Files
aimsight-labeling/auto/analyses/NORDEND_2026-09-30T0415_2026-10-13T0745/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

30 KiB
Raw Blame History

Labels: NORDEND, 2026-09-30T04:15Z → 2026-10-13T07:45Z (UTC)

Request: verify the 30 existing labels of the range (work/existing.jsonl): keep, correct, remove, add; the 3 spanning labels (work/spanning.jsonl) are read as context. DETECTORS: no.

Weather: no weather for NORDEND (data/NORDEND/weather/ does not exist; inventory.py lists no weather source). The weather is read on the images only.

Station: 15 min cadence, no missing instant and no instant without a pose in the range (inventory.py); export image quality (peak to snr) empty, so quality.py does not apply; no DIFF images. Three references CTR02, CTR03, CTR04; CTR01 a local reference read as a target. Image positions (x px): CTR01 1782, PR01 1827, PR09 1845 (left edge), PR07 2185 (middle-right, not on the left edge), PR16 2430, CTR04 2401 and PR08 2391 (bottom right).

Requalification of 2026-11-17 (labeling-rules §1, new rules: only what stands out from the noise of the corrected curves): 21 kept as they are, 9 rewritten, 3 removed; 30 labels in labels.jsonl. Details in section 2 and in "## Requalification" below. Sections 1 and the verdict table describe the previous verification and stay as they were, except for the three removed labels.

