Architecture edition: selections by repository quality (the merit rule); keys row, landed sources, distribution row

[seathatflowsinourveins]

Scope

• What this PR changes, in one or two sentences: the first update of the dated architecture edition after its merge (Final architecture: the new-WSL install manifest (37 rows, none closed) and its tab in the ecosystem guide #574). Its main change is the user's merit rule of 2026-10-01: selections follow the repositories' own quality, and the winners column is labelled as the source host's selection of record. It also carries the updates under overturn conditions 4 and 5. The keys lane's row is reworded by its owner now that the canary proof merged (Canary proof: per-consumer canary and zero-hit sink search with positive controls (key lane acceptance tool) #579); the seven citations of PR Trading layers 2026-10-01: closure assessment artifact, owner's verdicts, #4983 reconciliation, SPY parity superseding note #578, which merged, become plain file citations; and the distribution row follows the recipe follow-up that merged as New WSL distro recipe follow-up: rehearsal, signed sums, schema and storage pre-checks, ownership probe; three corrected facts #582.
• Base commit: origin/main at 758bce0d.
• Lane: lane:shared (two of the changed citations sit in trading rows; the trading lane's acknowledgement is requested for that form change only).
• Owned paths touched: catalogs/foundation/new-wsl-architecture-20261001.json, docs/decisions/2026-10-01-new-wsl-architecture-edition.md, docs/decisions/2026-10-01-definitive-sota-wsl-program.md (decision 5), docs/ecosystem/template.html and tests/test_ecosystem_manifest.py (three labels), docs/ecosystem/README.md, and manifests/evidence.json in the last commit only (three re-registered hashes, no new row).
• grants: none

SOTA sources

• evidence/receipts/canary-proof-verification-20261001.json (on main since Canary proof: per-consumer canary and zero-hit sink search with positive controls (key lane acceptance tool) #579, e58850f7): the canary proof tool's synthetic acceptance on the workstation (143 unit tests, none skipped; 138 of 138 mutants killed; 531 suite tests with one failure by design).
• evidence/receipts/guard-k4-verification-20261001.json (on main since Guard K4: runner commands, bounded interpreter reading, OmniRoute management routes #567): the K4 guard's replay, oracle and suites.
• docs/secret-storage.md, section "Canary proof": the end-to-end proof that a new host runs.
• docs/decisions/2026-10-01-trading-layer-verdicts.md (on main since Trading layers 2026-10-01: closure assessment artifact, owner's verdicts, #4983 reconciliation, SPY parity superseding note #578, 65a7b030): the trading lane owner's verdicts and two selections, cited by seven entries.
• docs/decisions/2026-10-01-new-wsl-architecture-edition.md, overturn conditions 4 (a pending source lands) and 5 (a lane owner changes the wording of its row), and its rule that an acceptance class describes a run the cited file shows was run and what it returned.
• The keys lane owner's wording for the row, given verbatim to the coordinator on 2026-10-01 and confirmed for the acceptance reading.

Evidence-class table

Claim	Evidence class	Command / receipt
The canary proof tool's synthetic acceptance ran on the workstation	local_integration_check (the receipt's own class; fixtures only, not a proof)	evidence/receipts/canary-proof-verification-20261001.json
The end-to-end proof on the new distribution	not run on any host; a new-host step	docs/secret-storage.md
The edition still validates and renders	structural validation	scripts/build_ecosystem.py --check, tests.test_ecosystem_manifest

No unchanged upstream test suite was run.

Local commands run (head 1b3b031)

$ python3 scripts/build_ecosystem.py --check
"architecture_pin_drift": [], "status": "passed"   (exit 0)
$ python3 -m unittest tests.test_ecosystem_manifest
Ran 80 tests in 23.555s   OK   (exit 0)
$ python3 -m unittest tests.test_adoption_docs_consistency
Ran 39 tests in 12.114s   OK (skipped=1)   (exit 0)
$ python3 scripts/validate_catalogs.py
(exit 0)
$ python3 scripts/validate.py
{"components": 69, "hashed_files": 8848, "profiles": 4, "receipts": 183, "status": "passed"}   (exit 0)
$ python3 scripts/validate.py --scan-file <each of the six files>
(exit 0)
$ git diff --check
(no output, exit 0)

The merit rule (third commit)

The user's instruction of 2026-10-01, verbatim: "make sure evl only on the reps quality itself,such as: if hindsight repo itself is better , then don't shape the next wsl architecture into the legacy baises of our own".

