Files
aimsight-labeling/auto/analyses/FLORIDA_2026-10-22T0315_2026-11-01T0000/analysis.md
T

5.8 KiB
Raw Blame History

Labels: FLORIDA, 2026-10-22T03:15Z → 2026-11-01T00:00Z (UTC)

Verification of the 19 existing labels (all reviewed 2026-11-17). Weather: no weather/ folder for FLORIDA, so no weather is available. Analysis cut short by the usage limit: the image check of the new label was not done.

Verdict per old id

  • FLORIDA-46 (22 03:15→03:45, Displacement): kept. Sellette 4 corr_x steps up at 03:30 and stays (day_22.png, table.txt). Neighbours are flat.
  • FLORIDA-47 (22 08:45→17:00, Thermal): kept. Pan drifts from about 08:45 to about 16:15 while the housing temperature goes from 16 to 20.6. In days.py, pose_pan reaches −9 noises at 14:15. Returned by 17:00.
  • FLORIDA-48 (22 11:15→13:30, Visibility): kept. r1_r is rejected at 11:30 and from 12:30 to 13:15 (inventory). Pont Ernen and Sellette 5 corr_x jump, then return at 13:30.
  • FLORIDA-49 (23 03:45→09:15, Visibility): kept. r1_r is rejected at 04:00, 05:45–06:00 and 06:30–09:00. Pont Ernen jumps from 03:45.
  • FLORIDA-50 (23 12:15→20:15, Visibility): kept. p1_r error_value rises from 12:15. The pose jumps (pan −0.099 and tilt −0.057 at 13:30). There are many rejections until 18:45 and the corr return around 20:15.
  • FLORIDA-51 (23 22:00→24 01:45, Visibility): kept. The tilt dips and Sellette 4 corr_y deviates. The curves return at 01:45.
  • FLORIDA-52 (24 01:45→02:15, Visibility): kept. r1_r and Pont Ernen are rejected only at 02:00.
  • FLORIDA-53 (24 07:00→14:45, Visibility): kept. r1_r is rejected from 07:15 to 14:30 with no pose.
  • FLORIDA-54 (24 21:45→25 21:15, Visibility): kept. Sellette 5 is rejected from 22:00. The last rejection is Sellette 5 at 19:45. Sellette 4 corr_x recovers from −0.25 at 19:45 to −0.05 at 21:15. Sellette 5 corr_y is still about +0.1 until 21:45, but the evening of the 26th has the same level at the same housing temperature (about 19 °C), so this is not a rain remainder. The bound is acceptable.
  • FLORIDA-55 (26 08:30→12:00, Visibility): kept. r1_r is rejected from 08:45 to 11:45.
  • FLORIDA-56 (26 11:45→19:00, Thermal): kept. Rotation and pan move with the housing temperature (16 → 21). Sellette 5 corr_y stays about +0.1 into the evening while the housing temperature stays about 19: the level matches the temperature, so this is not a separate event.
  • FLORIDA-57 (26 14:30→15:15, No data): kept. Two steps (14:45 and 15:00) are absent for all sensors.
  • FLORIDA-58 (27 01:45→11:45, Visibility): kept. Pose skewed at night (tilt and scale steps at 02:00–03:00 while the temperature is flat); r1_r rejected from 06:15 to 11:30.
  • FLORIDA-59 (27 09:00→09:30, No data): kept. The 09:15 step is absent.
  • FLORIDA-60 (27 11:30→20:00, Thermal): kept. Pan reaches −6.3 noises at 12:00 and the temperature peaks at 25.5. Sellette 5 corr_x goes down in the afternoon.
  • FLORIDA-61 (28 07:00→10:30, Visibility): kept. r1_r is rejected at 07:15 and from 08:15 to 10:15.
  • FLORIDA-62, 63, 64 (29, 30, 31, Thermal): kept. In days.py, Sellette 5 corr_y reaches +7.3, +6.7 and +7.0 noises and rotation −8.0, −6.9 and −7.4 noises. The bounds match the turning points.

1. My labels

2026-10-28T10:15Z → 2026-10-28T19:45Z (added)

  • Phenomenon of the catalogue: The camera moves through a thermal effect
  • Precedents: FLORIDA-56 and FLORIDA-60 (same station; the thermal label starts at the last rejection of the morning r1_r loss) and FLORIDA-62 to 64 (same shape); precedents.py not run (usage limit).
  • Family: Thermal displacement. The pose and Sellette 5 corr_y follow the housing temperature (19.7 → 24.3) and come back with it. error_value stays at its level.
  • Measured: in table.txt, rotation goes from about +0.01 in the night to −0.014 at 12:00–14:00, then is back to +0.005 at 19:30. Tilt goes from +0.006 to +0.017. Pan is flat around −0.04. Sellette 5 corr_y goes from about +0.045 at night to +0.12–0.14 from 10:30 to 15:15, then is back to +0.07 at 19:45. In days.py on Sellette 5, corr_y reaches +5.0 noises at 13:00, which is at the threshold of the labeling rule. The start is hidden under FLORIDA-61: rotation is already −0.008 at the first pose, at 10:30.
  • Images: none (not looked at, usage limit).
  • Confidence: medium. The cycle is clear on the hourly pose, but small and at the 5-noise threshold.

2. What I left without a label

  • 23rd, 09:15–12:15. Rotation goes to −0.02 with the temperature rising to 20.5, then the mist starts. This is −3.1 noises in days.py, and Sellette 5 corr_y does not move: under the 5-noise rule, so no thermal label.
  • 24th. No cycle above 5 noises, and the pose is mostly absent.
  • 25th. The thermal cycle is masked by the rain (FLORIDA-54); after it, the pose relaxes along with p1_r drying, and the two cannot be separated.
  • Night of the 25th → 26th. Sellette 4 and 5 are near the calm baseline (table.txt). A first figure with a per-day median suggested a shift; this was an artefact of that median.
  • Sellette 6 and Massif 2. Covered by the spanning Target issue labels (FLORIDA-1, FLORIDA-14), which hold within the range: rejections throughout.

3. The documentation

  • The 5-noise rule decided the 23rd and the 28th.
  • Per-day medians mislead on a station whose night level shifts; a fixed calm baseline was needed (my own scripts/table.py).
  • FLORIDA has no weather data.

Difficulties

  • Usage limit reached: no images looked at for the added label, no precedents query, no GIF.

Sources read

instructions/analyst.md, tools/GUIDE.md, tools/DATA.md, labeling/families.md, labeling/labeling-rules.md, labeling/reading-rules.md, labeling/stations.md, labeling/auto/FLORIDA.jsonl (ids and bounds), work/existing.jsonl, work/spanning.jsonl. Tools: inventory.py, overview.py, days.py --bounds; scripts/daily.py, table.py, s5.py.