量化主线进度 #36

Activity

1. Pigbibi commented on Nov 3, 2026

   

   Pigbibi

   on Nov 3, 2026

   ContributorAuthor

   More actions

   协调指令与回复统一放在 #38（本议题只记进度，不记协调过程）。

   #38 第六节列了三项待答：R01 部署路径归属、R02 部署 workflow 的 Secret 名字对不上、R03 ENABLE_MAIN_PUSH_CLOUD_RUN_AUTOMATION 与实际路径矛盾。三项都阻塞今晚部署。
2. Pigbibi commented on Nov 3, 2026

   

   Pigbibi

   on Nov 3, 2026

   ContributorAuthor

   More actions

   M02b / #958: Option A applies under the owner standing rule (2026-11-04): no sizing algorithm change. The three pre-existing 7.5% inverse-peak sizing cases remain on a separate card; no sizing job. r4 changes only the drawdown cache, and native explicit 0.2 must remain bit-equal. The #958 r4 waiter after P01 continues.

   IBKR #966 is CLEAN at 3cd0a3de. Native governance still requires the owner-approved label; merge waits for that owner label and the #967 warmup fix. Agents add no label. No merge, deploy, Secret or IAM change is authorized by this note.
3. Pigbibi commented on Nov 4, 2026

   

   Pigbibi

   on Nov 4, 2026

   ContributorAuthor

   More actions

   Governance change notice — read before I touch anything in this org

   I am the Qoder session that finished the unified risk-gate rollout in the personal and
   GdeiAssistant repositories. Nothing in this organization has been changed yet, and
   nothing will be changed until you reply here.

   Proposal for this org (27 public repos, all with risk-gate in the required contexts):

   1. Replace the current .github/workflows/risk-gate.yml variant with the canonical
      hardened file. The org copy still hard-codes the approver login in the workflow,
      has no RISK_GATE_OWNER repository variable, no second approval label, and
      publishes no ai-review context — so today a Codex review is decorative: nothing
      on the required-context list reads the codex-review-state:v1 receipt.
   2. Publish ai-review from the receipt on the current head, with base drift judged
      by Git ancestry, and add ai-review to the required contexts once every repo
      carries the new file.
   3. Stop treating a third-party review app as an approver for high-risk paths
      (BinancePlatform currently lets its APPROVED review clear protected paths).
      Reviewer availability is not approval authority.
   4. Adopt the two-label model: the owner label claims a human confirmed the change;
      the agent label claims an agent cleared it, and the status description then reads
      NOT confirmed by the owner. Both labels are backed by one repository variable for
      the approver login, set once at organization level.
   5. Keep the SHADOW central engine (governance-acceptance.yml) exactly as it is:
      report-only, pinned, never blocking, never a merge authority.

   Per-repo change shape: one PR per repository touching only
   .github/workflows/risk-gate.yml plus the policy paragraphs that describe it, no
   protection or ruleset edits in the same PR. Settings changes (required context,
   organization variable) come as a separate, separately reviewed change.

   What I need from you

   Reply in this thread with anything in flight that touches these files, so we do not
   collide. From a read-only survey, these look like governance edits in progress — I have
   not opened, closed or commented on any of them:

   • .github ci: enable coderabbit approval for high-risk prs #27 — adds a third-party review app as a high-risk approver and a
     pr-review-signal.yml; conflicts directly with item 3 and with the retired
     CodeRabbit policy. Recommend closing or re-scoping rather than merging.
   • .github feat: Draft exact-head collaboration review prerequisites #39 — edits governance-acceptance.yml and the governance docs.
   • PoliticalEventTrackingResearch #88 — report-only central governance pilot.
   • QuantAIService #450 — SHA-bound review plus owner override, touches the
     third-party review config and .github/governance/.
   • LongBridgePlatform #483 — opened by the cross-repository automation identity.

   Also confirm two things I should not assume:

   1. Which merge entry is authoritative per repository. The survey shows every author and
      every merge executed under the shared owner login, so the GitHub record cannot tell
      an agent action from a human one; that is exactly why the label names and the status
      description have to carry the distinction.
   2. Whether you want the required-context addition handled per repository or through one
      organization-level ruleset.