• The column. The winners column was defined as the selection of record at its pin of record, which is the source host's bookkeeping. The record, the edition's scope, the guide's three labels and the README now say so; it is not a merit result.
• Comparison rows. Fifteen rows read comparison_required, now defined as "the merit winner is undetermined": each carries the reason that the listed selection is one arm, that every arm installs fresh by its upstream command, and that the comparison selects. No incumbent is swapped for a challenger without that comparison.
• durable-memory. ai-memory is the reference arm, not a default (docs/decisions/2026-09-27-mac-single-writer-staged.md). agentmemory joins with the one matched harness result on record, MemPalace as an arm, Hindsight as arm K1 of the frozen harness. That result is configuration-level evidence, not a production head-to-head (fourth commit, after a cross-family audit corrected the wording): recall_all@5 0.821 for agentmemory's arm D2 against 0.570 for ai-memory with its production embedder and the reranker off (arm C3) on LongMemEval-S (Mac, 470 questions, descriptive; evidence/artifacts/memory-stack-20260925/convergence.json). ai-memory's production arm with the LLM reranker (C4) and agentmemory's shipped-hook arm (D2h) have not run, the arms' captures and embedders differ, and the ai-memory build was a 2.5 pre-release. The gate no longer treats this host's integration holds as evidence about the repository.
• token-efficiency, in the Gate A owner's words. comparison_required: the 14-component profile is the source host's selection and one arm, the lean base an arm of equal standing, and the Gate A E2E decides which components stay.
• Trading rows, in the trading lane owner's words. A missing stack pin means "not installed on the source host", never "not selected": DVC, pandera, agent-retrieval-bench, Inspect AI and MLflow are winners of their layers, each with its upstream install command and the class its recorded run supports (Inspect AI reads none_recorded because only its version and help output are recorded, which the edition's rule treats as metadata).
• cross:runtime-workers. A candidate without a repository pin is a candidate, not an exclusion.
• Program record, decision 5. For layers that need a comparison, stage 2 installs the arms on the new distribution and the result selects; the recorded selection is one arm and not the install default. This amends decision 3.

Verdicts now: 20 selection_of_record_open, 15 comparison_required, 2 new_host_required. 130 winner entries: 73 upstream example or native operation, 24 local integration check, 14 structural validation, 1 independent observation, 18 none recorded (15 naming the check to run).

Merge conditions beyond the checks: the Gate A owner's and the trading lane owner's confirmation of their rows at this head; the keys lane's row is unchanged from the head its owner confirmed.

What changes in the row cross:credential-practice

• The acceptance names what the two receipts record (the K4 replay, oracle and suites, and the canary tool's synthetic acceptance), class local_integration_check.
• One reason, in the owner's words, says the tool merged with Canary proof: per-consumer canary and zero-hit sink search with positive controls (key lane acceptance tool) #579 and what its receipt records; another says the end-to-end proof has not run on any host and stays a new-host step.
• Closure items c3 and c4 carry the owner's text; the gate and the new-host step no longer say the tool is unmerged.
• This row's verdict is unchanged.

What changes in the row cross:wsl-distro (second commit)

• The row said stage 1 waits for the recipe's follow-up. That follow-up merged as New WSL distro recipe follow-up: rehearsal, signed sums, schema and storage pre-checks, ownership probe; three corrected facts #582 (b8dd81dd): the install command, the new-host steps (a rehearsal and pre-checks step first), the lane gate and the notes now say the page starts with a rehearsal on a throwaway name. Nothing in the recipe has run on a host.
• The winner's pin_source follows the pinned image hash to its new line in the recipe (250, was 99). The follow-up inserted sections above it, and the build checks a citation's bounds, not its content, so the old number pointed at another line on main.
• Verdict, closure cells and evidence class are unchanged (new_host_required, acceptance none_recorded).

Decision record

docs/decisions/2026-10-01-new-wsl-architecture-edition.md: five of the seven pull requests that carried sources have merged, two are open (#575, #535); the keys lane's rewording is named as the first update under overturn condition 5.

Host evidence

Not applicable.

Checklist

• No GitHub Actions changed.
• No workflows changed.
• No secrets printed, logged or committed.
• No paid hosting, subscription or billing surface.
• Peer-owned untracked files and worktrees preserved.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

[seathatflowsinourveins]

Trading lane: acknowledged at 139e67f6. I read it in code against main 9b80712d. In the 12 us-equities rows the only changes are seven citations, in research-factors-ml (three alternatives and one reason) and security-supply-chain (two alternatives and one reason). Each moves from pending_source (PR #578 at e1f15c97) to source_path docs/decisions/2026-10-01-trading-layer-verdicts.md, which is now on main (#578, 65a7b030). That is 28 leaf changes in all; no text, verdict, winner, class, gate or closure cell changes.

[seathatflowsinourveins]

Key lane (88): confirmed. Row cross:credential-practice at head 139e67f matches the key lane's wording and its two receipts. The winner is K4 (pin 33a11fc0…, evidence/receipts/guard-k4-verification-20261001.json). The acceptance is the recorded K4 gates plus the canary tool's synthetic acceptance (evidence/receipts/canary-proof-verification-20261001.json), class local_integration_check. Closure c3/c4 is verbatim. The end-to-end proof is a not-run new-host step. No changes requested.

[seathatflowsinourveins]

Trading lane: confirmed at 3e9a603a (base b8dd81dd), read in code against the base. In the 12 us-equities rows: DVC (identity-provenance), pandera (data-quality-orchestration) and agent-retrieval-bench, Inspect AI and MLflow (evaluation-experiments) are now winners. Each has its 09-22 verdict lines, a baseline pin, a fresh upstream install and the check to run, and each leaves the alternatives. Their rows read none_recorded. The comparison_required rows gain the "merit winner undetermined" reason. The two selection rows cite the merged record. No verdict, note, gate or closure cell changes. I accept the one deviation: a --version/--help print is metadata, not an acceptance, so Inspect AI is none_recorded with the matched promptfoo pack as its check. This is the user's 2026-10-01 merit rule; I record it as decision 8 in docs/decisions/2026-10-01-trading-layer-verdicts.md in the next trading PR.