maintenance: anonymise the named science target in the WFC3-IR hole regression

[Jammy2211]

Overview

Library repos must not carry science-specific paths or named science targets. The WFC3-IR mosaic-hole regression in autoreduce — the reason the IR adapter ships STScI's MDRIZTAB DQ-bit rows instead of drizzlepac's final_bits = "0" default — is named after the real target it was found on. Rename every reference to the generic regression it actually encodes (HST program 14653, F160W, five-exposure mosaic, 123-px hole at r = 5.3") and drop the target name. Comment/name text only: no behaviour change, no API change, no test-outcome change.

Plan

• Reword the DQ-bit rationale comment in the WFC3-IR adapter so it cites the program/filter/exposure-count rather than the target name.
• Rename the regression test to describe the dataset shape it guards (test_five_exposure_f160w_mosaic_would_not_have_holed) and reword its comment the same way.
• Reword the "Blobs (DQ 512)" paragraph in the WFC3 design doc so the failing dataset is identified generically.
• Sweep the whole repo for any remaining occurrence of the target name and fix anything tracked; leave the gitignored local drizzle cache alone.
• Run the autoreduce test suite to confirm the rename is behaviour-neutral, then ship.

Detailed implementation plan

Work Classification

Library (PyAutoReduce source + its own tests + its own design docs). No workspace follow-up: nothing user-facing changes, and the renamed test is referenced nowhere outside its own file.

Affected Repositories

• PyAutoReduce (primary)

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoReduce	main	clean

Recent branches: main only. worktree_check_conflict clean — no active task claims PyAutoReduce (the two open tasks claim autolens_profiling, PyAutoArray, PyAutoGalaxy).

Suggested branch: feature/anonymise-wfc3-ir-hole-regression-target
Worktree root: ~/Code/PyAutoLabs-wt/anonymise-wfc3-ir-hole-regression-target/

Implementation Steps

1. autoreduce/instruments/wfc3_ir.py (~L37) — in the dq_bits_rows rationale comment, replace (PJ011646, 5 exposures, a 123-px hole at r = 5.3") with (program 14653, F160W, 5 exposures, a 123-px hole at r = 5.3"); reflow the following comment lines so the block stays inside the line budget.
2. test_autoreduce/test_target_and_instruments.py (L157-161) — rename test_pj011646_would_not_have_holed → test_five_exposure_f160w_mosaic_would_not_have_holed, and reword its body comment to describe an HST program 14653 F160W five-exposure mosaic. Assertions untouched.
3. docs/design/wfc3.md (~L52) — in the "Blobs (DQ 512)" paragraph, replace the sentence subject PJ011646 (program 14653, F160W, 5 exposures, 2–6 px dithers) with a generic An HST program 14653 dataset (F160W, 5 exposures, 2–6 px dithers); reflow the paragraph.
4. grep -rIn -i pj011646 --exclude-dir=.git over the repo → must return nothing tracked. scripts/cache/cache_manifest.json is gitignored (.gitignore:17 scripts/cache/) and does not contain the name in any case.
5. Run the suite in the task worktree (source activate.sh, python3 -m pytest test_autoreduce -q) — every test must pass, including the renamed one, with the same count as before.

Key Files

• autoreduce/instruments/wfc3_ir.py — WFC3-IR adapter; carries the dq_bits_rows table and the comment explaining why it exists.
• test_autoreduce/test_target_and_instruments.py — the DQ-bits regression legs; the renamed test asserts final_bits & 512 at numimages = 5.
• docs/design/wfc3.md — the WFC3 design note whose "Blobs (DQ 512)" section narrates the failure.

Trade-offs / risks

• The Brain's feature pass flags "public-API change may ripple downstream" from its generic heuristics; that does not apply here — no exported symbol changes, and the renamed identifier is a pytest method referenced only in its own file.
• The regression's provenance stays recoverable: PyAutoMind history records (complete/, condemned.md, autonomy_log.md) are deliberately not touched — they are records of work that happened.

Original Prompt

Click to expand starting prompt

# Anonymise the named science target in the WFC3-IR hole regression

Type: maintenance
Target: PyAutoReduce
Repos:
- PyAutoReduce
Difficulty: small
Autonomy: safe
Priority: low
Status: formalised

# Anonymise the named science target in the WFC3-IR hole regression

Type: maintenance
Difficulty: small
Autonomy: safe
Priority: low

Library repos must not name science targets — "these repos shouldnt have science specific paths". The WFC3-IR mosaic-hole regression in autoreduce is named after the real target it was found on. Rename the references to the generic regression (HST program 14653, F160W, 5-exposure mosaic with a 123-px hole) and drop the target name:
- `autoreduce/instruments/wfc3_ir.py:37` (comment)
- `test_autoreduce/test_target_and_instruments.py:157-158` (test name `test_pj011646_would_not_have_holed` and its comment)
- `docs/design/wfc3.md:52`

No behaviour change. Mind history records (complete/, condemned.md, autonomy_log.md) stay untouched — they are records of work that happened.

[Jammy2211]

Branch pushed — PR withheld on Heart RED

Implementation is complete and pushed; the pull request is deliberately not opened because the readiness gate (ship_library step 3) came back RED.