   I will hold until you answer. If the answer is that this org keeps its own mechanism,
   say so and I will treat these repositories as out of scope.

   中文摘要

   本人（Qoder 会话）已完成个人与 Gdei 仓库的统一闸门铺开，quant 组织暂时一行未改，等你回复后再动。计划是：把组织内旧版 risk-gate.yml 换成加强版（批准人改读仓库变量、补 ai-review 读 Codex 回执、按祖先关系判 base 漂移、共用 head 失败关闭），并把第三方 review 应用从高风险放行资格里撤掉；中心 SHADOW 引擎保持只读不阻塞。改动按每仓一个 PR，只碰 workflow 与策略文档，保护设置另开 PR。请在 #36 这条评论下回复在飞改动（尤其 .github #27、#39、QuantAIService #450、PoliticalEventTrackingResearch #88、LongBridgePlatform #483）、确认各仓合并入口，以及必需检查是用逐仓保护还是组织级 ruleset。
4. Pigbibi commented on Nov 4, 2026

   

   Pigbibi

   on Nov 4, 2026

   ContributorAuthor

   More actions

   Architecture milestone inventory sync — 2026-11-05

   Goal: record the replacement architecture inventory and its acceptance boundaries. These rows synchronize the local proposal; they do not authorize implementation or dispatch. Historical evidence is identified as such and is not a fresh remote acceptance claim.

   Milestone	Status	Evidence and acceptance boundary	Owner / next concrete step	Effort
   Architecture replacement baseline (NEW-ARCH-01..05)	INVENTORY_COMPLETE; NO_NEW_IMPLEMENTATION_BATCH	arch-top5-reaudit/out/REPORT.md; original October 31 definitions remain unrecovered.	Architecture coordinator: record the replacement baseline without renaming old identifiers.	0.5 engineering day for documentation
   Core runtime import boundary (NEW-ARCH-01; ARCH-03A/B)	MERGED; CONSUMER_ADOPTION_NOT_PROVEN	QPK #767 later merged readback; historical API correction preserves 31 exports and core-wheel recording without numerical extras. Earlier tests do not qualify later heads automatically.	Existing QPK owner: retain completion; only a demonstrated consumer gap can justify a separately scoped follow-up.	No new implementation proposed
   Reporting method identity (NEW-ARCH-02)	STOP_DUPLICATE_ARCHITECTURE_IMPLEMENTATION	arch02-scoping/out/REPORT.md; B9 evidence and merged QRS #1056 cover the concrete stale-label defect. Historical #758 overlaps response composition.	Existing B9/QRS reporting owner: retain ownership of remaining method/calendar decisions.	0.5 day handoff
   Configuration characterization (NEW-ARCH-03)	TESTS_MERGED; INJECTION_DEFERRED	LongBridge #694 merged; characterization only, not environment injection. Config/startup overlap: historical #664/#665/#670/#674; ambient shared parsers prevent a proven single-repo closure.	Existing configuration owner: no implementation dispatch; any future brief must first establish helper closure and fresh path ownership.	About 1 day scoping if separately authorized; original implementation estimate 3–5 days
   Startup and execution ownership (NEW-ARCH-04/05)	DEFERRED; NO_QUALIFIED_SMALL_BATCH	arch-top5-reaudit/out/REPORT.md; startup replay and execution event-order equivalence remain unproved.	Existing startup/execution owners: preserve the hold; do not change entrypoints, risk controls or fingerprints.	Original evidence estimates: 3–5 / 5–10 days; not scheduled
   Lifecycle projection (ARCH-04A)	DELIVERED_AT_RECORDED_TREE; CURRENT_PR_STATE_UNKNOWN	Firstrade #513; recorded 1800 tests / 118 subtests, ordinary and instrumented, with independent acceptance.	Established review-and-merge bot: read current #513 state/head only if disposition is still outstanding.	0.5 day readback
   Micro-extractions (ARCH-04B/C/D; ARCH-05A)	STOPPED	Parameter/ownership vetoes and owner pullback; no new implementation candidate qualified.	Architecture coordinator: retain stopped outcomes; do not revive display/telemetry helpers.	0 new implementation days
   Import provenance and alert configuration (ARCH-06 / ARCH-08)	MERGED_WITH_STATED_LIMITS	Firstrade #511 / Schwab #668; Firstrade #499 / Schwab #655. Historical missing scan evidence remains missing; selected parser lanes do not prove runtime adoption.	Existing maintainers: retain completion; avoid duplicate CI/governance work.	0 new implementation days
   Legacy numbering gaps (ARCH-02/05 parent definitions; ARCH-07)	INSUFFICIENT_EVIDENCE	Recovery report does not recover parent meanings; no attributable ARCH-07 definition found.	Coordinator: attach authentic existing references if available; do not invent requirements or block the replacement baseline.	At most 0.5 day link reconciliation

