Skip to content

Lead notes

The lead's working state, rewritten to the present before each reset. Last rewritten 2026-11-16.

First seating (2026-11-03): the handout's first work

  1. Old triage waker retired: done, 8c010c0, after triage's message that watch.inbox fired on the real S125–S127 edit.
  2. The seats in the docs: done, 17b9a3a (RELEASE_CYCLE § Seats, the CLAUDE.md doc map, seats/ in not_in_nav, which the seat PRs had left red).
  3. The id scripts: kept at seating; triage then moved to the kit's copies (150d916). Claim an RMn with the kit's classes/triage/item-next.py --note.
  4. Trust register (docs/TRUST.md): the maintainer, 2026-11-08: "lowprio, PM to reask after 1.0". Not started; handed to the pm to ask again after the 1.0 cut.
  5. Scope files are the pm's: noted. docs/ROADMAP_0_8.md, docs/ROADMAP_1_0.md, docs/proposals/.
  6. The kit pin: this seat moves it (bin/move-to-tag.py), on a release announcement that needs it or a fix waited on.

Standing rules from the person

  • HF uploads on the pm's sign-off (the maintainer, 2026-11-07): "you have HF upload rights on PM's signoff (basically, two-agent consensus instead of human)". Reference snapshots only, never a registry module or PyPI. Ask the pm with lane, label, dry-run file count and reason; quote its yes. RELEASE_CYCLE § Patches.
  • No cut with an unpublished lane (same day): JUST_DNA_NETWORK_TESTS=1 uv run pytest enricher/tests/test_prerelease_gates.py green at the cut commit (RM413).
  • Long jobs run niced (the maintainer, 2026-11-05: "memorize nice setting for the seat jobs"): every long command the lead or its agents start (full pytest suites, enrich/rebuild batches) runs under nice -n 19. Per-item batches run at -P1, at most -P2; pytest-xdist at most -n 4. Write this into every builder and merge brief. When the person or the sysadmin seat reports lag, renice -n 19 -g PGID the running groups at once. Why: a -P4 enrich batch beside a full suite drove load to 65 on 16 threads, and the person's remote desktop dropped frames.
  • Registry wakes (2026-11-04, amended 2026-11-05): just-dna-registry/liaison and just-dna-registry/triage wake our triage, and the registry liaison wakes the lead (re-added 275673f4). The two "false wakes" of 2026-11-04 were not wakes: a post to every seat never wakes through wake_from (the kit tooling seat, on winternewt/agent-seat-blueprints#269). The person's Zulip account for DMs is id 1213223.

  • Upstream adoption (2026-11-04): the liaison decides adoption blockers and hands the lead an adoption map. Pin only when a new major is proven to break something, until support is integrated. If the liaison is unavailable and work must move, the lead may pin as a safety. (RM341's cap was reverted under this rule.) Sent to the seat-builder for LEAD.md and LIAISON.md.

In flight

  • 0.8.9 cut 2026-11-18, awaiting the publish: cut commit 13b3be36 (format, compiler, enricher 0.8.9; format moves for the release record alone), merged into main (48aa8ae3, then RM453 in 19a7cdb0, which is 0.8.10's). Dists built from 13b3be36 into the root dist/, twine check passed. After the publish: tag v0.8.9 on 13b3be36, GitHub release from scratchpad/release_notes_0_8_9.md, README.md:94 and CLAUDE.md:84 to name 0.8.9. Branches cut/v0.8.9 (superseded) and land/v0.8.9 are the person's to delete.
  • Lanes behind source (liaison, 2026-11-17): done 2026-11-18. clinvar_2026-11-15 and mitomap_2026-11-09+2026-11-03 published by the person (the classifier refused the seat, kit #529), pulled here; mitomap_miss rebuilt, the old copy moved to /data/just-dna-cache/mitomap_miss.before-2026-11-18 (the person's to delete).
  • 0.8.8 released 2026-11-17: enricher 0.8.8 on PyPI (02:17Z), tag v0.8.8 on 655f1ebc, GitHub release, README and CLAUDE.md name it.
  • main pushed 2026-11-04 by the pm at the person's word (…150d916), the lead's commits included. Auto-mode rule 11 (kit v0.1.117) is what lets this seat push its own; the person pastes it.
  • Leftover local branches build/RM341 and build/RM343: cherry-picked onto main, so -d refuses them; -D is the person's.

  • 0.9 (2026-11-16): every decision made (ff09d949, the pm's answers to the critic; RM431 ask 4 local-only, never published, decided the same evening). Build plan: /tmp/claude-1000/-data-sources-just-dna-format/cb39872f-9ac8-49ad-96ce-7354acea34da/scratchpad/BUILD_PLAN_0_9.md (scratch; this line is the record): branch 0.9 off main; wave 0 LookupClients leaf module + parse_release; wave 1 RM443, RM412, RM404, RM437 in parallel; wave 2 RM430; wave 3 RM431 asks 6, 5, then 4. Then one four-leaf round (two at most), the pm sorts, the cut. unattended: starts when the person is away.

