Files
aimsight-labeling/auto/analyses/NORDEND_2026-09-17T0545_2026-09-30T0415/analysis.md
T

12 KiB

Labels: NORDEND, 2026-09-17T05:45Z → 2026-09-30T04:15Z (UTC)

Task: verify the 23 existing labels of the range (work/existing.jsonl, all reviewed by the user on 2026-11-17) and look for what they miss. Result: 23 kept, 0 corrected, 0 removed, 1 added.

Weather: data/NORDEND/weather/ does not exist: no MeteoSwiss weather for this station (no Swiss position); the causes were read on the images and the curves only.

Station: export columns peak to snr empty over the window: no image-quality metrics, quality.py not applicable. 1243 instants, every 15 min, none missing, a pose at every instant, an image every 15 min. Three references CTR02, CTR03, CTR04, all retained at every instant except 20/09 09:00-09:45 and 10:15-11:30 (CTR04 rejected).

Verdict per old id

Old id Range (UTC) Family Verdict Evidence
NORDEND-126 17/09 05:45 → 18/09 06:30 Thermal displacement kept pan drops from 06:30, tilt up from 06:00 (max 09:15), scale up; PR18 corr_x about -0.07 px at 09:15, PR13-PR16 small dips; next cycle starts at the turning point 18/09 06:30 (pan 17.7 → 15.5 mdeg at 07:00, tilt leaving -120). Remark: PR18 corr_y rises about as much as on 19/09 (+0.04 px at 09:30), where the label says "rises"; "hardly moves" is a soft wording, not wrong enough to cancel the review
NORDEND-127 17/09 20:00 → 22:45 Obstruction kept PR07 rejected 20:30, 21:45-22:15; PR14, PR15 corr_x +7, +8 noises at 20:15, PR11, PR16 at 20:30, PR16 corr_y -8.6 noises at 22:00-22:15, PR07 corr_x +7 noises at 22:30, nothing at 22:45: bounds right
NORDEND-128 18/09 06:30 → 19/09 06:15 Thermal displacement kept turning point 06:30; fast part 07:00-07:30; PR18 corr_x -0.083 px at 08:00, PR16 -0.054, PR15 -0.048, PR14 -0.034, PR18 corr_y +0.047; next start 06:15 (pan 14.9 → 13.4 mdeg at 06:45, tilt -118 → -115)
NORDEND-129 18/09 14:45 → 16:00 Visibility kept CTR01 rejected 15:00-15:45 (4), sunny-afternoon signature; bounds right
NORDEND-130 19/09 06:15 → 20/09 07:30 Thermal displacement kept PR18 corr_x -0.078 px at 09:00, PR16 -0.052, PR18 corr_y +0.040; tilt max 08:45
NORDEND-131 20/09 07:30 → 21/09 06:00 Thermal displacement kept weak day: tilt only +5 noises (13:30); pan up all night (+0.02 deg at 21/09 01:15), tilt down all night (-0.022 deg at 00:45); the visibility drop 08:45-12:00 is labelled apart (132)
NORDEND-132 20/09 08:45 → 12:00 Visibility kept 08:45 still normal on every sensor (the rolling flags at 08:45 come from a median pulled by 09:00); CTR04 rejected 09:00-09:45, 10:15-11:30, compensation on two references (CTR02, CTR03 corr at 0); 13 to 19 targets rejected at times; PR08 corr_x -0.48, PR06 -0.42, PR16 -0.40 px; 10:00 one-measurement return when CTR04 comes back; 11:45 all back except PR07 corr_x +14 noises; 12:00 back: end right
NORDEND-133 20/09 15:15 → 15:45 Visibility kept CTR01 rejected at 15:30 alone
NORDEND-134 21/09 06:00 → 22/09 06:30 Thermal displacement kept tilt max +0.042 deg at 09:00, pan min 14:00; PR18 corr_x -0.125 px, PR16 -0.071, PR15 -0.062, PR14 -0.049, PR18 corr_y +0.056. Remark: on 22/09 the tilt already steps at 06:15 (-137.6 → -132.3 mdeg), the pan leaves at 06:00 → 06:15, PR03 and PR05 corr_y rise at 06:15-06:30; the turning point is 06:00, two steps before the cut 06:30 between 134 and 136; both are Thermal displacement, so every instant stays under the right label: not changed
NORDEND-135 21/09 14:30 → 16:15 Visibility kept CTR01 rejected 14:45-16:00 (6); my thumbnail of 21/09 15:00 shows the rock around CTR01 in full sun
NORDEND-136 22/09 06:30 → 23/09 06:00 Thermal displacement kept tilt max +0.052 deg at 08:15; PR18 corr_x -0.126, PR16 -0.069, PR15 -0.065, PR14 -0.051, PR18 corr_y +0.065
NORDEND-137 22/09 14:30 → 16:15 Visibility kept CTR01 rejected 14:45-16:00 (6)
NORDEND-138 23/09 06:00 → 24/09 07:00 Thermal displacement kept tilt from -137 mdeg at 05:45 to -83 at 07:30 (fast part 07:15-07:30); PR18 corr_x -0.141 px, PR14-PR16 -0.066 to -0.076 at 08:00, PR18 corr_y +0.075; pan back up all night
NORDEND-139 23/09 14:15 → 15:45 Visibility kept CTR01 rejected 14:45-15:00; corr_x already -0.07, -0.06 at 14:00-14:15, CTR02 corr_x +0.016 → -0.017 at 15:00-15:15 (largest jump of CTR02, ten noises), CTR03, CTR04 at the same instant
NORDEND-140 24/09 07:00 → 25/09 06:00 Thermal displacement kept weakest day: PR18 corr_x -0.062, others -0.02 to -0.03, PR18 corr_y +0.035; pan up all night (+0.021 deg at 25/09 03:15)
NORDEND-141 24/09 11:15 → 26/09 12:45 Obstruction kept CTR01 rejected from 11:30, PR09 from 11:45, PR01 at times from 25/09 08:00 to 26/09 11:45; last rejections of CTR01 and PR09 at 26/09 12:30, 12:45 clean; from 24/09 06:00 to 11:15, PR09 and CTR01 at their usual level and error_value. Error_value of CTR02 and PR17 up (60 and 220 noises) without effect on the corr of the others, except the one-measurement common x dip of 25/09 14:00 (section 2)
NORDEND-142 25/09 06:00 → 26/09 06:00 Thermal displacement kept tilt max +0.045 deg at 07:45; PR18 corr_x -0.128, PR16 -0.084, PR15 -0.060, PR14 -0.048, PR17 -0.038, PR18 corr_y +0.082
NORDEND-143 26/09 06:00 → 27/09 06:30 Thermal displacement kept tilt max +0.053 deg at 09:00; PR18 -0.131, PR16 -0.074, PR15 -0.070, PR14 -0.052 at 08:45
NORDEND-144 26/09 14:30 → 16:00 Visibility kept CTR01 rejected 14:45-15:45 (5)
NORDEND-145 27/09 06:30 → 28/09 05:45 Thermal displacement kept tilt max +0.045 deg at 09:45; PR18 -0.117, PR16 -0.073, PR15 -0.059, PR14 -0.047; next turning point 05:45 (pan 19.6 → 18.9 mdeg, tilt -125 → -122)
NORDEND-146 28/09 05:45 → 29/09 06:30 Thermal displacement kept tilt max +0.042 deg at 09:15; PR18 -0.109, PR16 -0.068, PR15 -0.047, PR14 -0.044
NORDEND-147 29/09 06:30 → 30/09 04:15 Thermal displacement kept tilt max +0.034 deg at 12:15, PR18 -0.093, PR15, PR16 -0.049; ends at the range edge, continued by NORDEND-149
NORDEND-148 29/09 14:30 → 16:00 Visibility kept CTR01 rejected 14:45, 15:15-15:45; 15:00 accepted at error_value 0.36 with corr_x -0.062 (inside); 16:00 error_value 0.36, corr_y 0.164 against 0.20 around, about three noises, like the usual afternoon noise: end kept

