Repository navigation
Conversation
S01-S03 all use 1crn_small.pdb -- a single chain numbered 1..N, which is exactly the shape that hides stabiliNNatorApp#23: ranked residues were labelled by indexing the concatenated sequence with a residue number, so a single chain starting at 1 labels correctly by accident. 3ft7.pdb has chains A/B numbered 8-52 and mislabels 86 of 88 proline rows against the old code. Adding it makes the multi-chain case permanent coverage rather than a one-off manual check after this rebuild. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DEWo4xvL59PHC1QUvtTV9F
…or fixes live Two matrix runs against the deployed image: 52 positive cases, then the full set with --include-negative. No failures in either. Both app rows were confirmed on folding_260901.1.sif by reading the container path out of the user's own two morning jobs, so no probe submissions were needed for that half. The two fixes this rebuild was for are confirmed in production: - #22 TSV type: S01 and S04 result folders list `tsv` for both _summary.tsv files, where they used to land `unspecified`. - #23 residue labels: 0 mislabeled on the raw multi-chain 3ft7, with the counterfactual "old sequence[pos-1] lookup would mislabel 86/88". The same verifier reports 0/86 on a *cleaned* 3ft7 — contiguous renumbering makes the two lookups coincide, so only the raw structure exercises the regression. N03 had been passing for the wrong reason: it pointed at inputs/not_fasta.txt, which was absent from the workspace, so it failed on "file is empty or unreadable" rather than the FASTA check it exists to test. Fixture uploaded; N03 now quotes the fixture's first line. Also corrects the standing claim that the alpha bundle blocks the merged UI changes. That rested on grepping for ViewPredictStructureReport, a symbol absent from the bundle, from alpha's ActionBar.js, and from our own checkout — it proved nothing. Alpha does serve the new typed-row form. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DEWo4xvL59PHC1QUvtTV9F
coconut-stabilinnator generated reports from the pre-fix template and the label verifier exited 0 on them. It recomputes resLabel from the report's embedded data JSON and never inspects what the template rendered, and the bug lives in the template — so our "residue labels confirmed" rested on a check that could not distinguish a fixed image from an unfixed one. Their 690ffce (stabiliNNatorApp #24, test-only, unmerged, not in folding_260901.1) scans the report's inline script for DATA.input.sequence[...pos-1]. Re-run against the same S04 report pulled from the workspace: rendering ok, discriminating input yes, exit 0. The conclusion stands; the evidence for it did not. Recording both. Also notes which inputs can exercise the regression at all: raw 3ft7 gives 86/88, a cleaned contiguously-renumbered copy collapses to 0/86 and exits 3 under --require-discriminating, and S01-S03's crambin reports NO — so the flag belongs on S04 only, once #24 merges. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DEWo4xvL59PHC1QUvtTV9F
Both reproduce against folding_260901.1 from this side: the rendering-aware label check (690ffce) and tests/verify_mmcif_rejections.sh (f9007cc, 3/3 against the deployed App-StabiliNNator.pl). Neither is wired in yet — the matrix and the acceptance suite would otherwise depend on an unmerged branch. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DEWo4xvL59PHC1QUvtTV9F
…e it Asked for on 2026-09-03 and never run. It is now larger than when it was requested: that day's two verification runs added 107 matrix runs (214 objects) to the same directory. p3-rm is blocked by the auto-mode permission classifier as a permanent deletion, so this cannot be closed from a Claude session at all — it needs wilke. Recording that explicitly rather than leaving it to look like an ordinary pending checkbox that a later session might assume it can clear. Complete script regenerated at ~/clean_apptests.sh: outside the repo, but outside /tmp too, so it survives the reboot. Everything it deletes has its verdict committed in docs/test-reports/matrix_20260903_*.json. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DEWo4xvL59PHC1QUvtTV9F
Run on wilke's explicit instruction. 107 matrix runs (214 objects), the July/August hand-made StabiliNNator runs, the two flat-upload PDBs and _inputs/ are gone. inputs/ still holds all seven fixtures and the 09-03 runs are untouched. Verified by listing the directory afterwards, not by the script's exit status: p3-rm printed nothing at all on success, so exit 0 alone would not have distinguished "deleted" from "did nothing". Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DEWo4xvL59PHC1QUvtTV9F
The old coconut ColabFold server is gone, but UniRef30 2302, the envdb, mmseqs2.sif and the colabfold conda env are still on the host. Recorded where they are and that the envdb index may never have been built. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DEWo4xvL59PHC1QUvtTV9F
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
One matrix addition and the production verification record for
folding_260901.1.sif. No service code changes —test_data/,docs/, STATUS/TODO only.inputs/3ft7.pdb. Exists so the Test: full multi-tool CWL workflow execution (gap #5) #23 residue-label regression can actually fire: on a cleaned, contiguously-renumbered structure offset-lookup and identity-lookup coincide and the counterfactual collapses to0/N. S01–S03's crambin cannot exercise it either (one chain numbered from 1).docs/test-report-260903-api.md— the verification writeup, with the three results JSONs.What was verified (2026-09-03)
folding_260901.1--include-negativep3-ls -lTon S01 and S04 result folders readstsv(wasunspecified)rendering: ok,discriminating input: yes, counterfactual86/88on S04StabiliNNator cases landed on four distinct
computehosts across the two runs (fir, pear, bio-compute-04, alder), so that partition's image staging is no longer a single-host observation.Two tests that were passing for the wrong reason — recorded, not hidden
inputs/not_fasta.txt, which did not exist in the workspace, so it failed on "file is empty or unreadable" rather than the FASTA check it exists for. Fixture uploaded; it now quotes the fixture's first line.caa7dcd) recomputesresLabelfrom the report's embedded data and never inspects the rendered template — it exits 0 on a pre-fix report. The "confirmed" line in the first draft of the report rested on that. Re-verified with stabiliNNatorApp690ffce(CI: validate container builds on every PR (gap #10) #24, scans the report's inline script forsequence[pos-1]). Both the weak evidence and the correction are in the report so the weak line does not get re-quoted.Deferred, deliberately
--require-discriminatingon S04 andtests/verify_mmcif_rejections.shin the acceptance suite both reproduce againstfolding_260901.1but are not wired in — they live on stabiliNNatorApp #24, which is unmerged. Tracked in TODO.Incidental
Task 23519232 (a real UI submission) sent
dna_fileas a directory: the workspace fixtures are typedreads, the DNA selector's type filter excludes that, the browse dialog looks empty, andWorkspaceObjectSelector.js:845takes the folder. The service caught it correctly. Two BV-BRC-Web follow-ups recorded in TODO.🤖 Generated with Claude Code
https://claude.ai/code/session_01DEWo4xvL59PHC1QUvtTV9F