  • 0.9 progress (2026-11-18): branch 0.9 at 6a12f7e1, pushed (worktree .worktrees/09 on disk, locked). Every wave and both follow-ups merged (RM431 fu 4d5fe26a, RM430 fu 15668c1e); review round 1 (four-leaf, 033b7213): 11 classes, no blocker, all fixed (ea3dcf07) plus the pm's flip of item 3 (6a12f7e1, an unreadable pull record moved aside and re-pulled, the pm's cf12e864). Suite 8119 passed. Next, quota permitting (7-day 93%, hold 90%; a wait-quota is registered): review round 2, the last, brief at scratchpad/review_0_9_round2_brief.md; then its fixes plus mitomap_miss stale_reason and a walk over derived lanes (pm 757f5e32); then the cut. Owed at the cut (scratchpad/BUILD_PLAN_0_9.md): enricher floor on format 0.9; INTEGRATION_0_9 lines for stale in LANE_STATES, proxy-on/HF-off gene lanes could-not-ask (registry S36), a host closes a shared holder after its last lookup; CHANGELOG line for cpic's trim; keep just_dna_enricher.key_compare.KEY_COMPARES stable (the registry reads it).

Open decisions

None open.

reflect after 2026-11-24: classes/base/BASE.md as of v0.1.231 (a done-claim is checked from where the person looks; read 2026-11-17, no concern). reflect after 2026-11-23: classes/base/BASE.md as of v0.1.224 (only a real fork goes to the person; read as its adversary 2026-11-16, no concern: the lead's only standing wait is the PyPI publish, a step that is the person's, not a choice).

