comes with a sentence anyone can read. Nothing needs to be turned on: the sentence is built from the record's codes and counts and is included automatically when the ledger is exported.

How it works

The ledger stores no free text, so the sentence is generated from the closed codes and the counts with fixed templates. That has three consequences:

Examples

RecordSentence
Silver run, 7000 / 6210 / 790, passedSilver layer, nightly run: 7,000 rows came in, 6,210 were valid and 790 were quarantined. Valid plus quarantined equals the input, so the integrity check passed. Supports HIPAA integrity (164.312(c)).
Silver run that does not add upSilver layer, nightly run: ... Valid plus quarantined do not add up to the input, so the integrity check failed.
CI tests, 53 / 53 / 053 automated tests ran: 53 passed and 0 failed. Supports HITRUST secure development and maintenance.
Row-level security test, 25 / 25 / 0Row-level security test: 25 probes were run as the wrong user; 25 correctly returned nothing and 0 leaked. Supports HIPAA access control (164.312(a)).
SQL guard allowed a queryThe assistant turned a question into a read-only database query. The SQL guard allowed it.
Answer with no cited sourceThe assistant answered a question without citing a source, so the answer was held for review.

Where you see them

Adding a new kind of record

A new label or control is added to the closed vocabulary together with its template and a test that checks the sentence. Until a template exists, an event is described generically ("An event was recorded.") rather than guessed.

The sentences describe what was recorded. They are not a conclusion about compliance.