Widen what the checker can see

0.12.0 → 0.12.1-draft

Attempt at a glance
Status	grading 8 of 12
Procedure touched	Steps 5 and 6
Tested against	150-field standard set · 8 of 12 runs, strict
Evidence	96.9% ±1.6 against 92.4% ±2 baseline
Spent	$17.44 · 27m
Owner	s.mendo

What changed in the procedure

before and after, by step

Step 5 · Consumed collateral · Step 6 · Insurance and flood

Take property value from the appraisal, then derive loan-to-value and lien position. Read the flood determination and set the insurance conditions.

WasThe checker verified each step against that step's entries alone, so a figure that only looks wrong next to another step's entry passed.

NowThe checker reads the whole worksheet and can contradict a figure using anything already recorded on the file.

What it did to the queue

the reviewer’s side of the loop

Would likely close the flood-zone stops, which are 1 of the 7 waiting. Nothing to claim yet — at eight runs the interval still overlaps the bundle it is trying to beat.

Thread

no verdict until every run is in

you

The checker is handed only the slice of the worksheet belonging to the step under review. That means it cannot see reserves when it checks collateral, or the flood determination when it checks insurance. Give it the whole worksheet and re-grade.

astro

Eight of twelve runs graded so far: 96.9% ±1.6. That interval still overlaps 0.12.0, so there is nothing to call — the remaining four runs should narrow it to about ±1.2.

No new failure shapes have appeared. Flood zone is still the one ambiguous parcel in the set.

astro

Nothing reaches the ledger until the run finishes. A partial result is not a record, and reading one as though it were is how a bundle gets promoted on noise.