Files
aimsight-labeling/auto/analyses/SNCF01_2026-10-04T0000_2026-10-14T0645/scripts/build.py
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

108 lines
10 KiB
Python

"""Builds labels.jsonl: kept labels copied verbatim from work/existing.jsonl (without id), corrected ones rewritten."""
import json, sys
O = sys.argv[1]
old = {json.loads(l)["id"]: json.loads(l) for l in open(f"{O}/work/existing.jsonl")}
KEEP = [9, 10, 12, 18, 19, 20, 21, 22, 23, 25, 26, 33, 35, 37]
SH = ("Deduced cause: heat shimmer, the air heated by the sun above the tracks and the ballast, which makes the spots "
"tremble (a fast noise, by day only, with the recognition error of these low targets rising), nothing in front "
"of the targets on the images. Do not trust the small movements of these curves over the range; this is not a "
"movement of the structure.")
new = {
8: ("2026-10-04T08:45Z", "2026-10-04T16:30Z", "Visibility",
"By day (04/10), sunny day (cast shadows on the wall on the image of 12:00): the corrected curves of the low targets "
"fluctuate from one measurement to the next beyond their noise, mostly CV.4 (corr_x and corr_y), F4.2, BG.2, BG.3, with "
"one-measurement jumps that come back at once; late in the morning, BG.3, F4.2, BG.2 and the P targets jump together in "
"corr_x for one measurement; CV.2 and CV.4 are rejected at times (missing points). At night everything is calm again. " + SH),
11: ("2026-10-05T07:00Z", "2026-10-05T15:15Z", "Visibility",
"By day (05/10), sunny day (cast shadows on the image of 12:00): the corrected curves of the low targets fluctuate from "
"one measurement to the next beyond their noise, the most strongly of the batch: CV.4 and CV.2 in both columns, F4.2, "
"BG.3, and in late morning the corr_y of P1, P2, P3 jump together for one or two measurements; CV.2, CV.4 and F4.2 are "
"rejected at times (missing points). Calm again before nightfall; the train of 16:00 is labelled apart. " + SH),
14: ("2026-10-06T07:00Z", "2026-10-06T13:15Z", "Visibility",
"By day (06/10), sunny morning (cast shadows on the image of 12:00): the corrected curves of the low targets fluctuate "
"beyond their noise, mostly CV.4 in corr_x, F4.2 in both columns and CV.2, with one-measurement jumps that come back at "
"once; CV.2 and CV.4 are rejected at times (missing points). The other targets stay within their noise; calm again from "
"early afternoon. " + SH),
15: ("2026-10-06T07:15Z", "2026-10-07T03:00Z", "Thermal displacement",
"Daytime thermal cycle (06/10), sunny day (cast shadows on the image of 12:00): from the morning, the pan and the tilt of "
"the camera drop strongly and its scale rises, until early afternoon; they come back slowly in the afternoon, the evening "
"and the first half of the night. The compensation absorbs only part of it: corr_x of P7 rises clearly, with corr_y of "
"P6 and P7, while corr_x of P1 drops and that of P5 rises; the curves come back to their night level late in the evening, "
"the pose in the night. Deduced cause: the camera and its mount move with the heat of the day (camera pose); the "
"structure, a masonry wall, is not supposed to move like that. No temperature is available to confirm it. Normal, "
"nothing to do."),
16: ("2026-10-07T07:30Z", "2026-10-08T04:15Z", "Thermal displacement",
"Weak daytime thermal cycle (07/10), overcast day (image of 12:00): in the morning, the pan and the tilt of the camera "
"drop a little and its scale rises a little, then the camera stays in that position all afternoon; it comes back in the "
"evening and all night, beyond its position of the previous night, until early on the 8th. The corrected curves follow, "
"less than on the sunny days: corr_x of P5 and P7 rises, corr_y of P7 and P6 too, corr_x of P1 drops; P7 comes back late "
"in the evening, P5 only after midnight. The three train passages of the day (13:00, 16:00, 19:45) are labelled apart. "
"Deduced cause: the camera and its mount move with the heat of the day (camera pose); the structure, a masonry wall, is "
"not supposed to move like that. No temperature is available to confirm it. Normal, nothing to do."),
17: ("2026-10-07T12:45Z", "2026-10-07T13:15Z", "Obstruction",
"One single measurement: most targets (the CV, the F4 except F4.5 and F4.6, and all the P) are rejected and their curves "
"are missing; the references are affected too (R1 rejected, BG.3.1 just under its threshold and left out), the "
"compensation retains none of them; the targets that stayed under the threshold jump a little beyond their noise, BG.3 in "
"corr_x, F4.5 and F4.6 in corr_y. Everything comes back at the next measurement. Cause seen: a train passing in front of "
"the wall and the targets, blurred, on the image of 13:00. The structure is not moving: point to ignore, nothing to do."),
27: ("2026-10-10T07:00Z", "2026-10-10T13:00Z", "Visibility",
"By day (10/10), sunny morning (cast shadows on the image of 12:00): the corrected curves of the low targets fluctuate "