Verdict per old id

Old id Verdict Evidence
NORDEND-149 kept pose leaves at 04:15 (batch start), pan −0.030° at 12:30, tilt +0.035° at 09:45, scale up, temperature 35→46 °C at 12:00; PR18 cx −0.074, PR16 −0.044, PR15 −0.040, PR18 cy +0.037; returns at the 01/10 start. Not named: the corr_y dip of PR01–PR03 (≈ −0.04), and a confusing "ends after the range"; not wrong enough to lose the validation
NORDEND-150 removed (merged into 151) PR07 stays disturbed between 150 and 151 (cx 20:15 −0.001 vs level 0.055, 20:45 0.019, rejected 21:00, back 21:45): one nuisance, not two; PR07 is not "on the left edge" (x 2185)
NORDEND-151 corrected (start 20:45→18:45, description) PR07 cx leaves at 19:00 (0.022 vs 0.053, error 0.076 vs 0.045), so start 18:45; merged with 150; thumbnails 18:30/19:15/20:15/21:15/22:30 show PR07 shrinking 19:15–21:15, CTR01/PR09 21:15–22:30
NORDEND-153 corrected (description) "the camera pose does not move more than usual" false: pan +5 mdeg at 06:45–07:15 and the right-hand targets jump (new label); PR07 is not on the left edge; bounds 03:00–08:00 confirmed (PR07 jump 03:15, CTR01/PR09 back 08:00)
NORDEND-154 kept CTR01 rejected 10:15 alone, back 10:30
NORDEND-155 corrected (description) "the corr_y … does the same, in the opposite direction" false: corr_y of PR13, PR06, PR07, PR03, PR14 rise +0.04–0.05 at night like PR18 cx (+0.061); PR18 cy drops a little
NORDEND-156 kept start 06:00 (pan leaves after 05:30, tilt from 05:45), temperature max 12:00–13:00, pan −38 mdeg, tilt +39, PR18 cx −0.086 at 10:15, PR16/PR15/PR14/PR17 follow; corr_y dip led by PR01, PR03, PR02, PR07 (PR05 named but smaller): approximate, not wrong
NORDEND-157 corrected (description) sun also behind CTR02/CTR03 (15:45 thumbnail); bounds kept
NORDEND-158 kept start 06:00, max temperature 11:00, PR18 cx −0.097 at 09:30, PR18 cy +0.052; corr_y dip wider than "a few targets" (PR01 −0.055, PR02 −0.052): approximate
NORDEND-159 corrected (description) sun behind CTR02/CTR03 (15:30 thumbnail, CTR03 drowned), CTR03 error 0.09–0.11, small common corr_y jumps, PR02 dips / PR01 rises
NORDEND-160 kept start 06:30, pan −39, tilt +45, PR18 cx −0.104 at 11:00, corr_y dip at 09:30 led by PR02, PR10, PR01, PR03
NORDEND-161 corrected (description) same as 159 (15:15 thumbnail, CTR03 drowned)
NORDEND-162 kept start 06:45, temperature plateau 11–13 h, PR18 cx −0.104 at 09:45, PR15/PR16 −0.056/−0.058, PR18 cy +0.056
NORDEND-163 corrected (description) same as 159 (15:15 thumbnail; CTR03 error 0.08–0.12 from 14:45)
NORDEND-164 kept start 06:30, temperature 25→44.9, pan −51, tilt +47, PR18 cx −0.128 at 10:30 (called "strongly", the largest of the batch)
NORDEND-165 corrected (start 14:45→14:00, description) common corr_y jumps with tilt jumps at 14:15 (+0.011, median of the targets) and 14:30 (−0.010); sun behind CTR03 already at 14:15 (thumbnail)
NORDEND-166 kept start 06:45, PR18 cx −0.111 at 10:00, PR18 cy +0.061; cobweb night labelled apart
NORDEND-167 kept cobweb seen at 19:45 and 21:00 (left of the image); CTR01 17:15→13:30, PR09, PR07, PR04 rejections
NORDEND-168 kept spider body at the top of the 19:45 image, PR01 rejected 19:45, all targets and CTR02/CTR03 jump
NORDEND-170 corrected (05:45→08:30 instead of 07:00→07:45, description) CTR04 error rises from 06:00 (0.095, 0.169, 0.319…), right-hand targets jump on corr_x at 06:00 (PR15 +0.04, PR14 +0.04, PR08 +0.055); after its return CTR04 stays high to 08:15 (0.299), back 08:30 (0.058); PR16, PR14 off at 07:45–08:15
NORDEND-171 kept weak day (temperature 29.5→35.1), pan −12, PR18 cx dips to midday
NORDEND-172 corrected (description) 15:00 thumbnail: sun behind CTR02/CTR03 too, small common jump
NORDEND-173 kept 07:15–15:15: error_value of 19–21 sensors up from 07:30, CTR04 rejected at times 08:45–12:15, CTR01 back 15:15
NORDEND-174 kept PR07 jump 17:15 (first touched point), CTR04 rejected 21:15–23:15, CTR01 back 08:15
NORDEND-175 corrected (description) "the tilt, unlike on sunny days, drops a little" false: the tilt rises by day (−132 → −124 mdeg) and drops only at night; the pan drops ≈ 14 mdeg ("a little", not "hardly")
NORDEND-176 corrected (start 15:00→14:30, description) "The other targets remain measured" false: PR07 rejected 15:15; PR16 cx +0.08 at 15:15, back 16:00; spots smaller 15:15–15:45; sun behind CTR03 at 14:45 with a tilt jump and a common corr_y jump (CTR02 cy +0.027)
NORDEND-177 corrected (description) "PR08, which alone drops very strongly … is not the camera cycle" false: PR08 cy dips ≈ −0.03 from 08:00 to 17:00, as on 10 and 12/10; PR16 −0.027, PR17 +0.040, PR01 +0.034: part of the cycle
NORDEND-178 corrected (start 15:15→14:45, description) common corr_y jump at 15:00 (PR01 +0.026, PR02 +0.035, PR10 +0.03, CTR02 +0.025) with a tilt jump; sun behind CTR03 at 15:00 and 15:30 (thumbnails), CTR03 error 0.082/0.119
NORDEND-179 corrected (end 13/10 07:45→12/10 19:30, description) a camera step at 19:45 (new Displacement label) stops the cycle; the temperature then rises (29.6→31.1) and the pose stays at its new level
NORDEND-180 corrected (end 16:00→16:15, description) at 16:00 the tilt jumps and PR01, PR02, PR10, PR11 drop (−0.04 to −0.05), back at 16:15; CTR03 drowned in the sun at 15:30 and 16:00 (thumbnails)

Added: 2026-10-01T06:30Z → 07:45Z (Visibility, daybreak jump of the right-hand targets); 2026-10-09T06:30Z → 2026-10-10T08:15Z (Thermal displacement, unlabelled cycle); 2026-10-12T19:30Z → 20:00Z (Displacement, camera step); 2026-10-12T22:30Z → 2026-10-13T00:15Z (Displacement, camera back-and-forth).

Counts: 13 kept, 16 corrected, 1 removed, 4 added; 33 labels in labels.jsonl (scripts/build_labels.py builds it from work/existing.jsonl; validate.py: 0 errors, 2 overlaps of distinct Visibility events).

