Deliberate native PSF-quality delivery repair — original science unchanged

13 October 2026. The previous goal turn made progress by independently measuring a marginal gas/radio overlap; that neither resolves infrared ownership nor supplies reliable temporal measurements. The important unanswered native-quality question remains at the original transport stop. This is a deliberate, prospectively frozen transport resumption under the human’s continued research goal, not a renamed experiment, automatic retry, new survey or looser scientific gate.

Preserved attempt and unchanged question

Preserve 518e8683 registration, 275d8221 terminal result, the exact stopped TAP request and its raw/no-body/resource evidence. The old request timed out before HTTP status; cause beyond that local timeout and remote query state are unknown. No live remote-job handle exists in its receipt, and this repair does not claim it stopped/restart/cancel a remote job. It does not resend that URL.

The same three cases, six fixed Catalog alternatives, original 12-arcsec cones, 29 native fields, unique <=1-arcsec geometry, exposure/time binding, all strict PSF-quality rules, epoch groups/365-day span and K>=2-of-3 criterion stay exact. Do not replace a blue point by the red alternative, add features/labels, train, relax a flag or reinterpret missing observations as physical nondetections. Reuse the exact original analyzer and independent native checker unchanged. The old unmeasured result remains a historical attempt, never converted to zero.

Justified changed delivery and format contract

IRSA documents a maintained older catalogue-search interface with column selection, cone units, maximum rows and VO Table output. It offers the same named catalogue without the previously failed ADQL URL. Its documentation recommends SCS/TAP to new users; this existing study tests the maintained interface once as a bounded alternative. No faster delivery or explanation of the TAP failure is assumed. Official catalogue API.

Freeze exactly three URLs to https://irsa.ipac.caltech.edu/cgi-bin/Gator/nph-query with catalog=neowiser_p1bs_psd, spatial=cone, literal original decimal-centre objstr, radius=12, radunits=arcsec, outfmt=3, outrows=5001, order=source_id and the same selcols. No name lookup, server-side quality cut, list upload, email request, redirect, fallback, pagination or literal-URL retry. Stop at the first transport/schema/identity/geometry/resource failure. Three resumed requests at most, four attempts total including the old failed first request. State one deliberate amended resumption explicitly; do not say there was no resumption or pretend four cumulative attempts were three.

Retain complete raw VO bytes and FIELD/unit/type metadata, including any extra server columns. Project the same registered fields into CSV in original column order using literal TABLEDATA text, no floating-point/int coercion, trimming, imputation, calibration or flag rewriting. Empty cells remain empty; explicit null/NaN spellings remain native. Preserve large integer strings and padded flags. Reject missing/duplicate fields, XML entities/DOCTYPE, binary serialization, HTML/error/overflow output, duplicate IDs or >=5001 rows. Exact raw source IDs, cntr construction and finite positions/MJD must pass before analysis. All returned centroids must meet the original cone check in the unchanged native verifier. The CSV is a lossless field projection; extra columns remain in the original VO.

An independent transport checker uses a separate DOM parser, compares every VO cell with normalized CSV, validates frozen URLs/parameters/full bytes and receipts, and permits native analysis only after all three deliveries are complete. If stopped, it verifies the terminal stop/prefix and marks the quality count/gate unmeasured. It never counts transport success as physical source evidence. Original native quality/geometry checking still uses Decimal thresholds and independent vector angles, without importing the scientific worker.

Synthetic controls precede native requests: exact counter >2^53, leading-zero flags, XML namespaces, empty and explicit null cells, extra native fields, missing/duplicate columns, forbidden entities, overflow/error status, truncated row sentinel and independent parser/CSV equality. No measurements are read in registration. Original input/script/stop hashes must match before execution.

Cumulative resources, stopping and delivery

Carry the original limits forward: 90 seconds per request, 2 MiB each raw VO and 2 MiB each normalized CSV, at most 6 MiB of each, acquisition stage <=8 MiB including originals/CSV/logs. Its guard gets only 508 seconds, keeping the old 91.19-second acquisition plus this attempt below 600. The original and repaired packets together stay <=16 MiB, cumulative work <=30 minutes, one CPU/no GPU, 256 MiB sampled RSS and 5 GiB disk reserve. New source validation is <=60 sec/1 MiB, analysis <=300 sec/2 MiB and native check <=300 sec/1 MiB; registration controls <=60 sec/1 MiB. Check cumulative size before each stage. Preserve failure prefixes. No paid cloud work, images, ancillary observations or original evidence deletion.

Run one registered resumption, independently verify its measured or unanswered outcome, snapshot coherent changes and stop. No further repair, alternative route, threshold change or successor follows automatically. All old model/gate/parent/ unknown states remain unchanged. A passed quality/cadence gate would establish only a necessary measurement route; no calibrated ownership, youth, physical precision/recall, new star, birth or robust detector is claimed. The full goal remains active and unmet. Atlas/E155/E121 and N20/c6828/archive restrictions persist.