Final architecture: the new-WSL install manifest (37 rows, none closed) and its tab in the ecosystem guide

[seathatflowsinourveins]

Scope

• What this PR changes, in one or two sentences: adds the "05 Final architecture" tab to the generated ecosystem guide (Evidence becomes 06) and its source, the dated new-WSL install manifest catalogs/foundation/new-wsl-architecture-20261001.json: 37 rows (20 foundation, 12 us-equities, 5 cross), each with the selection of record at its pin, the verdict, the five closure items with what is missing, the evidence class, reasons with sources, alternatives, upstream currency, the ordered new-host steps and the gates. None is closed.
• Base commit: b72c588f (edition facts read at 5597f9fa; Anti-pattern log: a historical SDK inventory is not the pin oracle for a live SDK (Codex runtime lane handoff) #565 changes no cited fact)
• Lane: lane:shared: the edition sits under a foundation path and carries the 12 trading rows, so it needs the trading lane's acknowledgement.
• Owned paths touched: scripts/build_ecosystem.py, docs/ecosystem/template.html, docs/ecosystem/README.md, tests/test_ecosystem_manifest.py, catalogs/foundation/new-wsl-architecture-20261001.json, docs/decisions/2026-10-01-new-wsl-architecture-edition.md, manifests/evidence.json (hot-file commit, last).

SOTA sources

• Reference implementation: this repository's token-topic loader and validator (scripts/build_ecosystem.py, token_topic_card) and template renderer (topicCard), reused for the new topic, including the single-inline-script CSP hash and safeHref links.
• Closure items: catalogs/landscape/research-state.json saturation.close_only_when; evidence classes: docs/acceptance-evidence-policy.md:26-33.
• Upstream currency: one releases read per winner on 2026-10-01, cited on each row (GitHub releases, PyPI, Hugging Face, poppler.freedesktop.org, postgresql.org, Launchpad).
• Harbor README (https://github.com/harbor-framework/harbor, uv tool install harbor, v0.23.0); nautechsystems/nautilus_trader issue #4983 and PR #5041 (both open, read 2026-10-01); the gitleaks README (feature complete); the Next.js v16.3.8 release page (https://github.com/vercel/next.js/releases/tag/v16.3.8, read 2026-10-01); Docker Engine documentation, rootless mode, prerequisites and troubleshooting (https://docs.docker.com/engine/security/rootless/, read 2026-10-01).
• Not on main, cited through their pull requests as structured pending_source entries: New WSL distro recipe: Canonical Ubuntu 24.04.5 image, cloud-init passwordless user, first boot proofs (definitive SOTA WSL, unit W2) #569 344a69ff (new-distro recipe), Receipt: Harbor token-tools E2E (288 trials), figures and hashes of the private originals #570 cd7db158 (Harbor receipt, blocks host_prerequisite_observations and catalog_protocol_metric), Trading research convergence 2026-09-26: 13 layers, GPT-6 live refutation, verbatim-checked votes (verdict-wave input) #358 b0eb7a11 (trading convergence record), Definitive SOTA WSL program record: foundation closure assessment (20 layers, none final for install), criterion decisions, program units #573 5adcc3a4 (program record and the foundation closure assessment), Qualify native runtime recipes and repair Sol routing and grader FTP dependency #535 1195e212 (runtime-worker roster), Review all replacement WSL architecture layers and retained gates #575 bbee2a8e (the Codex lane's second independent reviews of the 20 foundation layers and the theme crosswalk).

Evidence-class table

Claim	Evidence class	Command / receipt
The generator rejects each invalid edition contract (40 messages plus the close_only_when case)	local_integration	python3 -m unittest tests.test_ecosystem_manifest; failing first against the base generator: 41 failures, 5 errors
The tab renders the edition with no script errors	local_integration	the Node page harness (fixture test and the real page)
The build is byte-deterministic	local_integration	python3 scripts/build_ecosystem.py --check
Winner pins equal the stack pins; cited files exist and are tracked	local_integration	generator validation; the real-edition test (git ls-files)
Upstream currency per winner	source_review	release reads of 2026-10-01
No layer is closed; item statuses per row	source_review (model-authored closure assessments, one assessor and one adversarial refuter per layer; statuses vary by one or two cells between runs where a layer ran twice)	the foundation assessment in PR #573; the trading assessment with the trading lane owner
Token ratios on the layer protocol's own metric	native_proven on the workstation for the runs, script recomputation not independently reproduced	the Harbor receipt on PR #570, block catalog_protocol_metric
Hindsight's holds and AgentRelay's held automatic Codex PTY submission	reported by the production program, unpublished; carried as gates	PR #573's record, "Independent review of this record"