   Confirmed mainline evidence

   • UES#667 is MERGED at squash commit e1795b9db31b (fresh <PROJECT_ID> readback: merged at 2026-11-04 19:31:36 UTC). This is an M01 support-line result; it does not prove consumer adoption, runtime acceptance or deployment. The existing merge-commit review status remains unchanged.
   • O1–O9 DEFAULT_SPEC was confirmed as one package by owner reference t1227s6938 at 2026-11-05 03:57 SGT, as recorded in the approval register. The specification is frozen only. Collection, deployment, orders, unfreezing and any PAPER start require separate scoped approval. Live, Kelly and every stop-loss switch remain OFF as required invariants; no production inspection is claimed. G3 V2 remains TBD; PAPER/runtime acceptance remains NOT_PROVEN.

   Blockers: unrecovered legacy definitions and missing consumer/runtime evidence retain the limitations in the rows above. No governance-file changes or changes to issues/PRs #27 or #39 are included. Existing production rows are unchanged.

   Next step: the architecture coordinator retains these recorded outcomes; any future implementation requires a separately scoped brief justified by a demonstrated gap.

   Source: arch-inventory-r1/out/milestone-36-proposal.md.
5. Pigbibi commented on Nov 4, 2026

   

   Pigbibi

   on Nov 4, 2026

   ContributorAuthor

   More actions

   Canonical risk-gate rollout opened for review

   The owner directed continuation after the 2026-11-04 notice.

   • Set the organization variable RISK_GATE_OWNER to the owner login. It is configuration, not a credential.
   • Opened one PR per repository: replace risk-gate.yml with the canonical tested gate and update the AGENTS.md approval paragraph. No branch protection, ruleset, secret or deployment setting is touched, and I have not labeled or merged any PR. risk-gate stays red until owner-approved is applied, as designed.

   Excluded, as listed in the notice: .github (#27, #39), QuantAIService #450, PoliticalEventTrackingResearch #88, LongBridgePlatform #483, BinancePlatform (custom gate and third-party approver), quantstrategylab.github.io (no workflow) and gcp-readonly-audit (empty).

   PRs:

   • QuantRuntimeSettings#1076
   • QuantPlatformKit#775
   • FirstradePlatform#519
   • CharlesSchwabPlatform#677
   • QuantAdvisorResearch#113
   • InteractiveBrokersPlatform#970
   • SchwabTokenAutoRefresher#103
   • UsEquityStrategies#717
   • UsEquitySnapshotPipelines#578
   • AlpacaPlatform#31
   • QuantStrategyPlugins#114
   • QmtPlatform#92
   • CryptoLivePoolPipelines#243
   • IBKRGatewayManager#199
   • CryptoStrategies#306
   • HkEquitySnapshotPipelines#111
   • ResearchSignalContextPipelines#81
   • MarketSignalSources#58
   • HkEquityStrategies#276
   • CnEquityStrategies#368
   • CnEquitySnapshotPipelines#67

   Adding ai-review to the required contexts remains a separate protection change, to be made only after every repository carries the new file.
6. Pigbibi commented on Nov 5, 2026

   

   Pigbibi

   on Nov 5, 2026

   ContributorAuthor

   More actions

   Canonical risk-gate rollout — heads ready for the merge entry

   Twenty-one pull requests are open, one per repository, replacing the organization risk-gate.yml with the canonical gate and updating the approval paragraph in AGENTS.md. FirstradePlatform also fixes dot-review-submit.yml: the gate's post-push and scheduled sweeps dispatch per-PR evaluations that complete as default-branch runs, where the dot review selector cannot recover a PR head.

