Record the fixed N1 research start line and dated interim plans

docs/decisions/2026-10-06-north-star-start-line.md

@@ -0,0 +1,122 @@
+# North-star R&D start line — October 6, 2026
+
+The command center selects an independently reviewed, adjudicated S4 on the same 80 native-bar slots as the R&D start line. This records the CC's evidence-led delegated decision; it paraphrases the reported user delegation and contains no verbatim user quotation. It does not declare that the start line has been met.
+
+The current board partitions into **27 recorded operational closures, 40 currently open foundation prerequisites, 12 open interim rows, and one upstream distribution hold**. The 40 are the remaining prerequisite rows, not a replacement denominator: all 80 stay in S4. There are 68 foundation-prerequisite flags across the full inventory, including the closed rows and distribution hold.
+
+Before the CC opens a new R&D wave:
+
+- The 40 remaining prerequisite rows must be READY, BY_DESIGN, or covered by an accepted dated exclusion; prior closures and unresolved critic corrections are also reviewed in S4.
+- Memory-owner and the 11 token rows may be INTERIM, each with a dated plan to finish alongside R&D. Required plans are not claimed fulfilled: token comparison run dates are still unverified.
+- Base-distribution keeps its upstream-release hold.
+- The CC adjudicates the independent review of the complete same-80-slot S4. Installation, a PR merge, tool visibility or an unsigned aggregate is insufficient.
+
+The interim token set is context-supply, command-output, output-compression, code-index, code-graph, repo-packing, structured-data, doc-conversion, api-docs, trace-viewer and token-lane-carriers. Memory is the twelfth interim row. ccusage and statusline remain prerequisite rows.
+
+## Corrected forecast and alternative
+
+**Earliest: Friday, October 9, 6:00 PM EDT (22:00Z)**, on the organic path, only if every step succeeds on its first attempt. **Likely: Tuesday, October 13, 6:00 PM EDT (22:00Z)**; reported range **October 12–14**, with low confidence. These are conditional R&D-start forecasts, not a measured delivery date or an owner commitment.
+
+The adversarial refuter requires the tools-wave relaunch, organic observation, queue/repair/application work, valid #786 evidence, prerequisite closure, then independent S4 review and CC adjudication. The new window is October 6, 8:30 PM EDT (00:30Z on October 7). The paper heavy-work blackout remains October 6, 3:50–8:10 PM EDT (19:50Z–00:10Z next day). Future windows and actual quotas must be checked before any run.
+
+The alternative is to require full **79/80 before R&D**. It has **no evidence-based date**: memory and token comparisons have separate unfinished plans, and the token A/B has no established run date. The old board's October 8 lower bound and October 9 full-readiness target are superseded as a forecast. Full 79/80 remains a separate objective; the 80th distribution row stays externally gated.
+
+This decision is overturned by an independently reviewed S4 that fails a required predicate, a CC ruling reopening a recorded closure, evidence that an interim capability is indispensable to the next R&D task without a dated completion plan, or a new evidence-bound CC decision changing the start criteria. A revised date requires measured progress and a new adversarial dependency/throughput review.
+
+## Pulled-forward prerequisite and critic rulings
+
+Betterleaks' accountable board owner is **github-ci-finalize**. After GO-RELAUNCH, that owner must complete the authorized P1 swap test before October 9 or record an accepted dated exclusion. Use **October 8, 8:00 PM EDT (2026-10-09T00:00Z)** as the conservative exclusive UTC boundary for the pull-forward, rather than the ambiguous legacy month-day-only gate. This records the dispatch requirement, not a run or an exclusion.
+
+Two counted closures require a CC ruling. Preserve the operational 27 pending that ruling:
+
+| Slot | Critic issue and current counterevidence |
+| --- | --- |
+| credential-guard | Rebaseline critic lines 20–21 required the designed fresh Codex canary or a ruling on one-leg evidence. The later retained job-1 result lines 3–7 supplies Codex item exit 0, canary absent, an explicit-present control exit 1 and launcher-presence exit 0. CC must bind it with the separate Claude evidence and decide whether the gate is met. |
+| convergence-validators | Critic lines 25–26 identifies a conflict between the infrastructure definition and the adjudication's session-invocation rationale. CC must justify the intended CI-only consumer with evidence or require the model-chosen per-script organic proof. Passing local tests do not settle this classification. |
+
+No autonomous demotion or readiness promotion follows this documentary record.