No unchanged upstream test was run; every new-host step is unexecuted.

Local commands run (head a93b3f9, on main 000aae7)

$ python3 -m unittest tests.test_ecosystem_manifest
Ran 80 tests in 23.833s   OK   (exit 0)
$ python3 -m unittest tests.test_adoption_docs_consistency tests.test_catalog_freshness_propose
Ran 119 tests in 69.898s   OK (skipped=1)   (exit 0)
$ python3 scripts/build_ecosystem.py --check
"architecture_pin_drift": [], "status": "passed"   (exit 0)
$ python3 scripts/validate_catalogs.py
(exit 0)
$ python3 scripts/validate.py
{"components": 69, "hashed_files": 8819, "profiles": 4, "receipts": 182, "status": "passed"}   (exit 0)
$ git diff --check
(no output, exit 0)

Review rounds

Round 1. The builder's edition was read in full by the coordinator and corrected (subordinate ids, the holds carried as gates, scheduling and supervision as comparison_required, the owners' verbatim corrections, each foundation row citing the Codex lane's review).

Round 2 (the review bot's three threads and an independent code review). The build now ties a row's evidence class to its winners (a none_recorded winner makes the row none_recorded; otherwise a class at least one winner carries), parses missing into its cN: segments, binds the five closure-item texts by sha256, lets a stack pin move pass as recorded drift, treats a landed pending source as a cited file, and keys layers on (catalog, layer_id). Winners carry a role in the owner's words.

Round 3 (an independent review of the evidence labels). A reviewer checked all 119 winner acceptance classes against their cited sources: 32 claimed more than the source records, 13 commands were only version or status prints and 6 could not be settled. The classes had been assigned from the kind of command, not from a recorded result. The edition now applies one rule, written into the record and the README: a class describes a run that the cited source, or one file it links, shows was run on a host and what it returned; a check that is only prescribed, planned, not run or failed is none_recorded, and it may still name the check to run on the new host. 55 entries changed. The lane owners re-cited their winners to the files that record the runs (the trading lane in comment 5930139395, the Gate A owner for its three rows, the keys lane for the K4 guard).

Coordinator's batch on top. The trading lane owner's selections of 2026-10-01 for the two layers that had none (research-factors-ml: EdgarTools and skfolio; security-supply-chain: Syft, Gitleaks and Grype; neither closes anything). cross:wsl-distro re-rated to new_host_required now that the recipe is on main, with stage 1 waiting for the recipe's follow-up. Every foundation and us-equities row cites the Codex lane's second independent review at PR #575's head 51cb79ba; the record names the difference between those reviews' selected sets and this edition's winners (20 of 32 layers) as a residual gap for the re-record pass. Citations of pull requests that have merged (#569, #570, #573, #358) are plain file citations; pending sources carry current heads (#575, #578, #535).

Verdicts: 20 selection_of_record_open, 14 comparison_required, 2 new_host_required, 1 provisional. None closed.
Winner acceptance entries: 125. 72 upstream example or native operation, 24 local integration check, 14 structural validation, 1 independent observation, 14 none recorded (11 naming the check to run).

Merge conditions beyond the checks: the trading lane owner re-read its 12 rows and confirmed them for merge with one correction, which this head carries (comment 5930788443); the Gate A owner confirmed the content of its three rows and repeats its script check at this head.

Open after merge:

• A cross-family read of the generator, template and test changes: queued for the gateway pool's reset; none has run.
• When Review all replacement WSL architecture layers and retained gates #575, Trading layers 2026-10-01: closure assessment artifact, owner's verdicts, #4983 reconciliation, SPY parity superseding note #578 or Qualify native runtime recipes and repair Sol routing and grader FTP dependency #535 merges, the build labels its sources as landed.