"beyond their noise, mostly CV.4 in both columns, CV.9 and CV.2 in corr_y, F4.2 and F4.5, and in mid-morning BG.3, P2, "
"P3 jump together in corr_x for one measurement; they come back at once. Calm again from early afternoon. " + SH),
28: ("2026-10-11T10:00Z", "2026-10-11T15:30Z", "Visibility",
"By day (11/10), overcast day (image of 12:00), the weakest agitation of the batch that still crosses the noise: CV.4 in "
"corr_y, F4.2, CV.2, and P1, P2 in corr_y make a few one-measurement jumps just beyond their noise, coming back at once; "
"the train of 14:15 is labelled apart. Calm at night. Deduced cause: a weak shimmer of the air above the tracks by day, "
"the same signature as on the sunny days, much weaker; nothing in front of the targets on the images. Do not trust the "
"small movements of these curves over the range; this is not a movement of the structure."),
29: ("2026-10-11T08:30Z", "2026-10-12T02:45Z", "Thermal displacement",
"Daytime thermal cycle (11/10), overcast day (image of 12:00): from the morning, the pan and the tilt of the camera drop "
"and its scale rises, until early afternoon; they come back in the late afternoon, the evening and the first half of the "
"night, beyond their position of the previous night. The compensation absorbs part of it: corr_x of P7 rises clearly, "
"with corr_y of P6 and P5, while corr_x of P1 drops; the curves are back at their night level in the evening, the pose in "
"the night. The train of 14:15 is labelled apart. Deduced cause: the camera and its mount move with the heat of the day "
"(camera pose); the structure, a masonry wall, is not supposed to move like that. No temperature is available to confirm "
"it. Normal, nothing to do."),
30: ("2026-10-11T14:00Z", "2026-10-11T14:30Z", "Obstruction",
"One single measurement: most targets (the CV, the F4 and all the P) are rejected and their curves are missing; the "
"references are affected too (BG.3.1 and R1 rejected), the compensation retains none of them, and the targets that "
"stayed under the threshold, BG.2, BG.3 and F4.5, jump strongly in corr_x and corr_y for the time of the measurement. "
"Everything comes back at the next measurement. Cause: a passage in front of the targets and the references; the camera "
"takes its image on the hour, none at that measurement; deduced from the signature, the same as that of the trains seen "
"on the images of the other passages. The structure is not moving: point to ignore, nothing to do."),
31: ("2026-10-12T06:30Z", "2026-10-12T07:00Z", "Obstruction",
"One single measurement, early in the morning: CV.8 and F4.2 go above their threshold (CV.6 too), their curves are "
"missing, and the recognition error of most CV and F4 targets and of the reference BG.3.1 rises; CV.4, left under its "
"threshold with its recognition error dropping, jumps clearly in corr_x for that measurement, and comes back at the next "
"one. The references hold, the other targets stay within their noise. Cause deduced from the signature: a partial passage "
"in front of the low targets, CV.4 latching onto a light or a reflection; no image at that measurement (the camera takes "
"one on the hour). The structure is not moving: point to ignore, nothing to do."),
32: ("2026-10-12T07:30Z", "2026-10-12T14:45Z", "Visibility",
"By day (12/10), sunny day (cast shadows on the image of 12:00): the corrected curves of the low targets fluctuate beyond "
"their noise, mostly CV.4 in both columns, F4.2, CV.2, F4.5, F4.6, and in mid-morning BG.3, F4.2, F4.6, P2, P3, P4, P6 "
"jump together in corr_x for one measurement; they come back at once; CV.2 and CV.4 are rejected at times (missing "
"points). Calm again in the afternoon. " + SH),
34: ("2026-10-12T15:45Z", "2026-10-12T16:15Z", "Obstruction",
"One single measurement: most targets (the CV, the F4 and all the P) are rejected and their curves are missing; the "
"references are affected too (BG.3.1 and R1 rejected), the compensation retains none of them, and the targets that "
"stayed under the threshold, BG.2, BG.3 and F4.5, jump strongly in corr_x and corr_y for the time of the measurement. "
"Everything comes back at the next measurement. Cause seen: a train passing in front of the wall and the targets, "
"blurred, on the image of 16:00. The structure is not moving: point to ignore, nothing to do."),
36: ("2026-10-13T07:45Z", "2026-10-13T14:00Z", "Visibility",
"By day (13/10), sun and clouds (image of 12:00): the corrected curves of the low targets fluctuate beyond their noise, "
"mostly CV.4 in corr_x, F4.2, F4.6, CV.2, and around midday P1, P2 and F4.2 jump together in corr_x for one measurement; "
"they come back at once. Calm again in the afternoon. " + SH),
}
out = []
for i in sorted(KEEP + list(new)):
if i in KEEP:
d = dict(old[f"SNCF01-{i}"]); d.pop("id")
else:
s, e, f, desc = new[i]
d = {"start": s, "end": e, "family": f, "description": desc}
out.append(d)
with open(f"{O}/labels.jsonl", "w") as fh:
for d in out:
fh.write(json.dumps(d, ensure_ascii=False) + "\n")
print(len(out))