+
+## Closed user-item batch, with execution still distinguished
+
+The CC reports git identity resolved by global, main-checkout and lane-worktree readbacks using the intended GitHub identity and a GitHub noreply address. The older plan identity is stale. This removes a user-decision dependency, not git's native operation/use gates.
+
+D01 and D04 have authorization under the prior comprehensive-resolution direction. Apply and readback are still CC-owned and pending. The private exact-step packet names D01's supported native plugin removal and D04's reviewed additive profile writer; current main's create-only profile step cannot be claimed to fill an existing profile. The source prerequisite is explicit. No host change is performed by this PR.
+
+ccusage is **offline-only**, the monitoring plan's G6 cross-check; it needs no new sign-in. Retain its offline acceptance, controls and accounting boundaries.
+
+Playwright site sign-ins receive a dated **start-line exclusion**, because the other web tools cover web retrieval. This does not exclude every browser modality or establish native browser acceptance. Chrome system-package installation is covered by the reported workstation authorization; the CC/operator owns any apply action. This lane performs no sudo operation.
+
+Existing Wave 1 proceeds only as runtime-qualification evidence under its own gates. Draft-only preparation explicitly permitted by the CC remains separate. No new R&D wave begins before this start line. Evidence is carried to the trading home through the already selected setup path.
+
+## Evidence and publication
+
+Primary authority: command-center item `task-ns2604-coop-20261006T175007Z`, SHA256 `9d251e3416ec4081ad1889c52493e592372ab70fd159b63d9f1aa8f6763e3f2f`. The supporting workflow `wf_f004715f-5a4` has SHA256 `dc294eb428aaabfcf2070208cb6f65edf8c1f01ca5744b1b81e4b03e6fbb3f70`; its refuter and 23-finding completeness critic are retained. These are source-review/adjudication records, not new acceptance execution.
+
+The [board](../../evidence/artifacts/ns2604-native-bar-20261006/board.json) records the new start-line forecast separately from the unknown full-79 date, named dispatches, scoped user-item resolutions and unresolved critic rulings. Historical 30/80 and unmeasured final S4 remain separate. Root has no installation, provider, trial or configuration authority from this source change.
+
+Publish through the co-op GPT read, CC acknowledgment and 5f queue. Registry and dashboard checkpoint stay in the last commit. Native dashboard generation readback remains a separate CC/co-op step.
+
+## Amendment (2026-10-07): date-certain review and opening unit
+
+This amendment supersedes the conditional start forecast and campaign-based waiting in the earlier sections. Their decided text and historical observations remain intact. The command center's START RULINGS, review-ns2604-coop-20261007T072919Z (SHA256 `6b0bde35825d0dbba8d55f1238cb0628621974eb41cb56b1f143580e7c3858e8`), govern the opening unit and the dated readings below.
+
+### Fixed review and adjudication
+
+The independent S4 review starts **no later than midnight EDT on October 8 (2026-10-08T04:00Z)**, on the same 80-slot board as it stands then. It does not wait for all open original gates to close. Each still-open original gate receives the CC's dated exclusion or removal from the start gate that day; this record adds no gate or regrouping.
+
+The CC adjudicates the start **by October 8, 6:00 PM EDT (2026-10-08T22:00Z)**. Only a **safety finding in the paper-engine path** or a **sign-in that only the owner can perform** may hold adjudication past that deadline. The authority is task-ns2604-coop-20261007T002252Z, point 1. This records a deadline and its two allowed holds, not a completed review or a declaration that R&D has started.
+
+S4 uses upstream evidence and clean upstream installation/configuration, the required integration smoke, and eligible organic client records. No local comparison campaign decides a row. Service, timer, dashboard and UI rows use their required one smoke; requested smokes are not organic counts. The interim completion plans are recorded separately below.
+
+### N1: data integrity, then N2 and N3
+
+The unit opened by adjudication is **N1, the data-integrity unit**, owned by **rnd-r1**, with **5f's custody for trading-lane landings**. It applies Layer 1.5 in [the pinned trading rules](https://github.com/seathatflowsinourveins/native-agent-stack/blob/6e6df2372bf7f11bf5bf5289b8bb71828d601912/blueprints/us-equities/AGENTS.md#L35), landed by #827 at `6e6df2372bf7f11bf5bf5289b8bb71828d601912` (lines 35–46).
