Metadata
| Status | done |
|---|---|
| Assigned | agent-30 |
| Agent identity | 3184716484e6f0ea08bb13539daf07686ee79d440505f1fdf2de0357707034c3 |
| Created | 2026-05-01T18:39:15.309972074+00:00 |
| Started | 2026-05-01T18:40:07.719859823+00:00 |
| Completed | 2026-05-01T18:48:14.886630906+00:00 |
| Tags | grant,urgent,v2-pivot, eval-scheduled |
| Eval score | 0.82 |
| └ blocking impact | 0.93 |
| └ completeness | 0.85 |
| └ constraint fidelity | 0.85 |
| └ coordination overhead | 0.90 |
| └ correctness | 0.80 |
| └ downstream usability | 0.82 |
| └ efficiency | 0.85 |
| └ intent fidelity | 0.90 |
| └ style adherence | 0.88 |
Description
Description
The v2 pivot (2026-04-24) replaced the KRAS cancer spine with PHR discovery in pangenomes. The Google.org submission is due TODAY. The risk: v1 framing surviving in any artifact a reviewer or visitor could see while the v2 application is being evaluated.
The submission-readiness-audit task is already auditing workgraph_google_application_FINAL_v2.md itself. This task is the COMPANION sweep — everywhere ELSE.
What to do
Grep the entire repo and any tracked sibling artifacts for v1-spine terms. Flag every hit with file path + line number + recommended action (delete / rewrite / archive / leave-with-note).
Search terms (case-insensitive)
- KRAS, KRAS G12D, G12D
- MRTX1133
- pancreatic cancer, pancreatic, oncology, tumor
- Boltz, RFdiffusion, RFDiffusion, DiffDock
- 'drug discovery competition', 'deep research competition' (the v1 demo framing)
- 'wrote this proposal with WorkGraph', 'drafted using WorkGraph', 'incorporated using WorkGraph' (the cut recursion claim)
- 'CRISPRme provided off-target analysis for the BCL11A guide used in Casgevy' (and close paraphrases — the inaccurate Casgevy framing)
- 'PI' / 'lead PI' (style rule violation, often correlates with old text)
Where to look
- Top-level:
workgraph_extended_outline.md(v1 — should exist but not be cited from canonical paths) - Top-level:
workgraph_google_application_FINAL.md(v1) - Canonical v2 docs:
workgraph_extended_outline_v2.md,STATE.md,README*.md,CLAUDE.md - Landing page:
index.html,website/index.html— CLAUDE.md flags 'What we've built' card #3 specifically .wg/notes and any tracking docs- Any other markdown / html in the tree
Classification of each hit
- OK (historical context) — explicit acknowledgment that the item is v1/superseded (e.g. STATE.md §5 decision log, CLAUDE.md 'Superseded' section). Leave alone.
- MUST REMOVE BEFORE PUBLIC TRAFFIC — anything a Google.org reviewer or website visitor could land on that still presents v1 as current. Top of list: landing page card #3.
- PUBLIC REPO HYGIENE —
poietic-pbc/deep-research-competition(public, stale). Recommend: archive the repo OR add a top-of-README banner pointing to v2 PHR demo. Do not delete (loss of git history). - CAN WAIT — internal-only files where v1 references won't be seen by reviewers in the next 24h.
Output
Write .wg/v1-residue-sweep-20260501.md with sections matching the four classifications above. For each MUST REMOVE item, include the exact replacement text or 'delete this block'. wg log a one-line summary (e.g. 'N must-fix, M public-hygiene, K can-wait').
Validation
- Every search term grepped across repo
- Landing page card #3 explicitly checked
- poietic-pbc/deep-research-competition repo status noted
- Each hit classified into one of the four buckets
- MUST REMOVE items have concrete fix text, not just a flag
-
Output file written to
.wg/v1-residue-sweep-YYYYMMDD.md
Depends on
Required by
Log
- 2026-05-01T18:39:15.309283330+00:00 Task paused
- 2026-05-01T18:39:50.385441910+00:00 Task published
- 2026-05-01T18:40:07.539969767+00:00 Lightweight assignment: agent=Default Evaluator (31847164), exec_mode=full, context_scope=task, reason=Evaluator (0.97 score, 20 tasks) best suited for systematic audit/classification: must grep entire repo, categorize v1 references into 4 judgment buckets, write structured report with concrete fixes, and log summary—fundamentally an evaluation task requiring quality assessment and decision-making
- 2026-05-01T18:40:07.719863901+00:00 Spawned by coordinator --executor claude --model opus
- 2026-05-01T18:40:18.541554086+00:00 Starting v1 residue sweep — checking messages, then beginning grep across repo
- 2026-05-01T18:40:53.602729425+00:00 Initial grep complete. Now checking landing page card #3 and full content of files with omitted long lines
- 2026-05-01T18:43:00.547816093+00:00 Grep complete. Found: index.html CTA + recursion-claim cards (MUST FIX), v1 outline+app retained as superseded (OK), public repo deep-research-competition not formally archived on GitHub + v1 README + v1 description (PUBLIC HYGIENE), STATE.md $1c$2$3 still pointers to v1 (CAN WAIT). Now writing report.
- 2026-05-01T18:47:52.133359232+00:00 Report at .wg/v1-residue-sweep-20260501.md complete: 4 must-fix in index.html (CTA + 3 What-we-built cards), 1 public-hygiene (deep-research-competition not formally archived; v1 description+README), 5 can-wait, 8 OK groups. Drop-in fix text included for all 4 must-fix items.
- 2026-05-01T18:48:14.791192664+00:00 Validated: every search term grepped, card #3 checked (already PHR-swapped, CLAUDE.md note is stale), public KRAS repo status noted via gh repo view, all hits classified into 4 buckets, must-fix items have drop-in fix text, output file at .wg/v1-residue-sweep-20260501.md. Marking done.
- 2026-05-01T18:48:14.886638892+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
- 2026-05-01T18:49:46.527466672+00:00 PendingEval → Done (evaluator passed; downstream unblocks)