Decision record

docs/decisions/2026-10-01-new-wsl-architecture-edition.md

Host evidence

Not applicable: no file under evidence/hosts/ changes.

Checklist

• New/changed GitHub Actions are pinned to a full commit SHA with a version comment (none changed).
• New/changed workflows declare top-level permissions: contents: read (none changed).
• No secrets are printed, logged or committed; no new required secret was added without a documented owner.
• No new paid hosting, subscription or billing surface was introduced.
• Peer-owned untracked files and worktrees were preserved (not deleted, moved or overwritten).

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-01T08:56:03.121761Z	3792b41	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[seathatflowsinourveins]

Trading lane: acknowledgement of the 12 us-equities rows (head 3792b418)

Acknowledged, with the changes below. Applied verbatim, they make the 12 trading rows final for this edition and replace each provisional_wording note. Source: the owner's verdicts of 2026-10-01, written from the trading closure assessment (read at #358 4d11709c), with the pins spot-checked against main b72c588f.

Verdict values. I accept the mapping:

• comparison_required for the seven comparison_required layers;
• selection_of_record_open for the three on_requirement_change layers;
• no_selection for research-factors-ml and security-supply-chain.

Field changes

1. execution-broker, the adaptive-paper winner.
   • Set component_id to adaptive-paper.
   • Set pin to dca821cca85dce3647fa7b488d5a23fbe5b85d4a, with pin_source "main at Adaptive paper: book Alpaca FEE activities in ledger cash accounting (unblocks account 2's IOVA recovery) #559 (blueprints/us-equities/adaptive-paper)".
   • Reason: the current pin is the sha256 of runner.py alone, and Adaptive paper: book Alpaca FEE activities in ledger cash accounting (unblocks account 2's IOVA recovery) #559 also changed safety.py and transport.py, so a single file hash does not bind the engine. The units run from a frozen clone at that commit.
2. execution-broker, the first gate (upstream). This is the final #4983 text, written after reading IBKR local acceptance (steps 2-4) prep: awaiting the user's paper IB Gateway sign-in #280's diff and re-reading the issues at 08:45Z:

   nautilus_trader #4983 (open upstream, read 2026-10-01) is a stale v1.227.0 report for rc5. At tag v2.0.0rc5 (1b0a49d2) the execution engine denies an order only when handles_order_venue is false, and the IB execution client returns true. That is source reading, not an rc5 order observation; PR IBKR local acceptance (steps 2-4) prep: awaiting the user's paper IB Gateway sign-in #280 moves the report to rc5_reclassified_reports. The open rc5 obstacles are on the recovery and risk paths that acceptance steps 3-4 exercise: #5007, #5057 and #5060 (open, read 2026-10-01); #4946 is closed. The rc5 IBKR stock-order path stays unqualified until native paper execution.

3. market-data-reference, add a user_side gate:

   Paid market data for the c3 comparison: the Databento arm (a paid entitlement) and Massive history beyond the free tier are the user's purchases.

4. storage-compute, add a user_side gate:

   The comparison's licensed inputs need the same paid-data decision as market-data-reference.

5. agents-models-workers, add a lane gate:

   GPT-6 disputes the earns_it label of ai-memory and SocratiCode (PR Trading research convergence 2026-09-26: 13 layers, GPT-6 live refutation, verbatim-checked votes (verdict-wave input) #358); retention is provisional until the held-out comparison runs.

6. evidence_class changes:
   • research-factors-ml → local_integration_check, because its verdict winners are checked through evaluate.py;
   • backtesting-engine → local_integration_check, because the SPY/LEAN parity replay is our integration with an explicit oracle and rc5 has no native execution here;
   • security-supply-chain → upstream_example_or_native_operation, for gitleaks in CI and pre-commit.

