# Labels: BREITHORN, 2026-10-22T16:00Z → 2026-11-01T00:00Z (UTC) Task: verify the 12 existing labels of the range (`work/existing.jsonl`), plus the spanning BREITHORN-13. Verdicts, one line per old id: | Old id | Verdict | Evidence | |---|---|---| | BREITHORN-16 | **corrected** (description) | The contrast-stretched thumbnails of 22T20 and 23T04 (`work/th_f`, `work/th_b`) show the dark outline of a vehicle, lights off, in front of the bottom left of the scene. The old text says "no object is visible" and "probable cause, not established". It also left out COLI22 in corr_x (−6 to −5 hundredths px against its level) and POUI22 (corr_x +3 to +4, corr_y +6 at 22T20). The bounds hold. | | BREITHORN-17 | kept | The housing temperature is at its trough on 24T00 (25.9 °C, 4 h median). Scale is low from 23T20 to 24T16. At 23T16–24T16, corr_y of REFPOU22 is up to +5.9, COLH20 up to +5.4, REFG44N down to −2.5. Almost everything is back at 24T20, fully at 25T00. The end is borderline but stays within one step. | | BREITHORN-18 | kept | A car is seen at 23T20 (one headlight) and 24T08 (two headlights). The data match the text: COLI22 corr_x −83, POUI22 corr_x +27/+23; COLI22, POUI22 and POUI24 are rejected at 24T04–12. The car is gone at 24T16. | | BREITHORN-19 | kept | The three-light vehicle with oblique pillars is seen at 24T16, 25T08 and 25T20. COLI24 and POUI22 are rejected until 25T20. At 26T00 another obstacle is there (label 20). | | BREITHORN-20 | kept | The oblique structure is seen at 26T00 and 26T04. REFH22, REFPOU22, COLH20''', COLI22 and POUI24 are rejected. All targets shift together in corr_x (−3 to −5). POUI22 corr_y is −57/−31. | | BREITHORN-21 | kept | Single cold measurement at 26T08 (26.1 °C, rotation at its extreme). corr_y: COLH20 +4.4 above its level, REFPOU22 +2.4, REFG44N −1.5. Back at 26T12, where only the scale is still relaxing. | | BREITHORN-22 | **corrected** (end 27T00 → 26T20, description) | The car with headlights is on the 26T16 image only. At 26T20 it is gone and the oblique structure is there (`work/th_g`). The 26T20 rejections (REFH22, COLI24, POUI22) and the POUI24 jump belong to the structure. | | BREITHORN-23 | **corrected** (start 26T20 → 26T16, description) | Its own text puts POUI22 rejected from 26T20 (rejection run 26T20 → 27T04), and the structure is on the 26T20 image. So 26T20 is touched and the start is the measurement before it. The description now names the 20:00 losses. | | BREITHORN-24 | kept | Cold at 28T04 (25.25 °C). corr_y at 28T04/08: REFPOU22, COLH20 and COLI24 up 5 to 6; REFG44N, REFI'21 and POUI24 down. Views clear on 28T04. There are small corr_x moves (COLH20''' −4 noises), but none is larger than those named. | | BREITHORN-25 | **corrected** (description) | Bounds right. The old text left out COLI24, the second-largest mover in corr_y (+7 at 29T00, flagged +5 noises), the corr_x jumps of POUI24 (+8) and POUH21 (+5) at 29T08 (spots intact on the fixed-frame thumbnails), and COLI22 in corr_x (−6 to −7). | | BREITHORN-26 | kept | Car with headlights at 29T20 and 30T00. COLI24 is rejected; POUI22 corr_x is +52/+48; COLH20''' corr_x is +17, then corr_y −23. Clear at 30T04. COLI22 (−11 in x) is called "a little" next to POUI22: acceptable. | | BREITHORN-27 | kept | Cold from 29T20 to 30T08 (trough 24.25 °C at 30T04). At 30T04: COLH20 +8 noises, COLI24 +7, REFPOU22 +5, POUI24 −4. Back at 30T12. | ## 1. My labels ### 2026-10-22T16:00Z → 2026-10-23T08:00Z (BREITHORN-16, corrected) - **Phenomenon of the catalogue**: A vehicle or machine that stays in front of the targets; An obstacle covers part of the spot (COLI22). - **Precedents**: BREITHORN-15 and BREITHORN-4 show