1. My labels

2026-09-30T04:15Z → 2026-10-01T10:00Z (kept, NORDEND-149)

  • Phenomenon of the catalogue: The camera moves through a thermal effect
  • Precedents: the NORDEND cycles of the batch (156–166), for the wording
  • Family: Thermal displacement: the pose follows the housing temperature and comes back
  • Measured: scripts/cycle.py: temperature 35.1→46.0 (12:00); pan −0.030° (12:30), tilt +0.035° (09:45), scale +4.1e-4; corr_x PR18 −0.074 (10:00), PR16 −0.044, PR15 −0.040, PR14 −0.030; corr_y PR02 −0.040, PR03 −0.036, PR01 −0.036, PR18 +0.037
  • Images: none
  • Confidence: high

2026-09-30T18:45Z → 2026-09-30T23:45Z (corrected, NORDEND-151 with 150)

  • Phenomenon of the catalogue: Part of the scene goes dark at night (faint spots, nothing in front)
  • Precedents: NORDEND-150/151 themselves
  • Family: Visibility: spots fade without an obstacle
  • Measured: PR07 corr_x 18:45 0.053, 19:00 0.022, 19:15 0.160, 19:30 0.017, 20:15 −0.001, 20:45 0.019, rejected 21:00, 21:45 0.054 (level 0.055, noise 0.0055); error_value 0.045→0.076 (19:00), 0.171 (19:15); PR16 cx 0.137→0.189/0.184/0.166, back 0.135 at 20:00; PR15 +0.03 at 19:15. CTR01 error 0.30 at 19:15, rejected 21:00–23:30; PR09 error 0.084 (19:15) → 0.53 (23:00), rejected 22:15–23:15. Pose quiet
  • Images: thumbnails 18:30, 19:15, 20:15, 21:15, 22:30, 00:30 (figures/night_spots_3009.png): PR07 small 19:15–21:15, whole at 22:30; CTR01 and PR09 smallest 21:15–22:30; PR16, PR05 unchanged
  • Confidence: medium (cause not established; merge per labeling-rules §3, "two nuisance episodes … where targets stay shifted make only one")

2026-10-01T03:00Z → 2026-10-01T08:00Z (corrected, NORDEND-153)

  • Phenomenon of the catalogue: Part of the scene goes dark at night
  • Precedents: none useful beyond the station's own
  • Family: Visibility
  • Measured: PR07 cx 0.127 at 03:15, rejected 03:30 and 06:45–07:00; PR09 rejected 04:15–05:30, 06:15, 06:45–07:45; CTR01 04:15–04:30, 05:15–05:30, 07:00–07:45; back at 08:00
  • Images: thumbnails 02:00, 03:30, 05:00, 07:30, 08:30: PR07 almost gone at 03:30, PR09 smaller at 05:00, PR09 and CTR01 very faint at 07:30; PR05 whole
  • Confidence: medium

2026-10-01T06:30Z → 2026-10-01T07:45Z (added)

  • Phenomenon of the catalogue: none exactly; close to "A reference loses its brightness, without an obstacle", and to NORDEND-170 (CTR04 at daybreak)
  • Precedents: LENZSPITZE-83 (part of the scene loses light before daybreak, targets jump: Visibility), JUNGFRAU-50/89 (a reference seen less well, small common jumps: Visibility)
  • Family: Visibility (change of light at daybreak, nothing in front; the curves come back)
  • Measured: corr_x 06:30→07:00: PR16 0.143→0.196/0.192, PR15 0.126→0.172, PR06 0.072→0.115, PR08 0.059→0.103, PR14 0.128→0.172, PR18 0.223→0.271; corr_y PR16, PR06, PR08 ≈ +0.05 at 07:00; back at 07:45. Raw x 06:30→07:00: CTR02 +0.05, CTR03 +0.07, PR01 +0.04, PR16 +0.04, PR18 +0.08, CTR04 −0.01. Pan 5.4→10.1/11.7 mdeg, then 6.7. Reference error_values flat (CTR04 0.026–0.030)
  • Images: zone thumbnails 06:15–07:45: 06:45 darker background with bright spots, 07:00 daylight; whole image 07:00: overcast, scene clear
  • Confidence: medium (that CTR04 is the odd one is deduced from the raw positions)

2026-10-01T10:00Z → 2026-10-01T10:30Z (kept, NORDEND-154)

  • CTR01 rejected at 10:15 only. Visibility. Confidence medium.