Spanning label NORDEND-24 (PR04, Target issue, 17/08 → 01/10): within the range, PR04 is rejected at times on 17-20/09, almost without interruption from 20/09 13:15 to 28/09 10:15, then more rarely on 28-29/09; its accepted corr are noisy (noise 0.017 against 0.005 for its neighbours). The label holds within the range.

1. My labels

The 23 kept labels are not repeated here (verdicts above). The added one:

2026-09-17T14:30Z → 2026-09-17T15:00Z

  • Phenomenon of the catalogue: The sun in the field or behind a spot (Visibility); a target left under the threshold keeps its jump (reading-rules.md, section 5)
  • Precedents: NORDEND-4, -7, -10, -13, -16 (CTR01 rejected on sunny afternoons, sun seen on the rock around it) and the kept 129, 135, 137 of this range: same target, same hour, same cause; here CTR01 stays just under its threshold, so its point stays
  • Family: Visibility, like every CTR01 sunny-afternoon label of the station (same target, same family); not Obstruction (nothing in front: the lit background), not Displacement (cause seen)
  • Measured: CTR01 corr_x between -0.059 and +0.013 from 12:30 to 14:30, -0.073 at 14:45, -0.012 at 15:00: a one-measurement jump of -0.079 px (largest jump of CTR01 in the range, 8.4 noises), back at 15:00; error_value 0.20 at 14:30, 0.353 at 14:45 (threshold 0.4), 0.18 at 15:00; corr_y 0.191 / 0.197 / 0.190 (unchanged). No other sensor beyond 4 noises at 14:45 except PR17 corr_y -4. CTR01 is not rejected that afternoon
  • Images: thumbnails 17/09 14:00, 14:45, 15:15 and 21/09 15:00 (ctr01_sun_2026-09-17.png): at 14:45 the rock around CTR01 is bright (background 113 against 33 at 14:00 and 20 at 15:15), the spot drowning, the same picture as 21/09 15:00 (rejected then); at 15:15 the area is in shade and the spot sharp
  • Confidence: high for the cause (seen); medium for the need of a label: one point, but clearly apart from the noisy afternoon of CTR01 and at the very hour it is rejected on the other sunny days

