Skip to content

Latest commit

 

History

History
107 lines (94 loc) · 7.16 KB

HANDOFF.md

File metadata and controls

107 lines (94 loc) · 7.16 KB

Handoff — 2026-07-26

Right now: a live SLURM run is in flight

run_Chaeto_subset.sh (--config configs/Chaetothyriales_subset.csv --cluster_tool mmseqs --project Chaeto_subset) is actively executing on the cluster. As of 07:38 this session, PROFILE_SEARCH:BUILD_FAMILY_PROFILES:BUILD_CHUNK and PROFILE_LOSS_SEARCH:BUILD_FAMILY_PROFILES:BUILD_CHUNK tasks (famsa+hmmbuild per family chunk) are running/queuing on the preempt partition — ~100 tasks in flight, most PD on AssocGrpCpuLimit (queue is throttled, not stuck). Do not resubmit this job; check progress with squeue -u $USER | grep PROFI and tail -f logs/chaeto_subset.log / .nextflow.log, or just resume watching — nextflow's own -resume will pick up any already-completed chunks if it needs restarting.

When it finishes, outputs land in results/Chaeto_subset/ and get collated to view/Chaeto_subset/. A prior run already produced view/Chaeto_subset/{report,core,losses, novelties}.html (committed in cdc082a) — this run is likely a rerun after the family-size fix in 773ecf6 ("build out family profiles but deal with too large families") and the extract_family_seqs.py fix in f6f5476, so expect the outputs to change.

Uncommitted work — needs a decision before it's lost

git status shows modified-but-uncommitted:

  • configs/Chaetothyriales.csv and configs/Chaetothyriales_subset.csv — removed 3 IN genomes with no matching data (Exophiala spinifera JCM 15939, Herpotrichiellaceae sp. CTeuk-1624, and one more) and reclassified 3 species from IN → OUT (Aspergillus fumigatus Af293, Zymoseptoria tritici, Coccidioides immitis WA_211 — these are Pezizomycotina outgroup markers that were miscategorized as Chaetothyriales ingroup). This is exactly the kind of edit that should be committed once you're satisfied it's correct — check git diff configs/Chaetothyriales*.csv before committing, since the live run above was launched against whichever version of these files was on disk when nextflow read them (confirm which one that was before trusting the run's output).

There are also many untracked new files from recent work that have never been committed: bin/convert_bfd_samples.py, config_support/, configs/pezizo5.csv, db/modelorgs/*, db/pfam/, db/uniprot/, docs/make_samples_intermediate.md, run_Chaeto.sh, run_Chaeto_subset.sh, run_pezizo4.sh, steps_Chateo.md, tests/test_extract_family_seqs.py, plus .living/ and .claude/ (the mycelium scaffold + hooks — see CLAUDE.md, these are meant to be checked in). example/, stories/, skillpacks/, busco_pezizo5/, data/, and logs/ are almost certainly working-tree/data directories, not things to commit — sanity check each before staging anything broadly (don't git add -A).

Where the ADR-0002 family-profile pathway stands

  • Phase 1 (mmseqs+HMM profile pathway, issue #3) is merged (PR #4, .living/decisions.md 2026-07-21). todo/TODO_REGISTRY.md still lists it in-progress — flip it to complete next time you touch that file.
  • AVX2 famsa crash — already fixed. .living/learnings.md and .living/decisions.md record famsa SIGILLing on the cluster's Abu Dhabi (c[01-30]) nodes, deferred as a known issue on 2026-07-21. It's since been fixed: nextflow.config's slurm profile now has a withName: '.*BUILD_CHUNK' block pinning to -C ryzen|broadwell|cascade on the preempt queue. No action needed here — but the .living/ entries are now stale and should get a short "resolved" follow-up note pointing at the nextflow.config fix, so a future session doesn't re-discover the same bug.
  • "Too large families" handling was being worked on as of the last two commits (773ecf6, f6f5476, both 2026-07-25 20:39) — modules/build_family_profiles.nf gained oversized-family skip logic, bin/extract_family_seqs.py gained matching logic to pull family members out of the mmseqs cluster. This hasn't been validated against a completed run yet — that's what the in-flight Chaeto_subset job above is for. Once it completes, check results/Chaeto_subset/*/oversized_families.tsv (per family_max_members config comment in nextflow.config) to see how many families got skipped and whether that's reasonable for this dataset.
  • Phase 2 (todo/cross-method-support-column.md — sweep + BUSCO/controls validation battery) is still open, not started. The full end-to-end pezizo5/Chaetothyriales parity check between --cluster_tool pairwise and --cluster_tool mmseqs (shape-parity, reports render, BUSCO/negative-control recall) called for in Phase 1's acceptance criteria has still never been run — this Chaeto_subset job is the closest thing to that so far, but it's the mmseqs path only, not a pairwise-vs-mmseqs comparison.

Suggested next steps, in order

  1. Let the running job finish (or check on it) before doing anything else that touches results/Chaeto_subset/ or work/.
  2. Resolve and commit the configs/Chaetothyriales*.csv edits — confirm the 3 dropped-for-missing-data rows and 3 IN→OUT reclassifications are intentional and final, then commit with a clear message (e.g. "fix: drop genomes with no annotation data, move 3 misclassified Pezizomycotina markers from ingroup to outgroup").
  3. Triage the untracked files — commit the real deliverables (bin/, docs/, run_*.sh, tests/, configs/pezizo5.csv, .living/, .claude/), decide whether db/modelorgs/*, db/pfam/, db/uniprot/ belong in git or should be .gitignored (they look like large reference databases — check size before committing), and leave data/, logs/, example/, stories/, skillpacks/, busco_pezizo5/ alone unless you know they're meant to be tracked.
  4. Update .living/decisions.md/learnings.md with a short "AVX2 fix shipped" note and flip todo/TODO_REGISTRY.md's Phase 1 row to complete, since both are currently stale relative to what's actually in nextflow.config and merged to main.
  5. Once Chaeto_subset is validated, decide whether to scale the same --cluster_tool mmseqs run to the full configs/Chaetothyriales.csv (run_Chaeto.sh currently runs --cluster_tool pairwise — the default — not mmseqs; you'd need a variant or a flag change to actually exercise the profile pathway at full scale).
  6. Start Phase 2 (todo/cross-method-support-column.md) once Phase 1 is confirmed working end-to-end: the sweep grid, BUSCO/negative controls, and pairwise-vs-mmseqs cross-method concordance check are all still outstanding and are what will actually validate whether the mmseqs pathway's presence calls are trustworthy.

Quick reference

  • Live run: squeue -u $USER | grep -i profi, logs at logs/chaeto_subset.log and .nextflow.log.
  • Config diff pending decision: git diff configs/Chaetothyriales.csv configs/Chaetothyriales_subset.csv.
  • ADR: docs/adr/0002-family-profile-search-pathway.md.
  • Todos: todo/TODO_REGISTRY.md, especially profile-search-pathway.md (Phase 1, needs status flip) and cross-method-support-column.md (Phase 2, next up).
  • Session log for this session: .living/log/2026-07-26-001-novinvenio.md.