2026-10-01T10:00Z → 2026-10-02T06:00Z (corrected, NORDEND-155)

  • Phenomenon of the catalogue: The camera moves through a thermal effect
  • Precedents: reading-rules §3 cites NORDEND-155 itself
  • Family: Thermal displacement
  • Measured: temperature 36.1→38.0 (13:00)→28.1; pan −9 mdeg by day, +22 at night; tilt +11 / −19; scale −3.4e-4 at night. PR18 cx 0.225→0.289 (05:00); PR16 +0.024, PR17 +0.023, PR15 +0.021; corr_y 09:00→05:00: PR13 0.050→0.104, PR06 0.059→0.104, PR07 0.081→0.127, PR03 0.121→0.156, PR14 0.013→0.053; PR18 cy −0.199→−0.215; CTR02 cy +0.014, CTR03 −0.010
  • Images: none
  • Confidence: high

Kept thermal cycles: 2026-10-02T06:00Z → 10-03T06:00Z (156), 10-03T06:00Z → 10-04T06:30Z (158), 10-04T06:30Z → 10-05T06:45Z (160), 10-05T06:45Z → 10-06T06:30Z (162), 10-06T06:30Z → 10-07T06:45Z (164), 10-07T06:45Z → 10-08T09:30Z (166), 10-08T09:30Z → 10-09T06:30Z (171)

  • Phenomenon of the catalogue: The camera moves through a thermal effect
  • Family: Thermal displacement
  • Measured: morning turning points on pan and tilt at 15 min (scripts/matrix.py pan|tilt 2 11): 02/10 pan leaves 06:00, tilt 05:45; 03/10 pan slides from 04:45, tilt 05:45–06:00; 04/10 tilt 06:15, pan 06:30–06:45; 05/10 pan peak 06:15, tilt 06:45; 06/10 both after 06:30; 07/10 tilt 06:30–06:45; all within one step of the bounds. Extremes per day: work/cycles.txt (verdict table)
  • Images: none
  • Confidence: high (the targets named for corr_y are approximate, not wrong in direction)

Sunny afternoons, corrected: 2026-10-02T15:30Z → 16:15Z (157), 10-03T14:45Z → 16:15Z (159), 10-04T14:45Z → 16:15Z (161), 10-05T14:45Z → 16:15Z (163), 10-06T14:00Z → 16:15Z (165), 10-08T14:45Z → 15:15Z (172), 10-10T14:30Z → 16:00Z (176), 10-11T14:45Z → 15:45Z (178), 10-12T14:00Z → 16:15Z (180)

  • Phenomenon of the catalogue: The sun in the field or behind a spot (CTR01), and A reference loses its brightness (CTR02, CTR03): one nuisance
  • Precedents: JUNGFRAU-50/55/89 (the sun behind a reference, small common jumps, Visibility; validated)
  • Family: Visibility (families.md: the lit background behind a spot is the same nuisance as the sun in the field)
  • Measured: CTR01 rejections from the inventory (02/10 15:45–16:00; 03, 04, 05/10 15:00–16:00; 06/10 15:00–15:30 and 16:00; 08/10 15:00; 10/10 15:15–15:45 and PR07 15:15; 11/10 15:30; 12/10 14:15 and 15:30–15:45). CTR03 error_value ×1000 (scripts/matrix.py CTR03:error_value 13 17.5): ≈ 60–70 usually, 82–137 in the sunny windows (not raised on 02/10). Common corr_y jump of the targets (scripts/common_jump.py corr_y 13.5 17): ±0.007–0.009 on 03–05/10; 06/10 14:15 +0.011, 14:30 −0.010, 15:15 +0.011; 11/10 15:00 and 15:30 +0.017; 12/10 16:00 −0.015, with tilt jumps of 8–13 mdeg. PR02 cy dips ≈ −0.03 at 15:30–15:45 on 03, 04, 05, 06, 11, 12/10; PR01 cy rises 15:15–15:45. 10/10: PR16 cx 0.152→0.238 at 15:15, back 0.152 at 16:00; 14:45 tilt −138.4, CTR02 cy +0.027. 11/10: figures/sun_jumps_1110.png. 12/10 16:00: PR01 cy 0.134, PR02 0.140, PR10 0.086, PR11 0.082, back at 16:15
  • Images: CTR02/CTR03 thumbnails 02/10 15:45, 03/10 15:30, 04/10 15:15, 05/10 15:15, 06/10 14:15 and 15:15, 08/10 15:00, 10/10 14:45/15:00/15:30, 11/10 15:00/15:15/15:30, 12/10 15:30/16:00 (figures/sun_behind_references.png): the rock behind them sunlit, CTR03's spot drowned (except 02/10 and 08/10, visible), in shade at 10/10 15:00 and 11/10 15:15; CTR01 zone 05/10 13:30 (shade) and 15:30, 08, 10, 11, 12/10 (sun)
  • Confidence: high

2026-10-07T17:00Z → 2026-10-08T13:45Z (kept, NORDEND-167), 2026-10-07T19:30Z → 20:00Z (kept, NORDEND-168)

  • Images: whole images 19:45 (spider body at the top, web) and 21:00 (web threads across the left). Confidence high.

