Otto: more reliable answers, safer tools, correct scheduling times

2026.10.18-otto-fiabilite-outils · 2026-10-18

Delivered build : cdab7a5ef6c8ab951219d1958ce527a9f9926a82

Changes

Otto no longer gets stuck on a mistyped value: a malformed reference or date is named and corrected within the same exchange, instead of being reported as a database failure and retried unchanged.
When a single piece of information is not readable by the assistant, it says so for that item and carries on reading, instead of giving up on the whole record and asking the person for references.
A rejected write is always reported as rejected: Otto no longer claims something was saved when nothing was written, and it does not invent the reason for the refusal.
Otto answers entirely in the language of the request, without switching language mid-answer.
Otto's answers are written in laboratory terms: no field names, internal references or stored values, and no explanations of its own inner workings.
A study prefilled from an order stays at quote stage by default, instead of being set to a progress stage nobody asked for.
The time of a scheduled session is given simply in the person's own time: the conversion, daylight saving included, is now done by the server rather than by the assistant.
Document import: batch extraction for long protocols, reading plan and extraction checks.
Charts: axis bounds and scale breaks.

Scheduling a session through Otto is still unstable on some requests: when it fails, Otto may still give a plausible but inaccurate reason.
A technical status value still shows up in roughly one answer in ten.