• Branch: feature/anonymise-wfc3-ir-hole-regression-target (PyAutoReduce, pushed to origin)
• Commit: b4ee3697097fe91fe332b12d61c8479a6756a3fb
• Worktree: ~/Code/PyAutoLabs-wt/anonymise-wfc3-ir-hole-regression-target
• Tests: python3 -m pytest test_autoreduce/ -q → 299 passed, 3 skipped, 34 warnings, 12.77s

Heart readiness verdict (verbatim)

pyauto-heart readiness --json at 2026-08-28T21:34:42Z — verdict: red, score 45, profile default.

red_reasons:

• release validation FAILED (stage integrate)

yellow_reasons:

• workspace validation not passing (2 failed, cloud#33179766004: autolens_test scripts/imaging/rectangular_mge.py, autolens_test scripts/imaging/rectangular_mge_rtu.py)
• manifest drift: session-start hooks (generated) — 32 mismatch(es) vs PyAutoMind/repos.yaml

None of these reasons touch PyAutoReduce, and this change is comment/test-name/doc text only — but the gate is the gate, so the PR waits for a human decision (clear the RED, or authorise the narrow corrective-PR exception in AUTONOMY.md).

What landed

File	Change
autoreduce/instruments/wfc3_ir.py	dq_bits_rows rationale comment now cites program 14653, F160W, 5 exposures, a 123-px hole at r = 5.3"
test_autoreduce/test_target_and_instruments.py	test_pj011646_would_not_have_holed → test_five_exposure_f160w_mosaic_would_not_have_holed; comment reworded. Assertions untouched.
docs/design/wfc3.md	"Blobs (DQ 512)" paragraph now reads "An HST program 14653 dataset (F160W, 5 exposures, 2–6 px dithers) failed packaging on exactly this…"

grep -rIn -i -e pj011646 -e 011646 --exclude-dir=.git over the repo now returns nothing. The only other place the string lived, scripts/cache/cache_manifest.json, is gitignored (.gitignore:17 scripts/cache/) and does not contain it in any case. PyAutoMind history records (complete/, condemned.md, autonomy_log.md) were deliberately left untouched.

PR body to use when the gate clears

Paste verbatim into gh pr create --label pending-release

## Summary

Library repos must not name science targets. The WFC3-IR mosaic-hole regression
— the reason the IR adapter ships STScI's MDRIZTAB `dq_bits_rows` instead of
drizzlepac's `final_bits = "0"` default — was named after the real target it was
found on. It is now identified by the dataset shape it actually encodes: HST
program 14653, F160W, a five-exposure mosaic with a 123-px hole at r = 5.3".

Three sites: the adapter's rationale comment, the regression test's name and
comment, and the "Blobs (DQ 512)" paragraph of the WFC3 design note.

## API Changes

None — internal changes only. The renamed identifier is a pytest method
referenced nowhere outside its own file; no exported symbol, signature, default
or behaviour changes.

## Test Plan
- [x] `python3 -m pytest test_autoreduce/ -q` → 299 passed, 3 skipped
- [x] `grep -rIn -i pj011646 --exclude-dir=.git` returns nothing tracked
- [x] Diff is comment / test-name / prose text only — no executable line changed

Generated by the PyAutoLabs agent workflow.

[Jammy2211]

PR open — under an acknowledged Heart RED

The human authorised opening and merging under the RED verbatim: "open prs under red and merge i acknowledge". The readiness reasons are unchanged from the comment above, and none of them touches PyAutoReduce:

• RED: release validation FAILED (stage integrate)
• YELLOW: workspace validation not passing (2 failed, cloud#33179766004: autolens_test scripts/imaging/rectangular_mge.py, rectangular_mge_rtu.py)
• YELLOW: manifest drift: session-start hooks (generated) — 32 mismatch(es) vs PyAutoMind/repos.yaml

The acknowledgement is recorded on the task entry in PyAutoMind/active.md (heart-ack:).

• PR: maintenance: anonymise the named science target in the WFC3-IR hole regression #74 — body pasted verbatim from the drafted block above, label pending-release, head feature/anonymise-wfc3-ir-hole-regression-target (b4ee369).

Opened in the same session under the same acknowledgement (unrelated task, PyAutoBrain#315):

• feat(clone): a sync mode, and the shared rename table birth and sync now both use PyAutoBrain#316
• feat: science-project memory — state.md, and a session-start block that reads it autolens_assistant#116
• feat: science-project memory — state.md as the head pointer a session reads first autogalaxy_assistant#20
• feat: science-project memory, plus the two birth-substitution gaps closed autocti_assistant#27
• feat: science-project memory — synced from autolens_assistant, rejects hand-resolved autofit_assistant#31

[Jammy2211]

Shipped

• PR: maintenance: anonymise the named science target in the WFC3-IR hole regression #74 — MERGED (merge commit 5ee7c4e1a8cb1b2a75863ff4c4193262f422a5cf)
• CI: Tests workflow green — unittest (3.12) success, unittest (3.13) success, unittest-nojax skipped (conditional leg). mergeStateStatus: CLEAN.
• Merged under an acknowledged Heart RED, human-authorised verbatim: "open prs under red and merge i acknowledge". RED reason release validation FAILED (stage integrate); YELLOWs workspace validation not passing (2 failed, cloud#33179766004: autolens_test scripts/imaging/rectangular_mge.py, rectangular_mge_rtu.py) and manifest drift: session-start hooks (generated) — 32 mismatch(es) vs PyAutoMind/repos.yaml. None touches PyAutoReduce.

The WFC3-IR mosaic-hole regression is now identified by the dataset shape it encodes (HST program 14653, F160W, five-exposure mosaic, 123-px hole at r = 5.3") in all three sites: the IR adapter's dq_bits_rows rationale comment, the regression test's name and comment, and the "Blobs (DQ 512)" paragraph of docs/design/wfc3.md. Comment/test-name/doc text only — no executable line changed; 299 passed, 3 skipped.

PyAutoMind history records (complete/, condemned.md, autonomy_log.md) were deliberately left untouched.