Skip to content

Phase 2: bundled real JWST dataset + the front door - #4

Merged
Jammy2211 merged 4 commits into
mainfrom
feature/phase-2-dataset
Aug 1, 2026
Merged

Phase 2: bundled real JWST dataset + the front door#4
Jammy2211 merged 4 commits into
mainfrom
feature/phase-2-dataset

Conversation

@Jammy2211

Copy link
Copy Markdown
Contributor

Phase 2 of PyAutoLabs/PyAutoBrain#188 — the real dataset and the user-facing front door. Stacked on #3 (4a); GitHub retargets on merge.

The dataset — dataset/imaging/cosj100020+015344/ (3.6 MB)

COSJ100020+015344: COSMOS2020 Classic 499609 / zCOSMOS 812167, spectroscopic z = 0.3422 (zCOSMOS-Bright DR3 conf 4.5, cited), log M* = 10.54 smooth early-type in the COSMOS-Web field. Four NIRCam bands (F115W/F150W 419² @0.03"; F277W/F444W 209² @0.06") reduced through PyAutoReduce from the cached program-1727 exposures — zero MAST traffic. Confirmed non-lens by COWLS cross-match (null at 55.5"; the match recovers the ring at 0.1" as control). Measured-only info.json (second-moments shape, curve-of-growth R_eff, band-to-band scatter as the error bar; no invented numbers). 448-line provenance README carrying the honest caveats: sky pedestal un-subtracted (5–19× noise — model it), tier-2b STPSF kernel (tier-1 ePSF rejected — star finder selects galaxies on this field; defect documented + auditable in reduction_manifest.json), SW coverage steps.

Front door

  • docs/make_readme_figures.py rebuilds the hero offline from the shipped FITS; every annotation read from info.json at runtime so the figure cannot drift from the data.
  • README v2: hero + the three decisions a user must own (mask extent, sky, the 2.6" neighbour), chat + coding-agent paths, two dataset-grounded starter prompts (prompt 2 is shipped verbatim as the frozen easy benchmark card — kept identical by a unit test, which now runs instead of skipping).
  • wiki/core/operations/dataset.md authored and indexed; PENDING/AGENTS/llms.txt reconciled.
  • Integration fixed a real staging defect: reduction_manifest.json recorded the rejected tier-1 ePSF as the shipped kernel — rewritten from the tier-2b record.

Validation (clean-venv wheel autogalaxy==2026.7.29.2)

--check-version clean · --scope all 0 missing · citations 921/0 · provenance 0 errors · lint clean · pytest 68 passed / 0 skipped · link sweep 694/0 · all four bands load via ag.Imaging.from_fits · figure script reruns byte-identically · sparse-cone citation sweep 0 violations.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EVSbY4u1JLfqRMgXYtnN9v

Jammy2211 and others added 4 commits August 1, 2026 17:08
dataset/imaging/cosj100020+015344: 4-band NIRCam cutouts (F115W/F150W
@0.03", F277W/F444W @0.06") of a z=0.3422 (spec, zCOSMOS DR3) early-type
in the COSMOS-Web field, reduced via PyAutoReduce from cached program-1727
exposures; measured-only info.json; 448-line provenance README (sky
pedestal un-subtracted; tier-2b STPSF kernel — tier-1 ePSF rejected with
the defect documented; COWLS non-lens cross-match). 3.6 MB.

docs/make_readme_figures.py rebuilds the hero offline from the shipped
FITS (annotations read from info.json at runtime); README v2 front door
with two dataset-grounded starter prompts; easy benchmark card (prompt
kept test-identical to README prompt 2 — both benchmark tests now run and
pass); wiki/core/operations/dataset.md authored + indexed; PENDING/AGENTS/
llms.txt reconciled.

Fixed in integration: reduction_manifest.json falsely recorded the
rejected tier-1 ePSF as the shipped kernel — rewritten from the tier-2b
record with the rejection kept auditable.

Validation: pytest 68 passed / 0 skipped; audit suite clean (922
citations); link sweep 694/0; figure script reruns byte-identically.

Epic: PyAutoLabs/PyAutoBrain#188 (Phase 2)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EVSbY4u1JLfqRMgXYtnN9v
… self-pin

The wiki-currency job's sparse clone has no workspace dataset/ dir, so the
dataset/.gitignore source path failed --check-citations there; the
autogalaxy_assistant pin now names the commit that actually introduced the
cited dataset files.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EVSbY4u1JLfqRMgXYtnN9v
@Jammy2211
Jammy2211 changed the base branch from feature/phase-4a-core-skills to main August 1, 2026 16:22
@Jammy2211
Jammy2211 merged commit 0c5ade5 into main Aug 1, 2026
2 checks passed
@Jammy2211
Jammy2211 deleted the feature/phase-2-dataset branch August 1, 2026 16:22
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant