Mix since 2026-10-17f: venture deepenings (fee calculators, schedulers, rosters), research follow-ups, redteam and process runs. Nothing contradicts the owner's quotes in goals.md. Venture friction repeats one point: the top allocator candidates are already deep, so runs add a guide page or a fee-step to an existing venture. That is product value, not a copy of one shape, but watch it.
Instruction lines: CLAUDE.md 337, prompt.md 89, review/prompt-1.md 130 (at budget), research/PROJECTS.md 1067. No file grew since the last report.
Friction check: the latest research/CLAUDE.md reports say platform/ci/pyenv.py fails under PEP 668 and prints noise. Not reproduced on current main: it falls back to --break-system-packages and a venv, and stdout is only the path (PY=$(...) printed /usr/local/bin/python3). The reports predate that fix, so no edit.
Not fixed: PROJECTS.md (~1,070 lines) stays the biggest file runs read. It is parsed by swe-value/swe-ledger, so a trim needs its own run.
Audit: nothing auditable (recent outputs mirrored or doc-only).
Data: derive.py fetch got 25 of 39 PRs; derived/ regenerated.
Follow-up on 2026-10-17f: E34 has too few research runs since to judge.