2. What I left without a label

  • 17/09 14:15, 19/09 10:45, 19/09 13:45, 20/09 07:30: error_value up on 5 to 10 sensors together for one measurement (CTR04, PR06-PR08, PR13-PR18), by day; corr at those instants within 4 noises of the rolling level, except PR02 corr_y +6 noises at 19/09 10:45 (one target, 0.04 px): an error_value rising without effect on the curves.
  • 25/09 14:00: PR08, PR16 corr_x -7 noises, PR14 -6, PR15, PR17, PR18 -5, PR06, PR12 -4, all in x, one measurement, back at 14:15. overview.py --rigid on 25/09 12:00-16:00: the CTR04 residual dx moves by +0.027 px at that instant (-0.016 → +0.012 → -0.026), the base shifts slightly while the camera is agitated in the warm hours (ty jumping by half a pixel from one measurement to the next): the afternoon agitation of the thermal cycle, under NORDEND-142, a few hundredths of a pixel, much smaller than the cycle itself.
  • 22/09 10:00-10:30: small common corr_y dip (PR10 -7 noises, others -4/-5), then CTR02, CTR03 corr_y diverging at 10:30 (±6 noises), CTR04 corr_x -7: within the warm-hours agitation of NORDEND-136.
  • Morning steps under the thermal labels (18/09 07:30-08:00, 20/09 06:45-08:00 PR07, PR16, 23/09 07:30, 25/09 07:30-07:45, 26/09 08:45, 27/09 07:45-08:15 PR18, PR16): one-measurement deviations of 4 to 8 noises of the targets that carry the cycle, during the fast part of the camera movement: part of the cycle.
  • 24/09 09:15-11:15: PR09 corr_y -5/-6 noises at times: its usual morning scatter (0.09 to 0.14 since 06:00), error_value at its level; the cobweb starts at 11:30.
  • 27/09 11:15: CTR01 corr_y +6 noises for one measurement, error_value 0.23 (its usual level): a single noisy point, not at the sun hour. 27/09 14:45-15:15: CTR01 error_value 0.33-0.35, corr unchanged.
  • CTR02, CTR03, CTR04 corr at 21/09 17:00, 22/09 10:30: references, 5 to 7 noises of thousandths of a pixel, no target follows.

3. The documentation

  • Helped: labeling-rules.md section 2 (the camera cycle starts at the morning turning point; one-step tolerance) to judge the cuts between the thermal labels; section 1 and reading-rules.md section 5 ("a target left under the threshold keeps its jump") for the added CTR01 point; families.md (the boundary "Target issue or episode") and the precedents to keep CTR01 in Visibility; GUIDE.md: ranges.py --rolling --instants and overview.py --rigid to find and explain the short collective deviations.
  • The cut 134/136 on 22/09 falls two steps after the turning point (06:30 against 06:00): beyond the one-step tolerance, but it moves no instant out of a Thermal displacement label; the rules do not say whether such a shift between two consecutive thermal labels is worth losing a validation. I left it.
  • PR18 corr_y in the thermal descriptions ("hardly moves" on 17/09 and 24/09, "rises" on 19/09) with similar amplitudes on 17/09 and 19/09: the wording is relative to the day, not to a fixed scale.
  • No weather for NORDEND (no Swiss position): the instructions foresee it.
  • No precedent for a single under-threshold point of CTR01 in the sun: the added label is the first.

Difficulties

none

Sources read

  • agents/agents/analyst.md, instructions/analyst.md, work/prompt.txt, work/existing.jsonl, work/spanning.jsonl
  • tools/GUIDE.md, tools/DATA.md; labeling/families.md, labeling/labeling-rules.md, labeling/reading-rules.md, labeling/stations.md (in full)
  • labeling/auto/NORDEND.jsonl (labels from 15/09 to 02/10)
  • precedents.py ("CTR01 … one-measurement corr_x jump in the afternoon without being rejected …")
  • Tools: inventory.py (window), overview.py (window; 25/09 12:00-16:00 with --rigid), ranges.py --rolling 4 --k 6 --instants 2, days.py --bounds --night 2-5 --smoothing 5, summary.py (20/09 08:00-12:30: corr_x, corr_y, error_value), thumbnails.py (CTR01, PR09 at 17/09 14:00, 14:45, 15:15 and 21/09 15:00)
  • My scripts: scripts/look.py (per instant, the sensors beyond K noises of the rolling ±4 level, and the rejections), inline reads of the pose and of CTR01, PR09, PR01 and the references through tools/scripts/data.py