Name yourself to give a verdict. Every verdict carries your name and the time, and a verdict without them is refused when it is applied.
from ingest/candidate_queue.py · candidate_queue's own order
Read the source ↗ · cell 46RCM3794
Location is ±25 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Read the source ↗ · cell 46QGD5279
Location is ±25 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±15 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±226 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±24 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Location is ±5.0 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Read the source ↗ · cell 46RCP7392
Location is ±15 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Read the source ↗ · cell 44RMQ9469
Location is ±15 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Read the source ↗ · cell 43QBD7442
Location is ±15 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Read the source ↗ · cell 46RCP7392
Location is ±15 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
Read the source ↗ · cell 43REQ3832
Location is ±221 km, coarser than one 1 km cell. Confirming it records the event; it does not write a 1 km training label.
No candidates. candidate_queue found no flood signal in the committed source snapshots that is not already labelled.
from hazards/flood/review.py · review.py's priority, one city after another
panel conflict: resolved label=1 via report (confidence=0.68, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.68, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.68, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.48, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.68, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.48, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.48, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.68, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.48, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.48, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.60, 3 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.60, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.60, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
panel conflict: resolved label=1 via report (confidence=0.60, 2 observations disagreed)
The sources disagree here. A human verdict overrides the model and the satellite label and enters the training panel as a verified observation.
Nothing waiting. Every conflicted or wide-band cell-period in the committed panels has a verdict.
from data/review/verdicts.csv · latest verdict first
No verdicts in the ledger yet. A verdict you give here appears in this column, then reaches the training panel once it is applied.
from log.obs_rejection + the ledger · reviewer rejections, then the ingest ledger
Duplicate of a record already in the log (same text, cell and hour)
Logged 2026-10-04 19:50 UTC in the rebuild this board was generated from
Kept with its reason. Rejected records are never deleted.
Duplicate of a record already in the log (same text, cell and hour)
Logged 2026-10-04 19:50 UTC in the rebuild this board was generated from
Kept with its reason. Rejected records are never deleted.
Nothing rejected. The ingest ledger this board was built from holds no rejected record, and no reviewer has rejected a candidate.
This board is the existing review loop, given a surface. A verdict here writes back as a verified observation that overrides the model and the satellite label (panel.resolve_cell_period).
A human verdict is the strongest label available and it is still one person's judgement.
Rejected records are kept with their reason, never deleted.
Ranking signals. Conflicted cell-periods: used. flood/review.py queues every conflicted panel row. Wide uncertainty band: used where a forecast CSV is committed; none for bengaluru, delhi, so their queue is conflict-only. Disagreement with an official warning: not modelled. the review queue does not compute a disagreement with an official warning yet, and the board does not invent one.
Sources: data/bengaluru/flood_panel.csv (committed 2026-08-05), data/delhi/flood_panel.csv (committed 2026-08-05); 6 source snapshots under data/sources/; ledger data/review/verdicts.csv (0 verdicts); log.obs_rejection (2 rows).
To apply downloaded verdicts: aqualens-flood review-verdicts verdicts.csv, then aqualens-flood panel --aoi <city> … --verified-csv data/review/verdicts.csv and aqualens-flood review-board --from-db. All reviewer tools →