Plan

  • 0.8.0 released, 2026-11-05. Published on PyPI by the person; tag v0.8.0 on e97339a3; GitHub release up; docs refreshed (INTEGRATION_0_7 archived) and merged. RELEASE_CYCLE step 10 now makes tag, GitHub release and docs refresh the lead's after every publish (auto-mode rule 13 covers the release, kit #283).
  • 0.8.1 released, 2026-11-07 (enricher only: RM397, RM398, RM400, RM401, RM403, RM407). Tag v0.8.1 on 6330f908, pushed. The GitHub release is not up (notes were in a session scratchpad, release_notes_0_8_1.md); auto-mode rule 13 is pasted now (2026-11-08), so the lead can create it.
  • 0.8.2 released, 2026-11-08 (all three packages; RM408, RM410, RM411, RM413-RM417). On PyPI (the maintainer); tag v0.8.2 on b98f7e9c, pushed. gnomAD constraint 4.1.2 published to just-dna-seq/gnomad_constraint (rebuilt with 0.8.2 from data/input/gnomad/, parquet byte-identical to the pre-bump build; the hub answers gnomad_v4.1.2_constraint). GitHub release up (the maintainer granted that one command; rule 13 was pasted afterwards, automode.py --check clean on 2026-11-08). Docs refresh merged (9c37d7db). Liaison's ledger updated (fe38e101). CLAUDE.md's @constraint-two-releases headline changed on the maintainer's yes (59491cbf). RELEASE_CYCLE step 10 complete.
  • 0.8.3 released, 2026-11-10 (all three packages; RM418, RM420, RM409, RM421, RM422, RM423). On PyPI (the maintainer); tag v0.8.3 on 725eb2bb, pushed; GitHub release up; README and the CLAUDE.md doc map name 0.8.3 (67a514b8). RELEASE_CYCLE step 10 complete. RM424 parked (a minor). The 0.8.3 worktrees and branches were removed by the maintainer, 2026-11-10. RM409's interrupt test failed once at load 30–48 (1.0 s bound): file if it recurs.
  • 0.8.4 released, 2026-11-11 (all three packages; RM425, RM426, RM427, RM428). On PyPI (the maintainer, 01:14 UTC); tag v0.8.4 on 36cbc20f, pushed; GitHub release up (notes corrected: one public attribute, GwasCatalogClient.size_refusals, also aec2b744 in the CHANGELOG); README and the CLAUDE.md doc map name 0.8.4 (9c4cf9cc). RELEASE_CYCLE step 10 complete. Desk #258 withdrawn (done); #259's default applied (307a5ae3). wt-cut-084 unlocked; cut/v0.8.4 handed to the maintainer for -D (cherry-picked, git cherry shows it on main).
  • 0.8.5 released, 2026-11-12 (all three; RM432, RM434, RM436, RM438). On PyPI 22:06 UTC (the maintainer); tag v0.8.5 on c036c700, pushed; GitHub release up; README and the CLAUDE.md doc map name 0.8.5 (a5308f60). RELEASE_CYCLE step 10 complete. The priming takes the package-index reader and the branch sweep on publish (6fcf74a5, desk #271, the maintainer: "treat builder as my word"); watcher pypi running. Merged worktree-agent-* branches deleted. For the maintainer: cut/v0.8.4, cut/v0.8.5, build/RM438 need -D (cherry-picked, on main by git cherry) after their worktrees go. RM437's doc half, RM439, RM440 built 2026-11-13.
  • 0.8.6 released, 2026-11-13 (all three; RM439, RM440, RM437's doc half). On PyPI 14:24 UTC (the maintainer); tag v0.8.6 on 2f03ab4a, pushed; GitHub release up; README and the CLAUDE.md doc map name 0.8.6 (48d287ce). RELEASE_CYCLE step 10 complete. The publish sweep took the merged cut and build branches; cut/v0.8.6 stays, held by its worktree wt-cut-086 (unlocked), for the maintainer to remove.
  • 0.8.7 released, 2026-11-15 (enricher only; RM442, RM444; sized by the pm). On PyPI 16:01 UTC (the maintainer); format and compiler stay 0.8.6. Tag v0.8.7 on 3b65ab7a, pushed; GitHub release up; README and the CLAUDE.md doc map name 0.8.7. RELEASE_CYCLE step 10 complete. test_first_seen's release list now takes every package's version (an enricher-only cut left format's on the old tag). The AVI interrupt test's load flake recurred and is fixed (d8c0e0f2, 10 s of work against a 4 s bound). cut/v0.8.7 and worktree wt-cut-087 (unlocked) are the maintainer's to remove.
  • 0.8.8 cut, 2026-11-16, unpublished (enricher only; RM445 alone; sized by the pm to cut at once). Cut commit 655f1ebc on main, pushed; format and compiler stay 0.8.6. dist/ holds the enricher 0.8.8 wheel and sdist, twine-checked (scripts/build-dists.sh needs bash, not sh). Gates: full suite 7446 passed, network gates 26, strict docs build green. After the maintainer publishes: tag v0.8.8 on 655f1ebc, GitHub release from the scratchpad's release_notes_0_8_8.md, README and the CLAUDE.md doc map name 0.8.8. For the maintainer to remove: worktrees wt-rm445, wt-cut-088 (unlocked); build/RM445, land/RM445 need -D (cherry-picked onto main); cut/v0.8.8 takes -d once its worktree goes.
  • Next patch on main, uncut (after the 0.8.4 cut commit): RM432 (ac3f1bcc, the lead's, 2026-11-11: two snapshot readers raise FileNotFoundError; a behaviour change for a direct caller), the GWAS docstring's 15/s budget (7a88e10f), RELEASE_CYCLE's floor rule (307a5ae3). Lesson from RM428: it added the public attribute GwasCatalogClient.size_refusals on main, which patch scope forbids; it shipped in the 0.8.4 cut and stays (the pm), and RM429 is the minor that surfaces it. A builder's own new attribute gets a leading underscore on main.
  • Carried by the 0.8.4 cut (was the next patch after 0.8.3): RM425, RM426, RM427 (e7fa81d6, triage's: the ClinVar skip names cache pull --only clinvar, and a walk checks every quoted command), the fence test's hub-registry fix (11f86d74) and the liaison's first weekly dependency bump (867e8462, 4e30ef46; huggingface-hub 2.4.0, duckdb held at 1.5.5).
  • Shipped in 0.8.3 (was the next patch): RM409 shipped 2026-11-08 (e0ceaa28: Ctrl-C stops the Open Targets mirror and the AVI build within a chunk); RM421 filed from it, sized by the pm and shipped 2026-11-08 (83182f88: a 120 s default timeout in stream_to_file, an AST walk refusing timeout=None). RM420 shipped 2026-11-08 (dd900ff3: cache pull/prepare follow the hub's release over a pulled cache; a local build is left alone; the pm sized it ahead of everything). The 0.8.2 GitHub notes and INTEGRATION_0_8 § 4 carry its correction. RM418 shipped 2026-11-08 (d28c74b5, one NCBI gate per process via net._service_gate). RM419 filed (the same class beyond NCBI); the pm parked it as a patch (88bda8eb): it unparks on a reported rate-limit refusal with two of our clients in one process, or the registry running enrichments as threads, then the Ensembl pair first.
  • Cache: /data/just-dna-cache/ensembl_variations is the split release. The old ensembl_variations.pre-0.8 was deleted by the maintainer, 2026-11-08. The just-dna-lite side still to be told.
  • Desk: none open. #210 and #259 withdrawn 2026-11-14 as default applied: (no downstream list; the release rule of 307a5ae3), #211 as moot: (5 local branches left, the publish sweep takes merged ones).
  • Not done, the person's: the worktree .claude/worktrees/lead-RM372/data/rm372/fence_check; merge commit cb6be8a4 has no Co-Authored-By trailer.
  • Records gap: no release records for 0.7.1 or 0.7.2.
  • Kit v0.1.136: in this shared clone, commit by path (git commit --only -- <paths>).
  • Parked: RM388, RM390, RM396 and round-3 findings without a number (ROADMAP.md). 0.9: RM385.