English: the app (server/, analysis.py, routes and JSON fields), instructions/analyst.md, docs; the table of shared renames

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
maximeandClaude Opus 5.5 committed 2026-11-16 12:40:03 +01:00
1 parent 1d80a77cb1
commit a3e6ade845
24 files changed
+1959 -1835

No files matched your search

+27
View File
@@ -0,0 +1,27 @@
# Diagnostic-App's instructions to the analyst
What the app asks of the analyst (agents repository, `agents:analyst`), on top of its trade. Launched by the Analysis tab of the Data page (`app/server/analysis.py`), with `DETECTORS: no`.
## Why this range
The user selected a range in the Data page and clicked "Analyze the range": they saw an anomaly there that the detectors catch badly, or they want a deeper analysis. Their hints are in the request (or "none"). Your labels are then used as the truth to improve the detectors ("Optimize the detectors"): they must be right.
## Your output folder
`labeling/auto/analyses/<STATION>_<start>_<end>/`, given by the request; your work and claude's log in `work/`, outside git.
- Your labels go in `labels.jsonl`; the server is the one that submits them to `labeling/auto/<STATION>.jsonl`. **You never write in `labeling/auto/<STATION>.jsonl`.**
- Then the user may write to you (the same session resumes): a question, a correction. An indication from them is written in the labels it touches ("according to the expert: …", in words) and in your analysis, section 3. If you change your labels, rewrite `labels.jsonl` in full: the server submits them again; if the diagnosis changes, update `report.md`.
## The report
On top of your labels and your analysis: `report.md`, the diagnosis the user will read in the page, in English:
- `# <STATION>: <the subject in a few words>`, then the question (their selection, their hints) in one or two sentences;
- "What the data show": each event dated (UTC), the sensors affected and those that do not move, the cause and its level of evidence (confirmed / strongly supported / consistent with / possible / insufficient evidence / not observed), the command behind each fact;
- the figures that prove, each inserted by `![caption](path relative to the output folder)` and followed by a line that says what is seen in it: curves (`plot.py`), thumbnails, whole camera images (copy them into the output folder with `cp`); a few, chosen, not all the working ones;
- "What remains not established"; "Labels": those you propose, one line each.
## Your answer
In 2 lines: the number of labels, the path of the report.