2026-10-08T05:45Z → 2026-10-08T08:30Z (corrected, NORDEND-170)

  • Phenomenon of the catalogue: A reference dropped: the base changes (family of the cause)
  • Precedents: none better than the station's own
  • Family: Visibility (reference seen less well at the change of light, nothing in front)
  • Measured: CTR04 error 05:45 0.048, 06:00 0.095, 06:15 0.169, 06:30 0.319, 06:45 0.293, 07:00 0.222, rejected 07:15–07:30, 07:45 0.390, 08:00 0.239, 08:15 0.299, 08:30 0.058. At 06:00 pan 22.2→26.4 mdeg, corr_x PR15 0.118→0.159, PR14 0.153→0.194, PR08 0.001→0.056, PR06 +0.044; at 07:15–07:30 all targets −0.1 to −0.5; after: PR16 cx 0.252 (07:45), PR16 cy ≈ −0.01, PR08 cx −0.02
  • Images: 07:15 image (overcast, sharp scene), as in the old label
  • Confidence: medium (under the cobweb night, which already disturbs the curves)

2026-10-09T06:30Z → 2026-10-10T08:15Z (added)

  • Phenomenon of the catalogue: The camera moves through a thermal effect
  • Precedents: LENZSPITZE-93 (weak cycle under an overcast sky, temperature staying up until the evening)
  • Family: Thermal displacement (labeling-rules §1: always labelled, even overcast; the Visibility labels overlap it)
  • Measured: temperature 25.0 (06:00) → 29–30 (15:00–00:00) → 27.2 (06:00); at the measurements where CTR04 is kept (NOREF=1 scripts/cycle.py): pan 31→16 mdeg, tilt −141→−127, scale +1.9e-4; 10/10 05:00 pan 23, tilt −134 (partial return); PR18 cx 0.30→0.26 (15:00–17:45), 0.25–0.29 at night; PR16 0.19→0.15–0.17. Chained with NORDEND-171 (ends 06:30) and NORDEND-175 (starts 08:15)
  • Images: those of NORDEND-173 (bad weather)
  • Confidence: medium (the day is mostly under Visibility labels; the start is hidden by the CTR04 losses)

2026-10-09T07:15Z → 15:15Z (kept, NORDEND-173), 2026-10-09T17:00Z → 2026-10-10T08:15Z (kept, NORDEND-174)

  • error_value of 19 sensors at 07:30 (start 07:15), CTR01 back 15:15; 15:15–17:00 clean; PR07 cx 0.035 at 17:15 (first touched), CTR04 rejected 21:15–23:15, CTR01 back 08:15. Confidence high.

2026-10-10T08:15Z → 2026-10-11T06:45Z (corrected, NORDEND-175)

  • Measured: temperature 29.2→33.1 (11:00)→25.1; pan 22→8.4 mdeg (13:00)→28.3 (11/10 05:00); tilt −132→−124 (09:00–12:00)→−139 at night; rotation +0.012°; PR18 cx 0.29→0.25–0.26 (09:00–13:00)→0.30–0.31 at night. Start kept: the pan leaves at 07:45, the tilt for good after 08:15. Confidence high.

2026-10-11T06:45Z → 2026-10-12T06:45Z (corrected, NORDEND-177)

  • Measured: temperature 25.1→31.6 (15:00); pan −16 mdeg, tilt +18, scale +2.1e-4; corr_x PR18 −0.041, PR17 −0.023, PR16 −0.022, PR14 −0.021; corr_y PR17 +0.040 (16:15), PR01 +0.034, PR08 −0.031 (12:45), PR16 −0.027, PR12 +0.026; PR08 against the others ≈ −0.03 from 08:00 to 17:00 (work/d1011/…residuals-corr_y-others…png), the same daytime dip on 10/10 and 12/10 (scripts/matrix.py PR08:corr_y). Confidence high.

2026-10-12T06:45Z → 2026-10-12T19:30Z (corrected, NORDEND-179)

  • Measured: temperature 25.4→37.5 (12:00); pan −29 mdeg (14:45), tilt +29, scale +4.1e-4; PR18 cx −0.076 (11:30), PR16 −0.038; PR08 cy −0.042 (16:15); pan back to 17.8 at 19:00 (morning 23–28), then the step at 19:45; temperature 29.6 (19:00)→31.1 (22:00), ≈ 31 all night. labeling-rules §2: a jump at night stops the cycle at the measurement before it. Confidence medium.