the same parking spot and the same targets, with the vehicle seen. WILDSTRUBEL-37 (a parked car, seen) is reviewed. They tell me to state the cause as seen. - **Family**: Obstruction. A few targets are hidden by an object while the rest is intact, so not Visibility. - **Measured**: COLI24 is rejected at 22T20, 23T00 and 23T04 (error_value 0.98–0.99). COLI22 error_value is 0.45–0.46 (level 0.13), still accepted. COLI22 corr_y is −11.7, −14.4, −14.8 and corr_x −6.3, −5.0, −5.0 (hundredths px against the 20–21 Oct level). POUI22 corr_y is +4.75 at 22T20 (level about −1.5) and corr_x +3.3, +4.7, +2.9. The references are flat. All of it is back at 23T08. - **Images**: Thumbnails at 22T16, 22T20, 23T00, 23T04, 23T08 and 29T20, contrast-stretched (15–60). A dark rounded outline with a horizontal edge sits in the bottom left at 22T20 and 23T04, at the place of the car of 29T20. COLI24 is hidden and COLI22 reduced. Nothing is there at 22T16 or 23T08. - **Confidence**: high for the bounds. Medium-high that the object is a vehicle: it is lights-off and faint, but it has the shape and place of the other parkings. ### 2026-10-26T12:00Z → 2026-10-26T20:00Z (BREITHORN-22, corrected) - **Phenomenon of the catalogue**: A vehicle or machine that stays in front of the targets; A reference dropped. - **Precedents**: BREITHORN-7 (a car, REFH22 rejected, compensation on three references). - **Family**: Obstruction. - **Measured** at 26T16: - REFH22 rejected (0.98); compensation on three references. - COLI24 and POUI24 rejected. - POUI22 accepted at 0.62, with corr_y +124 and corr_x +10. - COLI22 corr_x +5 against its level, COLH20 corr_y −5. - **Images**: At 26T16, a car with its headlights on in front of the bottom of the scene. At 26T20 there are no headlights and no car body; the oblique structure is there, with two small lights (`work/th_d`, `work/th_g`). - **Confidence**: medium-high. The car's departure between 16:00 and 20:00 is read on one image. ### 2026-10-26T16:00Z → 2026-10-27T08:00Z (BREITHORN-23, corrected) - **Phenomenon of the catalogue**: A vehicle or machine that stays in front of the targets; A reference dropped. - **Precedents**: BREITHORN-20 (the same structure on the night before). - **Family**: Obstruction. - **Measured**: - 26T20: REFH22, COLI24 and POUI22 rejected; POUI24 accepted at 0.66, with corr_x +22 and corr_y −4. - 27T00 and 27T04: REFPOU22, COLH20''' and POUI22 (0.75) rejected; REFG44N and REFI'21 corr_x −1.5 to −1.8 (level about +0.4); targets −2 to −3 at most. - 27T08: everything is back. - **Images**: The oblique structure is seen at 26T20, 27T00 and 27T04; the view is clear at 27T08. - **Confidence**: high. ### 2026-10-28T16:00Z → 2026-10-29T16:00Z (BREITHORN-25, corrected) - **Phenomenon of the catalogue**: The camera drifts away when the temperature of its housing drops, without a daily cycle. - **Precedents**: BREITHORN-14, -24 and -27 (same station, reviewed). Their wording is kept. - **Family**: Thermal displacement. The episode follows the housing temperature and comes back with it, and nothing is seen on the images, so not Visibility. The error_value rises only a little (REFH22 0.19–0.22, level 0.11; COLH20 0.15; COLH20''' 0.13–0.15) and there is no rejection. - **Measured**: - Temperature: 26.9 → 24.1 → 23.1 °C (29T04), then 26.4 at 29T12. - Pose: scale 0.99976 → 0.99965; rotation −0.0367 → −0.0423; tilt +0.0019. - corr_y at 29T00: REFPOU22 +10, COLH20 +9, COLI24 +7, POUI22 +8.7, POUI24 −8, REFG44N −4, REFI'21 −3, REFH22 −2.7. - corr_x: COLI22 −6 to −7; at 29T08, POUI24 +8 and POUH21 +4.5. - Back at 29T16. - **Images**: The view is clear at 29T04. At 28T16, 29T08 and 29T16 the fixed-frame