Notes, verbatim, one per row (each replaces the provisional_wording note)

• market-data-reference: Owner's verdict (trading lane, 2026-10-01): selection of record, open, distance 4. Blocking item: c3, the provider comparison on the frozen AAPL plan with point-in-time and availability evidence. Closure items follow the trading closure assessment of 2026-10-01 (read at PR Trading research convergence 2026-09-26: 13 layers, GPT-6 live refutation, verbatim-checked votes (verdict-wave input) #358's head 4d11709).
• identity-provenance: Owner's verdict: selection of record, open, distance 5. Blocking item: c3, the verdict_overturn_when comparison on blueprints/us-equities/point-in-time/fixture.json with the validate_source invariants; no gate blocks it, and it needs a recorded DVC install first. The verdict's own winner, DVC 3.67.1, has no pin of record.
• storage-compute: Owner's verdict: selection of record, open, distance 5. Blocking item: c3, a representative comparison (frozen universe, history, latency, rights, correction and availability semantics). A re-pin decision is pending: DuckDB 1.5.6 and EdgarTools 5.59.1 are published.
• data-quality-orchestration: Owner's verdict: selection of record, open, distance 5 (queue: on_requirement_change). Blocking item: c3. The Dagu-vs-Temporal kill comparison (gap-wave2 Integrate native tool evidence with explicit source-host scope #7) does not meet the same-fixture clause of verdict_overturn_when. pandera 0.33.1 is the verdict's data-check winner without a pin of record.
• research-factors-ml: Owner's verdict: open, with no selection of record for the layer itself. runtime-target.json is the engine and broker record and deliberately names no research, factor or ML component. Blocking item: c1, a selection or an explicit scope-out at the layer's re-record.
• backtesting-engine: Owner's verdict: selection of record, open, distance 4 (the other run graded c1 and c4 partial). Blocking item: c3. Five of six SPY parity cases wait on the undecided costs_and_routing contract; one_zero passed on 2026-09-23. The layer entry's "parity BLOCKED" text is stale and gets a dated superseding note. Cross-family (PR Trading research convergence 2026-09-26: 13 layers, GPT-6 live refutation, verbatim-checked votes (verdict-wave input) #358 at b0eb7a1): GPT-6 accepts provisional retention and disputes comparative merit.
• execution-broker: Owner's verdict: selection of record, open, distance 5. Blocking item: c3, IBKR acceptance-plan steps 2-4 on rc5 after the user's IB Gateway paper sign-in. Alpaca: paper execution at engine b528bb5 (2026-09-29: 11 of 13 trials passed; receipts via Paper receipts 2026-09-29: pre, rth and ext mover series, incentive monitor and overnight watch (mechanics-only) #564). dca821c has no paper execution yet, and its native-fault binding is stale until its re-run.
• portfolio-risk: Owner's verdict: selection of record, open and inconsistent, distance 5 (the other run graded c5 unmet). Blocking item: c1, settle skfolio as winner or conditional (catalogs/landscape/us-equities.json:2840 against :2773). Then c3: a fresh, single-execution, preregistered matched comparison, and a separately frozen skfolio upgrade experiment (1.2.9 to 1.4.10).
• evaluation-experiments: Owner's verdict: selection of record, open, distance 5 (queue: on_requirement_change). Blocking item: c1, reconciling the current choice (foundation workers and observability) with the verdict winners inspect-ai, mlflow and agent-retrieval-bench.
• agents-models-workers: Owner's verdict: selection of record, open, distance 5. This row follows the foundation memory and retrieval rows, whose closure governs. Blocking item: c3, a preregistered held-out financial, document and code retrieval comparison with abstention. The layer entry's pins (ai-memory 2.3.1, SocratiCode 1.14.0, CLI 0.155.1) lag the stack pins shown.
• observability-hosting: Owner's verdict: selection of record, open, distance 5 under the shared rule (queue: on_requirement_change). This row follows the foundation observability rows. Blocking item: c2, one frozen set (the Tempo conflict; Jaeger and OpenObserve dispositions), then c3 on the target host. Five pins are behind upstream; Grafana 13.2.3 is a security release, which is a foundation pin decision.
• security-supply-chain: Owner's verdict: open, with no selection of record (queue: on_requirement_change). Blocking item: c1, and the layer's own re-run trigger has fired (cosign verify-blob and OpenBao secret-lifecycle receipts). gitleaks upstream ships security patches only and names Betterleaks as its successor, which is a foundation pin decision.

