Files
2026-11-18 10:51:31 +01:00

3.9 KiB
Raw Permalink Blame History

CLARIDEN, October: Displacement labels requalified (significant only)

Request of the user (2026-11-18): too many small displacements, most going back the other way later; keep only the significant ones, on all CLARIDEN series.

Method

  • scripts/disp_size.py (work/disp_size.txt): per label and per target, the extreme of the 1 h rolling median of corr_x_mm / corr_y_mm from the level of the 6 h before, in mm and in own noises (median step), the net change at the end, and the month spread (P5–P95) of each target. PR-284 and PR_BG are discarded as noisy, with a large mm/px.
  • scripts/short_native.py (work/short_native.txt): the native-step extreme of the short labels.
  • scripts/month_plot.py (work/month_displacements.png): hourly medians of the targets concerned over October, with the labels, the camera tilt and the housing temperature.
  • Criterion: kept if a clean target moves by ≥ about 0.4 mm and ≥ 10 noises on its 1 h median, and keeps that level for days. Removed if the event is a wave that comes back, a remainder of a camera jump of one measurement, or ≤ about 8 noises.

Per label

id decision evidence
CLARIDEN-14 kept, corrected (pointer to 72 replaced) D-301-C x +1.1 mm (19 n), D-303-B y +0.36 mm (12 n), kept afterwards
CLARIDEN-23 kept, corrected (pointer to 32 replaced) D-304-C y +0.48 mm (14 n), D-303-C y +0.42 mm (11 n), D-301-C x about +0.5 mm, kept
CLARIDEN-3 removed D-303-B −0.25 mm, D-306-B −0.11 mm (13 n, but 0.1 mm), back with 44; D-302-B +0.7 mm is its own two-level jump
CLARIDEN-44 removed return of 3; D-301-C −0.84 mm is a one-hour spike (net −0.07 mm)
CLARIDEN-46 removed P-301-B y +0.39 mm, 6.9 n native, back the same day
CLARIDEN-73 removed D-303-B y −0.19 mm, 7.3 n native
CLARIDEN-72 removed partial return of 14: D-301-C −0.28 mm (5 n); its big D-302-B value is that target's own steps of 11/10 and 14/10
CLARIDEN-50 removed one measurement, D-301-A y −0.43 mm, 5.9 n
CLARIDEN-53 removed ≤ 5.2 n on the 1 h median, ≤ 0.2 mm
CLARIDEN-54 removed one measurement, D-302_A x +0.37 mm, 5.3 n
CLARIDEN-59 removed D-306-B y +0.05 mm, 6.4 n
CLARIDEN-62 removed D-303-B +0.17 mm (5.8 n), back with 29
CLARIDEN-29 removed D-306-B −0.11 mm, D-304-B −0.2 mm; wave with 62/30
CLARIDEN-30 removed D-306-B +0.05 mm (5.4 n)
CLARIDEN-31 removed D-306-B +0.13 mm, D-304-B +0.26 mm, D-304-A +0.38 mm (6 n), back with 63
CLARIDEN-32 removed partial return of 23, D-304-C −0.25 mm (7 n)
CLARIDEN-63 removed same dates and targets as 32 (−0.26 mm, 7 n)
CLARIDEN-36 removed D-306-B −0.06 mm, 7.8 n
CLARIDEN-1 kept as is P-300-B rejected at 4070/4070 instants

The labels removed or corrected are in the labels.jsonl of the analysis that owns them: CLARIDEN_2026-10-01T0000_2026-10-11T0000 (12 → 8), CLARIDEN_2026-10-11T0000_2026-10-20T0400 (11 → 6), CLARIDEN_2026-10-20T0400_2026-11-01T0010 (21 → 14, 23 corrected), and this one (14 corrected). labeling/auto/CLARIDEN.jsonl follows them since 2026-11-18 (the three period analyses submitted again: 46 → 30 labels, 2 Displacement; CLARIDEN-23 keeps its id; then CLARIDEN-9 restored: 31 labels, 3 Displacement).

Not labeled, for the record

  • D-302-B / PR_BG: two-level jumps (about 0.75 mm) on 1/10, 5/10, 11/10, 14/10, specific to that prism.
  • Across the 3→5/10 outage: D-304-C about −1.5 mm, D-303-C about −0.6 mm, with a change of camera tilt: CLARIDEN-9 (analysis CLARIDEN_2026-10-01T0000_2026-10-11T0000), deleted before this pass, restored on 2026-11-18 at the user's request, unchanged.

Sources read

Previous turn (docs, inventory, ranges); this turn: labeling/auto/CLARIDEN.jsonl, the labels.jsonl of the four CLARIDEN analyses, app/server/analysis.py (how labels are submitted), tools/scripts/data.py (loader), data/CLARIDEN through the scripts above.