[rusholme]

What and why

Task 4 of the unattended run breadth-2026-10-23 (specification): rapid_docs records what tasks 1 to 3 landed on Caltech-IPAC/rapid rebuild, as system behaviour with its numbers. Every earlier task ended TASK DONE (tasks 2 and 3 on their second attempts), so everything is documented as landed; nothing is reported as not landed.

Pull requests documented: task 1 psf #327 (the psf stage; the import draws psfid from the identity), #328 (the loop submits a psf job and holds that SCA's intakes), #329 (the intake's EPSFREF check), runbook #330; task 2 sca8 #331 (SSM retries), #332 (scripts/check_run.sh), #333 (publish set by set, bounded flush and close), #334 (association from the field's official chain under a resumed run), #335 (validate_run.py's stdevflux tolerance), runbook #336, #337 (consumer check starts at the run's first packet, client CRC off), runbook #338; task 3 sca18 runbook #339 with "Results of part 9", #340.

Addendum 3 is followed: every null-test count on the site (the breadth table, the F146 comparison, and the existing ramp and disjoint-reference bullets of runs 12 to 15) carries the artefact label in its sentence, citing the image study of one F146 epoch (PSF mismatch between the science ePSF and the resampled coadd, unmasked pixel defects, noise essentially none); the derived PSF's lower count is described as a smaller mismatch, not a cleaner image.

Files

• system/psf.md (new): scope and inputs (the file's meta.ref_file.epsf, fetched by name, the reference's shape, the detector/filter refusal), the reduction (in-focus G2V at the detector centre, block sum and why, centroid kept), declared outputs and headers, EPSFREF as the intake contract, settings, what it measured (the stamp beside dev's copy, the F146 comparison labelled as artefact counts, the job alone and under fan-out, the CRDS 504s), "Not decided here": the position, the reference PSF rule, a reference part in the key, the fetch.
• system/loop.md: the psf submission and hold as step 1 of the pass; association from the field's official chain (#334); Kafka set by set (#333); a six-filter measurements table and four bullets; "Not decided" narrows the bound bullet and gains the loop-stops-on-one-failure item.
• system/foundation.md: intro and stage list; the thin-chain table's psf row and the intake row's "left out" cell; a ### psf pointer; the psf memory row; a resumed live run continues from the dataset's official rows; what a Batch stage reads; Publication to Kafka (set by set, bounded flush, why) and the consumer check (start offset, CRC, retention); [psf] in the settings paragraph; "On SMDC" has nine parts, with a new Breadth part (the PSF, one dataset per filter and why: chains, fixes and resumes per band, no cross-band association or forced photometry; SCA 8; 18 SCAs; the checks); Validation gains check_run.sh; Measurements gains Breadth (the per-filter table from Results of part 9 and the task 2 and 3 PR bodies, labelled); "Not decided here": the other-filters PSF bullet and the psfs sequence bullet leave (closed by psf.md with decision-psf-from-epsf, and by #327's identity draw), the bound bullet is narrowed, and cross-band association, the null test's artefacts and the 107.5 s exposures enter.
• system/intake.md: the EPSFREF check as step 5 and in the inputs paragraph.
• system/decisions.md: decision-breadth-before-injection gains "each filter a dataset of its own with its own live run".
• system/index.md: the psf row and toctree entry; scope cells for decisions, foundation, intake, loop.
• AGENTS.md: psf in the list of stage pages.

Divergences from the brief, with the reason

• No new lines in decisions.md: both rulings the brief lists were recorded by foundation, loop, difference, intake, stage-contract, decisions, index: the housekeeping of 2026-10-22 and the reference bound #92, decision-psf-from-epsf (Ben, 2026-10-22 13:08 PDT) and decision-breadth-before-injection (Ben, 2026-10-22 16:03 PDT, 23:03 UTC). A second line would duplicate them; the per-filter-dataset part, absent from the existing line, is added to it.
• intake.md and AGENTS.md are edited, though the brief names neither for change: the intake's EPSFREF check is its behaviour now (the repository wins), and AGENTS.md lists the stage pages.
• loop.md and foundation.md also record #333, #334 and #337 beyond the brief's list (psf, dataset shape, check script, numbers): both pages stated one flush per pass, a run-scoped association rule and a consumer read from the earliest offset, all false after those PRs.
• No run ids, job ids, Batch ids or bucket names on the site; the runbook (part 9) carries them. Dataset-free wording ("one pointing's three epochs", "the wide-area run") replaces run numbers in the tables.

Build

Not built on the host (no Sphinx; provisioning is an install). docs.yml is the check. Locally: scripts/check-public-safety.sh OK; every {ref} target used resolves in decisions.md; the added text carries no em or en dashes and none of AGENTS.md's excess-vocabulary words; psf is in the index table and the toctree (the acceptance script derives its page list from system/*.md).

Measurements

Documentation only: no operation ran. The figures are tasks 1 to 3's PR bodies, the runbook's part 9 and "Results of part 9", and the image study LEDGER-images.md for the artefact label.

Improvements seen (not applied)

• reference.md still states its scope as (field, filter, SCA) (carried from foundation, loop, difference, intake, stage-contract, decisions, index: the housekeeping of 2026-10-22 and the reference bound #92).
• stage-contract.md does not name the psf stage's network read (the CRDS server); the foundation page now says what a Batch stage reads, and one sentence there would match.
• specification.md, tool.md and operations.md still describe the processing-date loop (carried from foundation, intake, loop, stage-contract, decisions, index: the HLWAS run from SOC L2 files to the Kafka topic #91 and foundation, loop, difference, intake, stage-contract, decisions, index: the housekeeping of 2026-10-22 and the reference bound #92).

What the next task must know

• New page system/psf.md; the Breadth part and Breadth measurements in foundation.md; status stays DRAFT on every page.

🤖 Generated with Claude Code