Audit trail: a protocol edit now records a single line; Otto confirms in the screen language and writes its date bounds correctly

2026.11.08-audit-protocoles-une-seule-ligne-et-otto-anglais-dates · 2026-11-08

Delivered build : af0dc2771b148597077d6036b5393fd373ef884f

Changes

• Audit trail: editing a protocol now records a single line in the audit trail. Until now it recorded two, identical and with the same timestamp: the second came from an internal technical write, performed just after the edit to release the editor working fields. The hash chaining was intact, and remains so: nothing had been altered or removed, an addition was simply made twice.
• Duplicate lines already recorded are kept as they are: not deleted, not altered, not annotated. Removing an entry from a hash-chained trail would require recomputing the chain, that is, rewriting the audit trail. A dated compliance note documents the period, the extent and the cause, and states that a pair of identical lines reads as one single edit.
• Otto: confirmation requests before a write (session scheduling, studies, clients, protocol copy, publication, assignment, stock movement, bulk edits of samples or reagents, analyses) are now worded in the screen language. They were always in French.
• Otto: for a question about a period ("today", "this week"), the read query uses an explicit interval. When the database rejects badly written date arithmetic, Otto receives the exact form to use and corrects on the next turn, instead of repeating the same query.
• Registers: a register write performed by Otto on behalf of a user is recorded as an agent act, with its turn identifier, on the generic register route as was already the case for figures.
• Library: the "Try it" preview of the “Scientific diagrams” card opens again; it was showing an error page.

Known limitations

• Duplicate lines recorded before deployment remain in the audit trail and stay visible in exports and reports.
• An export or indicator that counts PROTOCOL_UPDATED action lines over a period before deployment therefore keeps over-counting the affected edits.
• A write on a protocol whose only difference lies in the editor working fields and the last-write timestamp no longer leaves a trace: it does not describe the protocol. Any write touching another column stays audited as before.
• The date-writing hint given to Otto only applies to date or time errors reported by the database; a query the database accepts but which is wrong in substance is not detected.