   Every head below carries a current codex-review-state:v1 marker with has_unresolved_findings: false. risk-gate is red on all of them because the label is the remaining input — this is the old organization gate, which requires a live owner-approved added by the owner identity for protected paths. Workers do not label or merge here.

   The organization variable RISK_GATE_OWNER is set to Pigbibi, so the canonical gate keeps that attribution after it lands.

   repo	PR	head	codex-review-state:v1	risk-gate
   QuantRuntimeSettings	#1076	d32a53f6	completed unresolved=False current	failure
   QuantPlatformKit	#775	653e6827	completed unresolved=False current	failure
   FirstradePlatform	#519	2255c915	completed unresolved=False current	failure
   CharlesSchwabPlatform	#677	2e5862b5	completed unresolved=False current	failure
   QuantAdvisorResearch	#113	706d8a23	completed unresolved=False current	failure
   InteractiveBrokersPlatform	#970	2d0b4667	completed unresolved=False current	failure
   SchwabTokenAutoRefresher	#103	4ad0076e	completed unresolved=False current	failure
   UsEquityStrategies	#717	956924de	completed unresolved=False current	failure
   UsEquitySnapshotPipelines	#578	15ca522b	completed unresolved=False current	failure
   AlpacaPlatform	#31	74bbfb3c	completed unresolved=False current	failure
   QuantStrategyPlugins	#114	13ede1a5	completed unresolved=False current	failure
   QmtPlatform	#92	2df03e6b	completed unresolved=False current	failure
   CryptoLivePoolPipelines	#243	4d27eda1	completed unresolved=False current	failure
   IBKRGatewayManager	#199	01b0310d	completed unresolved=False current	failure
   CryptoStrategies	#306	61fec1ea	completed unresolved=False current	failure
   HkEquitySnapshotPipelines	#111	39d9c6c2	completed unresolved=False current	failure
   ResearchSignalContextPipelines	#81	275ca956	completed unresolved=False current	failure
   MarketSignalSources	#58	751b8534	completed unresolved=False current	failure
   HkEquityStrategies	#276	a6923b5d	completed unresolved=False current	failure
   CnEquityStrategies	#368	9683bae9	completed unresolved=False current	failure
   CnEquitySnapshotPipelines	#67	11749e41	completed unresolved=False current	failure

   After these merge, risk-gate also reads ai-review from the receipt on the current head, so a head without a completed review shows pending or failure by design rather than passing silently.

   Nothing else in this rollout touches branch protection, secrets, deployments or runtime settings; those would be separate, owner-approved changes.
7. Pigbibi commented on Nov 5, 2026

   

   Pigbibi

   on Nov 5, 2026

   ContributorAuthor

   More actions

   Canonical risk gate — remaining repositories and one shared fix

   Continuation of the 21-PR rollout. Three more repositories now carry the canonical gate, and a P2 that Codex found on PoliticalEventTrackingResearch#89 was fixed in the shared file and pushed to every rollout branch.

   The shared fix. refreshCollidedSiblings() used to run only after evaluate() resolved. A synchronize that moved a pull request off a shared head could then leave the released sibling on a stale collision error until its own event or the six-hour sweep, because SHARED_HEAD (or an API error) threw first. The refresh now runs from a finally block; a refresh failure is logged as a warning and never replaces the evaluation error. The gate file on all 24 branches is byte-identical after the fix.

   The Codex thread on PoliticalEventTrackingResearch#89 is outdated on the new head. BinancePlatform#486 also raised a dead allowlist: its Dependabot exemption read .github/scripts/dependabot-policy.json, which did not exist. The branch now adds that file with the same two workflows scripts/dependabot_auto_merge_gate.py already treats as eligible.

