fig5-sweepga-f16-validated-chop-rerun-after-real-audit

Metadata

Status	open ‖ paused
Agent identity	`3184716484e6f0ea08bb13539daf07686ee79d440505f1fdf2de0357707034c3`
Created	2026-06-22T19:56:58.747917953+00:00
Started	2026-06-22T20:32:35.097545567+00:00
Tags	`pedigree`, `fig5`, `sweepga`, `pafchop`, `slurm`, `validation`, `plot`, `eval-scheduled`

Description

Paused downstream rerun. Do not unpause until redo-sweepga-paf-filter-identity-scoring-audit has produced both required files and explicitly says the rerun may proceed.

Important stale state:

Do not reuse outputs from cancelled Slurm array 1705408 or old f16 chopped outputs created before PAF semantics validation.
The old task audit-sweepga-paf-filter-identity-scoring is invalid for gating because it did not produce SWEEPGA_PAF_FILTER_IDENTITY_AUDIT.md or the TSV audit summary.

Goal: rerun the Fig5 sweepGA f16 sensitivity using exact CIGAR-based PAF chopping, small configurable chunk sizes no larger than 10 kb unless the audit justifies otherwise, and the command line approved by SWEEPGA_PAF_FILTER_IDENTITY_AUDIT.md.

Required work:

Heavy chopping/sweep filtering only through Slurm, never on the head node. Shared-node jobs are preferred; use a few cores per task and /dev/shm for scratch where needed.
Use the repaired pafchop-rs build that recomputes qstart/qend/tstart/tend/col10/col11 and CIGAR-derived tags. Reject raw PAF rows that cannot be exactly split.
Run a small matrix of chunk sizes including 10 kb and at least one smaller size if practical, plus the f16 raw/un-chopped reference.
Apply sweepGA filtering exactly as approved by the audit, expected to be --num-mappings 1:1 --scaffold-jump 0 with identity/ANI scoring only if the audit passes.
Summarize whether chr3 target homology survives 1:1 filtering and compare to wfmash/minimap2/PGGB untangling evidence already collected.

Acceptance criteria:

Slurm job IDs and executable paths are recorded.
No whole-genome pafchop/sweepGA work ran on the head node.
Deliverables include a TSV summary and a plot suitable for updating the Fig5 sweepGA panel/key supplemental figure.
The report clearly says whether the result supports smaller chopping, f16 frequency, minimap2/wfmash cross-validation, or a sweepGA limitation.

Paused downstream rerun. Do not unpause until redo-sweepga-paf-filter-identity-scoring-audit has produced both required files and explicitly says the rerun may proceed.

Important stale state:
- Do not reuse outputs from cancelled Slurm array 1705408 or old f16 chopped outputs created before PAF semantics validation.
- The old task audit-sweepga-paf-filter-identity-scoring is invalid for gating because it did not produce SWEEPGA_PAF_FILTER_IDENTITY_AUDIT.md or the TSV audit summary.

Required work:
- Heavy chopping/sweep filtering only through Slurm, never on the head node. Shared-node jobs are preferred; use a few cores per task and /dev/shm for scratch where needed.
- Use the repaired pafchop-rs build that recomputes qstart/qend/tstart/tend/col10/col11 and CIGAR-derived tags. Reject raw PAF rows that cannot be exactly split.
- Run a small matrix of chunk sizes including 10 kb and at least one smaller size if practical, plus the f16 raw/un-chopped reference.
- Apply sweepGA filtering exactly as approved by the audit, expected to be --num-mappings 1:1 --scaffold-jump 0 with identity/ANI scoring only if the audit passes.
- Summarize whether chr3 target homology survives 1:1 filtering and compare to wfmash/minimap2/PGGB untangling evidence already collected.

Acceptance criteria:
- Slurm job IDs and executable paths are recorded.
- No whole-genome pafchop/sweepGA work ran on the head node.
- Deliverables include a TSV summary and a plot suitable for updating the Fig5 sweepGA panel/key supplemental figure.
- The report clearly says whether the result supports smaller chopping, f16 frequency, minimap2/wfmash cross-validation, or a sweepGA limitation.

Depends on

done .assign-fig5-sweepga-f16-validated-chop-rerun-after-real-audit

Required by

open .flip-fig5-sweepga-f16-validated-chop-rerun-after-real-audit

Log

2026-06-22T19:56:58.709814377+00:00 Task paused
2026-06-22T20:30:35.526867914+00:00 Task published
2026-06-22T20:32:35.097549895+00:00 Spawned by coordinator --executor codex --model gpt-5.5
2026-06-22T20:33:41.449273440+00:00 Evaluation: inspected WG state, branch history, and expected deliverable paths; no rerun outputs or task commits found.
2026-06-22T20:33:47.882211054+00:00 Agent 'agent-2675' killed — task auto-paused (use 'wg resume' to re-enable dispatch)
2026-06-22T20:40:34.509954601+00:00 Paused old LLM-routed rerun after it was assigned to an evaluator role and began grading instead of implementing. Replacement shell submission task fig5-sweepga-f16-validated-chop-submit-slurm-shell now owns the live Slurm run: array 1705617 and dependent summary 1705618.