thumbnails of POUI24, POUH21, COLI24 and COLH20 show the spots intact and in place. - **Confidence**: high. ### Kept labels BREITHORN-17, -18, -19, -20, -21, -24, -26 and -27 are copied unchanged, with their review field. Evidence is in the table above. ## 2. What I left without a label - 23T08–23T12: COLJ19, COLI20 and COLI21 corr_y rise by 3 to 5 together. This is the faster phase of the drift of BREITHORN-13 (tilt rising), not a separate event. - 30T20: COLJ19 corr_y +3.5 for a single point. COLJ19's corr_y swings ±2 to 3 on calm days (31 Oct – 2 Nov). Within its noise; no other sensor moves. - 31T04–31T08: scale is slightly low and COLH20 corr_y is +1. Under 5 noises, with flat temperature (26.6–26.75 °C). No cycle. - 30T12 → 11-01T00: calm. A rolling-level search (`ranges.py --rolling 3 --instants 2 --k 4`) finds no instant with two sensors off level. - Every rejected measurement of the range falls under a label: COLI24, COLI22, POUI22, POUI24, REFH22, REFPOU22 and COLH20''' (checked against the inventory runs). ## 3. The documentation - `families.md` boundaries "Obstruction, whatever the duration" and "Thermal or Visibility" settled the families. `labeling-rules.md` §2 (start = measurement before the first touched point) gave the BREITHORN-23 correction: its own description put POUI22 rejected at 26T20, before its start. - `labeling-rules.md` §5 "Nothing on the images is written only after looking": BREITHORN-16's "no object visible" did not survive a contrast-stretched thumbnail. The plain thumbnails are too dark here. On this station, `--contrast 15,60` on the bottom-left zone is needed to see a vehicle without lights. - Precedents: BREITHORN-15 and BREITHORN-4 (same parking, vehicle seen) supported the correction of 16. - **Spanning label BREITHORN-13**: holds within the range for dates and direction: tilt rising fast until 25 Oct, plateau 25–28, then a slower rise. Its list of drifting targets is incomplete. Between 20–21 Oct and 31 Oct – 1 Nov, these also drift: - COLI22: corr_x from about 0–1 to 3–4.5, corr_y from about 0.5 to 3–4; - COLH20''': corr_y from about 0 to 2.5–4; - POUI22: corr_y down from about −1 to −3 / −5, like COLI24 next to it; - POUI24: corr_x from about 0 to 2–3; - COLI24: corr_x from about 0 to 2–3, besides its corr_y. For POUI22 and COLI24, the change in corr_y falls under the parkings of 23–26 Oct, so it cannot be dated as a drift or a step from these data alone. - No weather: BREITHORN has no place in `stations.txt` (indoors, in a factory). No MeteoSwiss station was fetched. - No image-quality columns: the export's peak–snr columns are empty, so `quality.py` does not apply. ## Difficulties - `tools/GUIDE.md` (one tool command per call, no other script) conflicts with the analyst brief (own scripts allowed in `scripts/`). I followed the brief: `scripts/dump.py` prints per-instant tables of corr_x, corr_y and error_value. - Validate reports 3 overlaps between Obstruction labels. They are intended: successive obstacles share a boundary measurement (18/19 at 24T12–24T16, 19/20 at 25T20–26T00, 22/23 at 26T16–26T20). ## 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/BREITHORN.jsonl` (ids 4, 7, 13–27), `work/existing.jsonl`, `work/spanning.jsonl`, `~/aimsight-data/extraction/stations.txt`, precedents (`precedents.py`, parked vehicle). Data: `data/BREITHORN/displacement` and `camera` (19 Oct – 2 Nov), `metrics` (`metrics.py`), images and thumbnails (`work/th_a` … `work/th_h`), whole images 22T16 and 22T20. Tools: `inventory.py`, `metrics.py`, `ranges.py --rolling 3 --instants 2`, `thumbnails.py`; own script `scripts/dump.py` (outputs `work/corr_x.txt`, `work/corr_y.txt`).