+
+1. Inventory what the selected platform repositories already provide for security identity, splits and corporate actions, and delisting and terminal prices. Each inventory entry names its repository/pin and exact source locator, plus the source's limitations. Inspect and use existing upstream features before proposing glue. Subjects include the selected NautilusTrader 2.0.0rc5 catalog/adjustment surfaces, EdgarTools identity maps, alpaca-py corporate actions, Lean map/factor files and an already entitled vendor feed. These are inventory subjects, not availability or acceptance evidence.
+2. Produce the residual glue plan, with a maintained reference for each part and a demonstrated remaining gap. Foundation layers retain their upstream implementation; desk glue is confined to what the upstreams leave open.
+3. Accept the unit through a **historical-data end-to-end backtest using the selected NautilusTrader 2.0.0rc5 on a known-answer set**. No strategy number is cited before that acceptance. The retained split-inversion and no-data research can inform that set; a newly supplied research package is not historical availability evidence.
+
+**N2**, after N1's inventory and its applicable data gates, is the **D1 prospective holdout**, whose validation window opens **2026-10-14**. **N3** is the **catalyst Wave 1 rerun** when its data gates permit it. The SPY `one_stress` mapping remains under its separate mapping condition; it is not the unit this start opens. These are unit definitions and ordering constraints, not new runs or completed data gates.
+
+### One canonical essential profile
+
+The essential profile is one planned **`north-star-rnd` entry in [adoption/manifest.json](../../adoption/manifest.json)**, a task-scoped projection of the finalized catalog. **grand-catalog** writes it on **#820's chain after J820-FIX2**. This record references that carrier; it creates no second essential-tool list and does not claim the entry is already published.
+
+Its **single profile base commit is `e712bae606e8d27f010af8e3c13ded6b7a7c207a`**, resolved from `essential-profile-wf_383f6a1e-0cd.json#/profile/base_commit` (record SHA256 `6685923cc99f4457d5c0afa35f0e48cb4295b568709519e2a12ad41bb678ab5c`). The later trading-rule and START RULINGS references above do not define additional profile bases. The profile is evaluated against N1's task requirements; rnd-r1 supplies their install/smoke state or exact gap, rather than treating catalog inclusion as acceptance.
+
+### Dated readings and exclusions, re-checked at S4
+
+All of these readings are dated **2026-10-07** and re-checked at the fixed S4 review:
+
+| Reading or exclusion | Scope and control | Re-check or overturn |
+| --- | --- | --- |
+| N1 execution route | N1's executors are Codex CLI lanes using native sign-in. The gateway is not on N1's execution path and does not hold this start. Cross-family reads retain their established co-op GPT and 5f Claude routes under the dated deviation. | Re-check executor identity and the dated deviation at S4. Plan line 98 resumes for SDK-worker/research-runtime execution from N2 onward and after GW-ROUTE acceptance; this is not a general gateway waiver. |
+| Research process isolation | Network and credential restrictions apply to backtest/data-job processes, not to reading upstream documentation. Before N1's first backtest, rnd-r1 with codex-token-parity names supported installed-client launch arguments and smokes one job: network off for the backtest, scrubbed environment with no broker credential pointer; any data pull is a separate job naming its vendor hosts. | Actual launch/smoke proof is required; no configuration or successful job is asserted here. R&D first-turn exposure must show zero `codex_apps` tools through the native feature setting, or carry a dated exclusion. A disable instruction alone is not exposure proof. |
+| Landing distribution | 5f may land from the retiring distribution until retirement. Every landing requires the CC's individual ledger cue. | A landing without its cue overturns this reading; re-check custody and cue at S4. This lane performs no landing. |
+| Trading output repository | N1 outputs land in this repository under `lane:trading`, with 5f's custody. A later repository move retains its separate gates. | Re-check repository policy and custody at S4; no repository move is performed. |
+| Data entitlement | N1 uses sources already entitled or public. Paid additions remain the owner's decision and must be named. | Re-check every proposed source's entitlement and permitted observation; this is not a blanket purchase or provider-fetch approval. |
+
+Publication of this amendment follows the co-op's exact-head read with CI at that head, the CC's ACK and its slot ahead of the S4 review. It uses the existing #801 and changes no other PR. No early rebase, host action, native smoke, model/provider run or measured readiness promotion follows from this documentary amendment.