Everything else in the 12 rows stands as written: winners at stack pins, closure cells (last complete run per layer), alternatives, reasons, new-host steps and the run-variance statement. My acknowledgement covers the trading rows only. The template, builder and test changes are foundation-owned.

[seathatflowsinourveins]

Trading lane: re-read the 12 us-equities rows at 9bcc0365 in code against my acknowledgement (comment 5927943766). All 12 owner notes are verbatim, and no provisional_wording remains. The #4983 gate text is exact, the three gates are added and the three evidence_class changes are applied. The adaptive-paper winner is pinned to dca821cc under name; I accept that adaptation, and no stack.json row is needed for an in-repository engine. The trading rows are final for this edition; acknowledged.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3792b41813

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[seathatflowsinourveins]

Trading lane: answers on the overstated acceptance classes (items 1-7)

I accept the rule: a class describes a run that the cited source, or one file it links, shows returned. Under that rule five of the seven items are corrections with a path: the cited sources were the wrong files, not missing runs. Every path below was read on main 2fd9b31c (2026-10-01 ~11:20Z). Apply them verbatim.

1. alpaca-py. Correct, in two parts.

• market-data-reference, identity-provenance, storage-compute (the data-acquisition role): upstream_example_or_native_operation.
  • Source: blueprints/us-equities/authenticated-data/README.md:13-14: "The September 20 run used installed Alpaca-py 0.44.0 and made five authenticated GET requests. All returned HTTP 200."
  • The order-contract tests are not this role's evidence.
• backtesting-engine, portfolio-risk (the separate Alpaca path): local_integration_check.
  • Source: blueprints/us-equities/adaptive-paper/trials/pre-20260929/receipt.json. It records the 2026-09-29 paper series run through the adaptive-paper engine. That engine pins alpaca-py==0.44.0 (blueprints/us-equities/adaptive-paper/requirements.txt:3) and submits orders with alpaca.trading.requests.LimitOrderRequest (transport.py:175).
  • The class is local_integration_check, not native operation, because the runner is ours.

2. alpaca-py in execution-broker. Correct to local_integration_check.

• Source: the same receipt, blueprints/us-equities/adaptive-paper/trials/pre-20260929/receipt.json: bounded paper roundtrips with fills, all trials passed and flat. The rth- and ext-20260929 receipts are the same kind of evidence.
• The order-contract tests may stay listed as structural_validation beside it.

3. nautilus-trader in backtesting-engine and portfolio-risk. Correct to local_integration_check in both.

• backtesting-engine. Source: blueprints/us-equities/engine-nautilus/spy-parity/verdict-v2.json, which records:
  • case one_zero, verdict PASS, complete true, failed 0;
  • checks[49] expected and observed 2.0.0rc5, so it is a native NautilusTrader rc5 replay against the frozen LEAN oracle.
  • acceptance-plan.md:16 ("Planned, not executed") predates that 2026-09-23 run and is stale. This is the stale text the superseding note in my trading PR corrects.
• portfolio-risk. Source: evidence/artifacts/gap-wave2-20260923/us-equities__portfolio-risk/6-nautilus-riskengine-denial.json: an rc5 RiskEngine run, evidence_class local_integration, outcome advanced.
  • Its preregistration problem (the synthesis's risk) limits its use as a comparison, not its class.

4. nautilus-trader in execution-broker. Correct to upstream_example_or_native_operation.

• Source: evidence/receipts/ibkr-readonly-acceptance-20260923.json. The receipt records acceptance step 1: read-only, with execution disabled, against a signed-in paper IB Gateway. Both the official ibapi probe and the "NautilusTrader 2.0.0rc5 native adapter probe returned 'passed'".
• The class covers step 1 only; steps 2-4 stay open, as the row's gate already says.

5. edgartools in market-data-reference, identity-provenance, storage-compute. Correct to upstream_example_or_native_operation.