2026-10-12T19:30Z → 2026-10-12T20:00Z (added)

  • Phenomenon of the catalogue: A step of the camera that the compensation does not erase
  • Precedents: none close (FLORIDA-74/77/78 are one-measurement jumps of a reference)
  • Family: Displacement (a step that stays; the temperature does not explain it: +10 mdeg of tilt in one step ≈ 5 °C at the batch slope of 0.002°/°C, the temperature rose 0.6 °C)
  • Measured: 19:30→19:45: tilt −133.0→−123.6 mdeg, pan 17.2→15.3, sigma0 18.3→13.5 (×1e-4); overview.py --rigid 2026-10-12T19:30Z: ty +0.44 px for all, residual CTR03 dy +0.04, CTR04 dy +0.037; corr_y PR01 0.179→0.144, PR02 0.186→0.144, PR14 0.053→0.024, PR16 0.006→−0.023, CTR02 0.102→0.075, CTR03 −0.087→−0.063; PR18 cx 0.284→0.260; error_value CTR02/CTR03 +0.01, CTR04 −0.011, PR14/PR16 +0.02. Tilt −117…−125 mdeg until 13/10 07:45: no return
  • Images: zone 19:15–20:00, no obstacle; figures/camera_step_1210.png
  • Confidence: medium (cause unknown)

2026-10-12T22:30Z → 2026-10-13T00:15Z (added)

  • Phenomenon of the catalogue: none exactly; a camera step with its return
  • Precedents: none useful
  • Family: Displacement (a jump then the return without an established cause)
  • Measured: overview.py --rigid 2026-10-12T22:30Z: ty −0.07, −0.22, −0.56, −0.12, −0.06 px at 22:45–23:45; CTR03 dy −0.05, CTR04 −0.03 against the targets; tilt −122.6→−128.7→−137.2 (23:15)→−125.2; pan 19.2 at 23:15; corr_y PR01 0.138→0.179 (23:15)→0.147 (00:15), PR02 0.145→0.190, PR10 0.092→0.134, PR14 0.017→0.044; temperature 31.1 (22:00), 29.8 (23:00), 30.4 (00:00); PR04 rejected 22:45–23:00 (Target issue), PR17 step at 23:00 (NORDEND-169)
  • Images: whole image 23:15: night, window marks, nothing in front
  • Confidence: medium

2. What I left without a label

Removed at the requalification (2026-11-17)

  • NORDEND-154, 2026-10-01T10:00Z → 10:30Z, Visibility: missing point only. CTR01 is rejected for one measurement; no target leaves its level beyond 5 noises on corr_x or corr_y, neither against the level of the two hours around (--outside) nor against the local level (--rolling 6h): no range at all.
  • NORDEND-295, 2026-10-02T15:30Z → 16:15Z, Visibility (sun behind CTR01, CTR02, CTR03): mostly missing points (CTR01 rejected 15:45–16:00). Under the prescribed method for a moment inside a thermal day (--rolling 6h), no target beyond 5 noises in the range. Against the level of the hours around, single measurements reach 8–9 noises (PR11 corr_x −9.4 and PR08 corr_x −8.7 at 15:45, PR16 corr_y +7.8), but the same window shows deviations of the same size outside the label (PR08 corr_x +8.8 at 14:15): the afternoon agitation of the thermal day, not standing out from it.
  • NORDEND-301, 2026-10-08T14:45Z → 15:15Z, Visibility (sun behind the references): CTR01 rejected for one measurement. Against the level of the hours around: no range. With --rolling 6h: a single measurement of a single target, PR18 corr_y −5.9 at 15:00, in a window where such single-measurement departures occur about twenty times in twelve hours (PR02, PR03, PR07, PR14, PR18…): ordinary noise of the day.

Decision rule used for the moments of a sunny afternoon (under a thermal cycle): kept when at least one target goes beyond 5 noises both against the level of the hours around (--outside the label, ±2 h) and against the local level (--rolling 6h); 295 and 301 pass only one of the two, by a single measurement.