+
+#801's existing dependency on #791 is retained. The co-op's landing cue must reconcile that parent and T10's record-publication dependencies; this amendment does not authorize unstacking or changing queued #791.
+
+## Amendment (2026-10-07): dated interim completion bindings
+
+The original memory-plus-eleven-token interim membership remains. The [dated completion-plan bindings](../../evidence/artifacts/ns2604-start-line-20261007/interim-plans.json) name each existing source locator, source digest, accountable owner and next action. They are plans to finish alongside R&D, not completion evidence or an invented delivery date. ccusage and statusline remain outside this interim set. Re-check each plan at the fixed S4 review.
+
+| Interim slot | Accountable completion lane | Plan date |
+| --- | --- | --- |
+| `durable-memory/memory-owner` | memory-h2h / CC | 2026-10-06 |
+| `token-efficiency/context-supply` | overlap-token | 2026-10-07 |
+| `token-efficiency/command-output` | overlap-token / codex-token-parity | 2026-10-07 |
+| `token-efficiency/output-compression` | overlap-token | 2026-10-07 |
+| `token-efficiency/code-index` | overlap-token / codex-token-parity | 2026-10-07 |
+| `token-efficiency/code-graph` | overlap-token / memory-h2h | 2026-10-07 |
+| `token-efficiency/repo-packing` | overlap-token | 2026-10-07 |
+| `token-efficiency/structured-data` | overlap-token | 2026-10-07 |
+| `token-efficiency/doc-conversion` | overlap-token | 2026-10-07 |
+| `token-efficiency/api-docs` | codex-token-parity | 2026-10-07 |
+| `token-efficiency/trace-viewer` | overlap-token | 2026-10-07 |
+| `token-efficiency/token-lane-carriers` | codex-token-parity / overlap-token | 2026-10-07 |
+
+The memory plan retains its upstream re-drive trigger; D3r4 steps 8 onward and local token comparison campaigns do not gate the start. Requested integration smokes, eligible native ordinary counters and independently reviewed S4 remain separate evidence classes. The trace-viewer plan preserves its recorded operational service/UI observation; retaining an interim plan is not a new promotion or demotion. These bindings do not duplicate the canonical `north-star-rnd` projection and add no gate.

docs/decisions/2026-10-06-ns2604-native-bar-recount.md

@@ -15,3 +15,5 @@ A new accepted per-slot native receipt, landed dated scope decision or CC-applie
 CC item task-ns2604-coop-20261006T140508Z holds client-side installs/re-wiring for semble, codebase-memory, jcodemunch, qmd, headroom, SocratiCode and the Codex RTK hook while workflow wf_8703c159-c6d prepares the ordered install plan and names step owners. Backend, index and PR work continues. The hold includes MCP registrations, hooks/trust, instruction files, P2-9 and P0-1 grant diffs. Operational owner acceptance does not override that client-change hold or grant configuration authority.
 
 The later accepted overlap-codenav closure transfers Serena and structural-search to readiness-runner with both closed=false. Their contract map and original controls/exposure/AST coverage conditions remain, and Serena joins its separate plan hold wf_5e8f452c-484. Consume only the authorized frozen #786 v1.1/G13 and named install-owner evidence when supplied; no independent trials. Exclusion is conjunctive per client: full-run zero organic choice, a chosen complete-task covering owner, and one upstream-native fix/rerun changing nothing. An uncovered class remains NOT-READY. #780 is frozen at50bf3c544db95c5121efd3229938148e21763b67 with no push/rebase/close until CC revival; #758 remains co-op/5f custody. No count changes follow this scope transfer.
+
+The later [north-star start-line decision](2026-10-06-north-star-start-line.md) supersedes this record's calendar forecast: conditional R&D start earliestOctober9/likelyOctober13, while full79/80 has no evidence-based date. It retains all80slots, currentreported27 and finalS4unknown; two critic corrections awaitCCruling. NewR&Dwaves wait for the selected reviewed/adjudicated S4, while existingWave1 is runtimequalification. No historical numerator or source receipt is rewritten.