   Every head below carries a current codex-review-state:v1 marker with status: completed and has_unresolved_findings: false. risk-gate stays red on all of them because the label is the remaining input — the base gate is the old organization gate, which requires a live owner-approved added by the owner identity for protected paths. Workers do not label or merge here.

   repo	PR	head	marker
   CharlesSchwabPlatform	#677	59c9035c51d7	CLEAN
   InteractiveBrokersPlatform	#970	755c6c6bbb30	CLEAN
   FirstradePlatform	#519	da99150555fa	CLEAN
   LongBridgePlatform	#705	c59f0d96ab9d	CLEAN
   CnEquitySnapshotPipelines	#67	29e5d1175160	CLEAN
   CnEquityStrategies	#368	7fa93e486a6c	CLEAN
   HkEquityStrategies	#276	1a3662780ecf	CLEAN
   MarketSignalSources	#58	0f6c3c03816f	CLEAN
   ResearchSignalContextPipelines	#81	4217573938a4	CLEAN
   HkEquitySnapshotPipelines	#111	713dcb31095c	CLEAN
   CryptoStrategies	#306	c1d250a13264	CLEAN
   IBKRGatewayManager	#199	4d160ab98cc1	CLEAN
   CryptoLivePoolPipelines	#243	0ac7b96e0900	CLEAN
   QmtPlatform	#92	3d329a2bfa07	CLEAN
   QuantStrategyPlugins	#114	c71b77aa05ce	CLEAN
   AlpacaPlatform	#31	94c699ce8b0f	CLEAN
   UsEquitySnapshotPipelines	#578	56d901346711	CLEAN
   SchwabTokenAutoRefresher	#103	7261132c451a	CLEAN
   QuantAdvisorResearch	#113	8309b137fa61	CLEAN
   QuantPlatformKit	#775	c5713be8f032	CLEAN
   QuantRuntimeSettings	#1076	292f213d3a84	CLEAN
   UsEquityStrategies	#717	eb3b115ae8d7	CLEAN
   BinancePlatform	#486	38b8c14c8948	CLEAN
   PoliticalEventTrackingResearch	#89	cc7fb22d29ad	CLEAN

   What is new in this batch beyond the earlier 21:

   • LongBridgePlatform#705 — same old gate. Also fixes dot-review-submit.yml the way FirstradePlatform#519 does: the gate's push and scheduled sweeps dispatch evaluations that complete as workflow_dispatch runs with no PR in the payload, so the dot review selector now admits only pull_request_target runs.
   • PoliticalEventTrackingResearch#89 — same old gate. Its draft #88 adds a report-only shadow workflow and no gate file, so there is no file conflict.
   • BinancePlatform#486 — replaces a repository-specific variant. It drops the CodeRabbit approval path and the workflow_run trigger on PR review signal; CodeRabbit is not installed on this organization, so that path never produced an approval, and review acceptance now comes from the Codex receipt on the current head. pr-review-signal.yml stays in place as a required context that always reports success and can be retired separately once the required-check list changes. It adds the Dependabot policy file described above.

   Left untouched, with the reason:

   • .github — #27 (open, not draft) rewrites risk-gate.yml and adds .coderabbit.yaml; #39 is a governance draft. Editing the same file now would conflict with that work. #27 also reintroduces a CodeRabbit approval path, which conflicts with the decision to standardize on the Codex receipt; the merge entry may want to close it instead.
   • QuantAIService — #450 (draft) replaces the gate with .github/governance/gate.cjs, a policy file and a review-signal workflow. That supersedes a workflow-only replacement.
   • quantstrategylab.github.io — no risk-gate.yml, and no gate is wired into its required checks; it deploys Pages from pages.yml. Adopting the gate there is a new surface, not a replacement, so it needs a decision.
   • gcp-readonly-audit — empty repository.

   Outside the organization, the same shared fix is open on Pigbibi/PhotoStory#83 (78e17842), current-head CLEAN. Its gate path is protected, so it needs owner-approved from the owner; no agent added that label. Pigbibi/12345HotlineBots has the same gate on main but an open gate PR #235 is changing it, so I left that repository alone; the finally change applies there after #235 settles.

   The organization variable RISK_GATE_OWNER remains Pigbibi. ai-review is published but is not yet in any required-check list; adding it is a settings change for the owner.