Approved offline NEOWISE W2 flag interpretation repair

14 October 2026. Human instruction: “Please fix the NEOWISE flag-format mismatch using the data already downloaded, resume the research goal, and keep the app and blog updated with verified results while monitoring disk space.” This authorizes an offline representation repair and a corrected readout, with no further survey request. It does not authorize restricted Atlas/E155/E121 or archival operations.

The original 874 rows, original failed TAP request, three completed Gator requests, computational 0/3, incompatible-contract qualification and public/private delivery remain historical evidence. All outcomes and raw flag shapes are already consumed; this amendment is frozen before corrected execution, not an outcome-blind study or fresh physical evaluation. Do not label the repair a new discovery.

Exact change and unchanged science

Decode the first two quality characters as W1/W2. Accept exactly two uppercase characters from A/B/C/U/X, or exactly four with the same first-two alphabet and literal trailing XX. Return the second character. No padding, trimming, case conversion, numeric coercion, additional bands or missing-value imputation. Malformed/unknown encodings fail closed. Raw VO/CSV and reported strings remain byte-identical. The official NEOWISE supplement describes the band positions and unused XX suffix; the downloaded Gator rows carry the active two characters. The cause of that delivery/documentation difference remains unknown.

The required W2 grade remains A. Every other numerical/flag quality criterion, 1-arcsecond unique geometry, 1e-5-arcsecond boundary guard, duplicate-exposure rule, 60-second header consistency, six alternatives, fixed three best records, 30-day grouping, five rows/group, three groups, 365-day span and K>=2/3 gate remain exact. No substitution, extra feature, model training or threshold tuning. The producer uses an explicit decoder; the independent checker implements its own literal predicate and Decimal/vector geometry without importing that decoder.

Fresh synthetic controls precede corrected native execution: all 25 active flag pairs in both encodings, malformed lengths/alphabet/whitespace/null, unchanged SNR/chi-square/flags/error boundaries, and untouched row strings. A first guarded control launch was refused by sandbox process-priority setup before child creation; its receipt is preserved. The permitted fresh control stage runs no native data.

Bounds, verification and stopping

Reuse the exact cached three cones and 874 records only. New network observations: zero. One offline repair unit, <=30 minutes cumulative worker time, one CPU/no GPU, 256 MiB sampled RSS, 5 GiB reserve, <=16 MiB combined scientific/publication packet. Controls <=60 sec/1 MiB, analyzer <=300 sec/2 MiB, independent checker <=300 sec/1 MiB, final delta audit <=60 sec/1 MiB. Preserve failures and stop on identity, geometry, byte binding, interpretation or resource disagreement; no automatic corrective scientific rerun or successor after a terminal result.

Independently recompute all rows, counterpart counts, epochs and the primary gate. Then verify the new rows equal old rows except for the single photo-quality reason and its resulting acceptance boolean. Publish the measured corrected support, whether pass or fail, beside the original result and qualification. Rejected pipeline measurements are not physical nondetections. No physical ownership, youth, population precision/recall, star or birth is established by this gate. A successful parser repair does not complete the full detector goal.