From the previous verification

  • 2026-09-30T10:30Z and 2026-10-10T11:45Z: the error_value of 6–7 sensors of the right part (CTR04, PR06, PR08, PR15, PR16, PR18) rises for one measurement; corr moves by 0.01–0.03, within the agitation of the thermal day; under the thermal labels.
  • The afternoon agitation of the pose on sunny days outside the sun-on-references windows (common corr_y jumps of ±0.007–0.010 between 13:30 and 16:30): part of the thermal cycle; the thermal descriptions do not mention it (section 3).
  • PR08 corr_x ≈ +0.03 on the nights of 09/10 and 10/10: PR08's night levels vary as much on other nights (06/10 +0.021, 02/10 −0.016); not separated from the common night state.
  • After the step of 12/10 19:45 the corr_y of most targets stay ≈ 0.02–0.03 lower until the end of the batch: stated in that label.
  • PR04 (spanning NORDEND-24/152) and PR17 (spanning NORDEND-169): left to their spanning labels.

Spanning labels, within the range

  • NORDEND-24 (PR04, until 01/10 00:00): holds (rejections 12:15 and 14:15 on 30/09, "rare on 30/09").
  • NORDEND-152 (PR04, from 01/10): holds (rejections on 01, 02, 05, 07–12/10, none on 03, 04, 06/10; noise 0.0145 against ≈ 0.005, "twice as noisy" is understated).
  • NORDEND-169 (PR17, from 08/10 02:15): step at 02:30 confirmed (corr_y 0.113→−0.059); the "partial return on the evening of the 12th" is at 23:00 (cx 0.165→0.109, cy −0.052→+0.016). It falls inside the camera back-and-forth of 22:30–00:15 (added), which it does not mention; not wrong within the range.

Requalification

Script scripts/requalify.py (runs tools/scripts/ranges.py data/NORDEND --columns corr_x,corr_y --k 5 per label; outputs in work/requal/<id>_<mode>.txt, summary work/requal/summary.txt), then scripts/requalify_build.py (writes labels.jsonl).

  • Movement labels (Visibility, Obstruction, Displacement): --start/--end = label ±2 h, --outside <label> (mode short), and label ±6 h with --rolling 6h (mode rolling); the ranges overlapping the label are kept, the largest deviation of a target (references excluded) is reported.
  • Thermal labels: --start/--end = the label (mode thermal: deviation from the median of the cycle, which understates the amplitude against the night level), and --rolling 6h.
  • Extra checks: NORDEND-309 against the level before the step (--outside 2026-10-12T19:45Z/2026-10-12T22:15Z, work/requal/NORDEND-309_before.txt); NORDEND-302, PR18 corr_x against the night of 08→09/10 (--outside 2026-10-09T06:30Z/2026-10-09T17:00Z, work/requal/NORDEND-302_night.txt).

Largest deviation of a target in the range, in its own noises (short or thermal / rolling):

id family measured decision
149 Thermal PR18 cx −10.1, PR02 cy −10.3 / PR13 cy −11.1 kept
291 Visibility PR07 cx +17.7, PR16 cx +7.3 / PR16 cx +11.9 rewritten (description on the PR07, PR16 jumps)
292 Visibility PR07 cx +12.1 at 03:15, PR16 cy −8.2 at 03:15, PR07 cx +24.9 at 07:15 / +25.7 rewritten (description on the PR07, PR16 jumps)
293 Visibility PR16 cy +12.6, PR16 cx +11.7, PR06 cx +11.8 / PR16 cx +13.0 kept
154 Visibility none / none removed
294 Thermal PR15 cx −9.2, PR18 cx −8.8 / PR02 cy +7.5 kept
156 Thermal PR18 cx −12.0 / −7.1 kept
295 Visibility PR11 cx −9.4 (single) / none removed
158 Thermal PR18 cx −20.3 / −14.3 kept
296 Visibility PR13 cx +7.0 / PR10 cx +6.7, PR12 cy +6.7, PR15 cx −6.3 rewritten (effect first)
160 Thermal PR18 cx −19.7 / −13.3 kept
297 Visibility PR18 cx +8.3, PR01 cx +8.0 / PR10 cx +6.9, PR01 cy +5.9 rewritten
162 Thermal PR18 cx −14.0 / −8.5 kept
298 Visibility PR05 cy +8.7, PR02 cy −6.6, PR16 cx +6.3 / PR02 cy −5.7, PR09 cx −5.6 rewritten
164 Thermal PR18 cx −13.2 / −9.2 kept
299 Visibility PR13 cx +8.9, PR08 cx +6.7 / PR12 cx +5.6 rewritten
166 Thermal PR18 cx −9.0 (07/10 09:30, outside the cobweb) / cobweb up to −52 kept
167 Visibility PR15 cx −77 / −55 kept
168 Obstruction PR10 cx −23.7 / −20.2 kept
300 Visibility PR03 cy −78 / −42 kept
171 Thermal PR18 cx −14.3 / PR07 cx +8.9 kept
301 Visibility none / PR18 cy −5.9 (single) removed
302 Thermal PR18 cx −6.4 at 14:30–16:45 against the night before (clean measurements) kept
173 Visibility PR06 cx −136 / −57 kept
174 Visibility PR06 cx −121 / PR16 cx −63 kept
303 Thermal PR08 cy −16.3, PR18 cx −9.0 / PR16 cx +14.2 kept
304 Visibility PR15 cy +27.3, PR16 cx +17.5, PR18 cy −15.4 / PR16 cx +14.1 rewritten (targets named on the measurement)
305 Thermal PR08 cy −21.2, PR01 cy +10.0 / PR08 cy +11.0 kept
306 Visibility PR07 cy +7.2, PR01 cy +6.9 / PR01 cy +8.4 rewritten
307 Thermal PR01 cy +7.4, PR18 cx −6.8 / PR01 cy +9.0 kept
308 Visibility PR01 cx +11.2, PR01 cy +10.0 / PR01 cy +9.0 rewritten
309 Displacement level before: PR15 cy −15.0, PR01 cy −8.8 kept
310 Displacement PR03 cy +11.4, PR01 cy +10.6 / PR02 cy +7.3 kept

