Files
aimsight-labeling/auto/analyses/long-labels/NORDEND/analysis.md
T
maximeandClaude Opus 5.5 ddb17c2189 auto: labels requalified under the 5-noise rules (only what stands out from the noise): 32 analyses refined, the 6 paused periods verified, the long labels requalified with the fixes the period analysts found; 1153 -> 961 labels
- 32 analyses: each label kept, rewritten or removed by its analyst (section Requalification of the report; mm magnitudes where the targets have them)
- SNCF01 x3, WILDSTRUBEL 2026-09-01 -> 2026-10-06 x3: verified from their existing labels under the new rules
- long labels (no analysis): auto/analyses/long-labels/<STATION>/; NORDEND-169 and -195 run to 07/11, so NORDEND-283 and -284 (their continuation) are removed
- reading-rules.md, stations.md: the citations of removed or rewritten labels point to their successor, or to [expert]

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-17 23:58:00 +01:00

11 lines
4.9 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Requalification of the long labels: NORDEND
- **NORDEND-24** (Target issue, PR04, 17/08 05:15 → 01/10 00:00): **kept** as is. PR04 is rejected at 1866 of 4229 measurements (44 %), on 40 of 46 days, its own rejections (85–100 % daily from 18 to 28/09, 54–92 % on 21–23/08 and 06–08/09). Between the rejections its own curve leaves its level beyond the threshold, without its neighbours: `ranges.py --k 5` on the label, corr_x up to +8.8 noises (28/08 21:00, +0.11 px, about +5 mm), corr_y up to +7.5 noises (24–25/08); its own noise 0.013–0.014 px against 0.005 for PR03 and PR05.
- **NORDEND-152** (Target issue, PR04, 01/10 00:00 → 26/10 20:45): **kept** as is. 435 of 2467 measurements rejected (18 %), on 23 of 26 days, by day and by night; PR05 over the same range: 42 (2 %), on 4 days (the shared episodes 09/10 and 23–25/10). Its own noise 0.014 px against 0.005 for its neighbours ("twice as noisy" holds, understated). Its own curve beyond the threshold without its neighbours: corr_x −8.6 noises (25/10 21:45 → 26/10 09:45, about −5.5 mm), corr_y −9.9 and −7.1 noises (25/10 21:45 → 26/10 06:00, about −6.6 mm), the "last departure" of the text; the −35 noises of 25/10 01:30 are shared with PR03 and PR05 (rain, other label). Rejections stop on 26/10, the end on the curves holds.
- **NORDEND-169** (Obstruction, PR17, 08/10 02:15 → 01/11 00:00): **corrected** (end → 07/11 14:00, description), review removed. Beyond the threshold all along: `ranges.py --k 5 --outside 2026-10-08T02:15Z/2026-11-07T14:00Z` (level = the September plateau, 20/09–08/10), corr_y −28 to −44 noises (−0.17 to −0.27 px, −10 to −15 mm) almost without a break from 08/10 02:30 to 07/11 14:00; corr_x +20 to +23 noises (about +6.5 mm) from 08/10 to 13/10 00:15, then +7 to +27 noises in steps to the end. PR16 does not follow (its ranges are the daily cycle, a few noises). Rejections 37 of 2909 (1 %): the label stands on the movement, not on missing points. Corrections applied: (1) the end, 01/11 00:00, was a batch edge; the offset and the disc persist to the end of the analysed data, 07/11 14:00 (period analysis 24/10–07/11: night levels unchanged on 01–07/11, pale area under the spot on 04/11); (2) the "partial return on the evening of the 12th" falls inside the camera back-and-forth 12/10 22:30 → 13/10 00:15 (Displacement, period 30/09–13/10): at 23:00 PR17 corr_x 0.162 → 0.109, corr_y −0.032 → +0.016, while PR01 corr_y also rises (0.151 → 0.177) and comes back by 00:30 (0.146); PR17 does not come back (corr_x 0.09–0.10, corr_y about 0 until 01:30): its own step, coinciding with the camera movement, now stated.
- **NORDEND-195** (Target issue, CTR01, 16/10 19:00 → 01/11 00:00): **corrected** (end → 07/11 14:00, description), review removed. CTR01 rejected at 1237 of 2086 measurements (59 %) from 16/10 19:00 to 07/11 14:00, every day (23 of 23), 83–97 % on 23–24/10 and 31/10–02/11, 31–66 % on the other days; its own rejections (before 16/10 only sunny afternoons). Its accepted points do not leave their level by 5 noises (`ranges.py`, 0 ranges; noise 0.016–0.018 px), so the label rests on the rejection regime, kept as a target rejected most of the time every day. Correction applied: the end, 01/11 00:00, was a batch edge; the regime continues to 07/11 14:00 (01/11 96 %, 02/11 97 %, 03–07/11 33–66 %); the text "until the end of the month … almost without stopping in the last days" becomes the description of the extended range.
Overlaps to report: NORDEND-284 (PR17, 01/11 00:00 → 07/11 14:00) and NORDEND-283 (CTR01, same bounds), written by the analyst of the 24/10–07/11 period because 169 and 195 could not be changed, now lie entirely inside the extended 169 and 195: they become duplicates (one label per target and per regime) and should be removed from `labeling/auto/NORDEND.jsonl` when these corrections are submitted. (NORDEND-230, which that analyst had moved to 01/11, is no longer in `NORDEND.jsonl`.) Not changed here (out of scope: never written in `NORDEND.jsonl`).
Sources read: `tools/GUIDE.md`, `tools/DATA.md`, `labeling/families.md`, `labeling/labeling-rules.md`, `labeling/reading-rules.md`, `labeling/stations.md`, `input.jsonl`, `labeling/auto/NORDEND.jsonl` (NORDEND-230, 283, 284 and labels after 06/11), the analyses `NORDEND_2026-09-30T0415_2026-10-13T0745` (12/10 camera movements, NORDEND-152) and `NORDEND_2026-10-24T1630_2026-11-07T1400` (continuation of 169 and 195), `data/NORDEND/targets.csv`. Commands: `scripts/rejections.py` (rejection rates per day, through `tools/scripts/data.py`), `ranges.py --k 5 --min-measurements 6` on PR04/PR03/PR05 (both PR04 labels), CTR01, PR17/PR16/PR18 with `--outside` and the mm columns (`work/ranges_*.txt`), `summary.py --step 15min` on 12/10 21:30 → 13/10 01:30 (PR17, PR16, PR18, PR01). Labels written by `scripts/write_labels.py`; `validate.py`: 0 errors (one overlap flagged: 152 and 195, two different targets, expected).