• Source: blueprints/us-equities/catalyst-provenance/access-resolution.md:
  • :4: "Native EdgarTools 5.58.0 completed its full quarterly-index request";
  • :76-77: the command runs upstream edgar.get_filings and native bounded Filings.to_context;
  • :105: run 2026-09-24.
• The README's HTTP 403 is the earlier failure that this file records as resolved (:8-9).

6. adaptive-paper in execution-broker. Confirm local_integration_check. It is our runner, with native paper results.

• One addition for the row: the committed paper receipts are at engine b528bb55. The first runs at the pin of record dca821cc happened today: the account-2 recovery went flat at 06:14 ET, and the pre series is running. Their receipts land in my 10-01 receipts PR.

7. Shared winners in the three trading rows.

• opentelemetry-collector-contrib: confirm structural_validation.
• dagu: confirm the change of command to the recorded three-step workflow, with the class unchanged.
• prometheus: confirm.
• systemd: correct to upstream_example_or_native_operation for the historical host.
  • Source: blueprints/us-equities/adaptive-paper/trials/ext-20260929/README.md:360-371, "Units (systemd user journal and manager)". The systemd 255 user journal (journalctl --user -o json USER_UNIT=<unit>.service) and systemctl --user show record the timer-started paper-ext-20260929.service (started 16:00:05 ET, :65) and the alert unit's run.
  • The new-distro recipe is a separate, not-yet-run claim. Keep it in new_host_steps, not as this winner's acceptance.

With these corrections no trading winner is none_recorded. Each row's class therefore follows the weakest recorded winner, not none_recorded. Verdicts, notes, gates and closure cells stay as acknowledged.

[seathatflowsinourveins]

Trading lane: re-read of the 12 us-equities rows at fa8b71da

Read in code against my answers (comments 5927943766 and 5930139395), the SPY correction and the two selections.

Confirmed:

• My seven items. alpaca-py, nautilus-trader, edgartools, adaptive-paper, systemd, otelcol, dagu and prometheus are each cited to the files I named, with the classes I gave.
• The backtesting-engine SPY sentence is verbatim. The added reason about the rc5 EURUSD quickstart (evidence/receipts/native-nautilus-v2-20260920.json) is correct.
• research-factors-ml. comparison_required, winners EdgarTools 5.58.0 and skfolio 1.2.9, and c1 partial with my text.
• security-supply-chain. selection_of_record_open, winners Syft, Gitleaks and Grype; Grype is pinned by catalogs/foundation/automation.json. c1 is partial with my text.
• Your additions. The Codex-review reason on every row. Restic cited to evidence/receipts/native-offhost-restic-20260920.json, and Loki at structural_validation (a7's call), which makes the three hosting rows structural_validation.
• Unchanged. No verdict, gate or closure cell changed beyond c1 of the two selection rows.

One correction (apply verbatim):

• qdrant in agents-models-workers becomes upstream_example_or_native_operation, not none_recorded.
  • Source: evidence/hosts/nativestack-5975wx-20260925/nativestack-5975wx-20260925--qdrant--use--20260925.json. It records result pass: the running Qdrant server answered GET / with version 1.19.1, commit 6ab21ca, the catalog pin. This is on the historical host.
  • Supporting: the SocratiCode qualification receipts ran their Qdrant-backed upstream regressions (REQUIRE_QDRANT=1) against a disposable Qdrant 1.19.1 and exited 0 (evidence/receipts/socraticode-1160-qualification-20260929.json: 4 files, 12 of 12 tests).
  • The health check stays in new_host_steps for the new distro.
  • With this change, agents-models-workers reads upstream_example_or_native_operation, unless one of its other winners is weaker.

Optional, for consistency: portfolio-risk's skfolio cites blueprints/us-equities/research-evaluation/README.md; research-factors-ml cites the run record blueprints/us-equities/research-evaluation/receipt.json. Both describe the same run; citing the receipt in both rows is cleaner.

Scope. The classes of the foundation-owned winners (codex, claude-code, ai-memory, socraticode, qmd, serena, context-mode, rtk, grafana) come from the foundation rows. I did not re-verify their sources.

With the qdrant correction applied, the 12 trading rows are confirmed for merge.