Rewritten labels (none had a review field): their bounds and family are unchanged; the descriptions now lead with the effect on the corrected curves and no longer conclude on a missing curve (the rejections of CTR01, PR09, PR07 are kept as context only); the targets named are those measured above the threshold. Remark: the bounds of 291 (to 23:45) and 292 (to 08:00) were set on the rejections of CTR01 and PR09; under the new rules the curves move only until about 21:45 and at 03:15 and 07:00–07:15; the bounds were left as they are, as the request asks.

python3 labeling/scripts/validate.py labels.jsonl: 30 labels, 0 errors, 2 overlaps of distinct Visibility events (291/292 inside 292/293 at daybreak of 01/10; 167/300 cobweb night and CTR04 at daybreak).

3. The documentation

  • Helped: families.md (sun behind a spot = Visibility; thermal boundaries; Displacement for a camera step); labeling-rules.md §2 (bounds on the curves, not on the rejections: decisive for NORDEND-170 and the sunny labels; "a jump at night stops the cycle": NORDEND-179) and §3 (two nuisance episodes with targets still shifted make one: 150 + 151); reading-rules.md §5 (raw positions with --rigid before saying "the camera": daybreak of 01/10, steps of 12/10); stations.md (rock face, CTR01 local reference). The precedents JUNGFRAU-50/55/89 gave the wording for a reference with the sun behind it.
  • Misled or contradicted: the existing labels call PR07 a "left-edge" target; in the image it sits in the middle-right. The sunny-afternoon labels described only CTR01; the images show the sun behind the references CTR02 and CTR03 on every sunny afternoon, which makes all the targets jump together at some measurements. stations.md could say for NORDEND: "CTR02 and CTR03 have a rock face behind them lit by the afternoon sun".
  • The thermal labels of the sunny days do not mention the afternoon agitation of the pose (labeling-rules §5 asks for it); I kept them, the omission being general and the agitation small.
  • quality.py cannot be used (export quality empty): the "spot fades" statements rest on the thumbnails.
  • No weather for NORDEND, though the instructions ask for it.

Difficulties

  • days.py --bounds with a fixed night was of little use: the pan rises all night every night; I read the turning points on my own 15 min matrices (scripts/matrix.py).
  • 150 and 151 had been validated separately by the user; their merge loses the validation of both.

Sources read

  • Requalification: instructions/analyst.md, labeling/labeling-rules.md §1–2, work/requalify_input.jsonl, this analysis; tools/scripts/ranges.py (help), tools/scripts/inventory.py (distances, mm/px)
  • 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
  • data/NORDEND/ through inventory.py, overview.py (batch; 10/10–12/10 with --others; --rigid on 12/10 twice), ranges.py --rolling 4 --instants 3, days.py --bounds, metrics.py, plot.py, thumbnails.py; whole images 2026-10-01T07:00Z, 2026-10-07T19:45Z, 2026-10-07T21:00Z, 2026-10-12T23:15Z; my scripts in scripts/ (exports read through tools/scripts/data.py)
  • precedents (labeling/scripts/precedents.py): LENZSPITZE-83, JUNGFRAU-50/55/89, FLORIDA-74/77/78, LENZSPITZE-93/114