docs: consolidate research convergence and close practice workstreams - #1011

Scope

Record the research-convergence practice, its pending discovery/refutation dependencies and supported gatherer/worker routes; link the canonical CC review ruling and #1008 terminal-state protocol, with explicit owner receipts and reopen conditions. The existing native-stack-research template points ecosystem convergence to the recorded process and unresolved delta. Actual Claude/Codex row publication stays with cc-native-practice and the co-op, with concrete handoff and receipt requirements.

• Base commit: f1f6c0c20e214f1cee9aba3c5f0c1430d66e75ed.
• Lane: lane:foundation.
• Owned paths: docs/decisions/2026-10-11-research-convergence-practice.md, tracked evidence/artifacts/new-wsl-install-plan-20261002/config/research-harnesses-skill.md, tests/test_research_convergence_practice.py, and their native evidence-manifest registrations.

SOTA sources

• CC program-final-resolution-20261011.md WS2/WS3 (2026-10-11 14:00Z) and ruling-pr-wave-convergence-20261011.md, rules 1–9 with its 13:43:25Z addendum; full retained authority captures accompany the read packet.

• assafelovic/gpt-researcher v3.7.0, commit 0957c301ed06c2a5857b834358c7227c739041d4, cli.py:336–359; exact clean installed-source Git object captured.

• bytedance/deer-flow v2.1.0, commit 345f08be00c8a9495079b732a39b46aa9af1584e, embedded/client and native-headless sections README.md:1643–1663,1826–1846; exact clean installed-source Git object captured.

• OpenHands/software-agent-sdk v1.53.0, commit 54daf056bd863bb46f922a2fe9324dd736b37ff6, Makefile:34–37 and standalone example:9–29; exact Git objects captured separately from locally modified installed source.

• openai/codex rust-v0.160.0, existing source-bound Python worker examples/omniroute-codex-sdk/worker.py:5,41–42, and official SDK lifecycle documentation. All four exact sources are registered and synced in MIRRORS.json/MIRRORS.lock.json; the Codex SDK ref has its dedicated openai/codex@a956835d-sdk0160 mirror. Pinned objects were checked with lazy fetch disabled. No new route/source equality or native task is claimed.

• Landed #1008 terminal-state authority, squash 569bbc59db2b5f7c8a932357726efb7acca90b91, lines 69–122; canonical no-dates regression matches the landed test exactly.

Evidence-class table

Local commands run

python3 -B -m unittest tests.test_research_convergence_practice tests.test_resolution_protocol_docs tests.test_evidence_manifest tests.test_new_wsl_definitive_defaults tests.test_native_maintenance tests.test_practice_references
rc0;204 tests,3 skips

python3 scripts/validate.py
rc0;70 components,11676 hashed files,4 profiles,239 receipts

python3 scripts/evidence_manifest.py --check
git diff --check
rc0

gitleaks dir <owned-private-read-packet> --redact=100 --no-banner
rc0;no leaks found

The correction adds the canonical no-dates regex regression: it fails against the exact record at f1dfd97 with “dated comparison” and passes after the fix. Existing native skill/install contracts and repository integrity checks also pass. No new dependency, paid research, model job, live config update or landing occurred. The authorized landing rebase is recorded below. #993/#995 keep their current-head review-wave gates; #1008 and #1007 have landed; W6 qualification remains governed by the protocol's evidence joins.

Decision record

docs/decisions/2026-10-11-research-convergence-practice.md. WS3 becomes CLOSED only after accepted publication, owner slot/qualification receipts (or bounded no-matching-row receipts) and named dependencies; #978 landed as squash 333ec47; Discovery still awaits WS1 categories and their qualified output, and refutation enforcement awaits #993. The CC ruling and #1008 protocol remain the single rule/state authorities; REOPEN is a new practice-tool release or measured practice failure. FINAL architecture-layer qualification remains the distinct W6 evidence join.

Host evidence

No host receipt or platform-status change. Installed source/package observations are source review only; historical task and E2E scopes remain qualified by their original receipts.

Follow-ups

• Supply the cited immutable source/correction packet to readers, including retained authority, measurements, primary sources and source manifests; preserve original-to-masked identities (Astra fix1/fix2 P3-1).

• Preserve correction-specific primary and fail-first regression evidence separately from historical observations (Astra fix1/fix2 P3-2).

• Complete owner publication and dependency qualification before WS3 CLOSED: cc-native-practice and the co-op return exact slot IDs/refs, installed-template/launcher hashes and acceptance receipts, or bounded no-matching-row receipts; join CC acceptance and named dependencies (Astra fix1/fix2 P3-3).

• Retain the historical manifest-locator follow-up: bind 2026-10-05-dagu-mise-pin-move.md:81,91,227 to original evidence revisions or stable paths, preserving frozen evidence (Astra fix1/fix2 P3-4).

• Retain pre-existing machine-name labels as counted historical attribution; unchanged sites do not block this PR (Astra fix1/fix2 P3-5).

• Keep adjacent GPT Researcher/DeerFlow preflight, native exit/result and SDK startup/cleanup gates qualified during publication; every new installed route needs separate acceptance (Astra fix1/fix2 P3-6).

• Align the containment test's environment probe with its runner (Astra fix1/fix2 P3-7).

• Check the newly merged model-currency integration in a genuine worktree (Astra fix1/fix2 P3-8).

• Runtime-workers owns the separately routed worker-version/port drift (Claude F13); this record changes no deployed worker or gateway.

• Name a release channel for repository-script reopen conditions and concrete native manifest/recording tools (Claude FU-2).

• Source the five-report-reference predicate and align it with a retained repository oracle (Claude FU-3).

• Finish remaining body/template spacing and template wrapping on their next touch; the CC's specifically folded record tokens are corrected (Claude FU-4).

• Correct the layer protocol reopening locator to the Decision section; 569bbc5 has no separate reopen section (Claude FU-5).

• Extend regression coverage beyond N2 when that wider test work is undertaken; the current canonical no-dates regression covers N2 only (CC round-2 follow-up).

• Obtain genuine-worktree coverage for the archive-sensitive token grader (Astra fix2 P3-9).

• Reconcile the installed host guard copy with the intended source revision (Astra fix2 P3-10).

• Align the tracked research template's remaining historical Converge 373 public stars and schedule weekly review intake #978-pending wording with the landed intake status; the same-claim question is reported to the CC and the template is unchanged under the enumerated carry scope.

Checklist

• Required SOTA section and pinned primary source references included.
• Native registrations, validator, existing skill/reference tests and diff checks pass.
• Private source/authority captures and owner handoff retained for the CC read.
• Credential values, Codex configuration and peer-owned worktrees/catalogs preserved.
• New runtime execution and model/E2E acceptance are not claimed.
• CC review and same-head Claude/GPT/Codex wave complete; later fixes wait for all blockers or explicit CC release.

First correction — released complete review wave

Astra ACK at 0be7728 adds no blockers. CC-ruled B1 and F1–F8 are folded in ONE forward commit without rebase: pending #978/WS1 Discovery and #993 enforcement, canonical ruling/#1008 authority pointers, registered mirrors, tool/pin/failing-measurement reopen rows, exact known research-harnesses and skills-agent-docs identifiers plus bounded no-matching-row receipts, reordered skill insertion, and canonical catalogs/landscape/research-state.json with layer ID. The co-op’s research-queue.tsv is intake. P3-1’s extra paper-priority qualifier disappears with the rule restatement; P3-2 through P3-6 stay in Follow-ups. Optional F9–F12 add no scope; F13 remains owner-routed.

Current correction and all previously missing source captures are retained in lanes/convergence-practice-work/pr1011-full-wave-fix-20261011/, including complete original WS3 and Claude-preparation packets with their immutable source seals. Attach the complete correction archive to the delta readers and verify extracted source/transport bindings. No new live tool acceptance or owner-row closure is inferred. ONE confirmation micro per family precedes landing.

Second correction — released round 2

Head ab594c950723f59e44fcaf15d7497d517eb7310b is ONE forward commit on f1dfd978a34f992a3ffe85fde0f9507c21163b2c, without rebase. It pins the landed #1008 authority, changes the comparison to one pending its named inputs, adds the exact canonical no-dates fail-first regression, distinguishes first mirror sync from the later resync, and corrects the requested run-together tokens. The Scope describes the recorded practice and canonical links; the stale paper-queue claim was removed. Astra’s fix1 ACK adds no blockers or research-queue entries. Correction evidence is retained in lanes/convergence-practice-work/pr1011-round2-commit-ready-20261011/; supply its sealed primary/regression packet to the delta readers. One delta micro per family remains required before landing; owner and dependency qualification remain pending.

Landing carry — released after Astra fix2 ACK

Prior enumerated carry head a5f0aa0ed8e9cff3eaa40c2c2496f7ec34b30b77 rebases the three existing commits onto main 77185991ec9627e59473df5b4d7a11ca87b0461d. Only the CC-enumerated record lines and native registry entries change: #978 is landed at 333ec47, only WS1 remains pending, the Discovery pin uses the landed squash, resync observations retain unchanged pins, and the SHA token is split. The public_starred.py blob is identical at the old and landed pins. The source check also observed resyncs at 23:14:43Z–23:14:44Z, later than the CC's 22:34Z snapshot. Astra fix2 ACK adds no blockers. The co-op released ONE exact-lease push on ab594c9; source changes remain limited to the enumerated carry. The CC carry check, green CI and landing remain pending.

Registry carry after #975 landed

Current head 7bb720fa7499e49bf20d85591c232f32288b36b3 carries the three existing commits onto main 86fa384421b1ca50134bd56d89cc72e0f7547816. Every non-registry range-diff row is = and all three owned source blobs are unchanged. Native registration retains all main entries. Full validation (11691 files), the native final-stack check and the three-commit leak scan pass; the ONE exact-lease push on a5f0aa0 succeeded. The CC's carry check, current-head CI and landing remain pending.