[djbclark]

Records djbclark's rulings of 2026-11-02 (relayed by the djbclark-ade lead 7c58015b), resolving a contradiction in architecture-DEFINITIVE-v3.md: R1/R4 named tendcf-agent only for Android, while §1's diagram, §8/D18 and §9 ran the projector and index 'inside tendcf-agent' on every host.

Rulings: A1 tendcf-agent is Android-only, in Kotlin; Linux and macOS run CFEngine plus a small validate-and-project command run before cf-agent reads host_specific.json, no daemon. A2 that command is Python, promoting bin/projector.py. A3 built after schema v5 lands, before any further host is enforced.

Change (docs and comments only, 14 files): new decision D45; D18 amended; §1, §6, §8, §9, §9.6, §9.8, §12, §13 updated; dated amendment notes on the goal-file, projector and schema-v5 reconciliations (decision text kept); GLOSSARY entry; comment headers in bin/projector.py, bin/schema_lint.py and check.yml. The command is named tendcf-validate. Schemas and fixture data are untouched: the device-trust 'agent' entry (const 'tendcf-agent') is listed as an open point, not edited.

Open for the operator (DEFINITIVE §12, nothing decided here): trigger (exec_command or a separate step; every run or per new goal file); first-build validation scope vs the full §9 validator; the rest of the device-side gate on Linux/macOS (TUF fetch, local yes, stored baseline, policy-tree digest); the device-trust agent entry (id, what sha256 binds); the rfc8785 dependency; macOS python3 and install prefix; SQLite index owner off Android; the v5 detector-takeover point.

Local: xref_lint 0, schema_lint OK, py_compile OK.
Author: Claude Opus 5.5 high (omp slice); lead: Opus 5.5 high.
Reviews (design ruling, two independent seats):

• Fable 5.1 high (read-only): sound with changes, 15 findings, all applied in c430196 (F14 amended: python3 -I without -P). Fixes: the name tendcf-validate attributed to the amending session, not the operator; missed live spots (paper :1865, projector P-8); the v5 takeover point closed by A1; open-list facts (CLT python3 3.9.6 on the review machine < requires-python 3.11; minimal Debian has no python3; DC-11 interpreter chain; agent_map id timing vs v5; who writes schema_ceiling).
• astra (gpt-6-astra high, read-only): sound with changes, 7 findings, all applied in 2ea90b3 (6c narrowed to the placement clauses only). Fixes: per-run validation stated as ordering, with the validate-to-CFEngine handoff protocol open; schema descriptions (report-row :5/:124, goal-file agent_map) amended to D45, description text only; takeover point closed in schema-v5 too; interpreter trust boundary and -I as a recommendation, not settled; root execution as the current Linux deployment fact, privilege boundary open; superseded 'inside tendcf-agent' struck.
  The two reviews do not disagree on any ruling; astra refines four of Fable's fixes.

Operator decisions now listed in DEFINITIVE §12 (none taken here): validate-to-CFEngine handoff protocol; runtime trust boundary (interpreter, stdlib, rfc8785) under DC-11; privilege split; first-build validation coverage; owners of the remaining device duties off Android (TUF fetch, local yes, stored baseline, schema_ceiling row); the agent wire id and digest scope, best decided before v5 freezes.

---

Note

Low Risk
Comment and architecture-doc updates only; no runtime, schema structure, or projector behavior changes. Residual risk is future implementers misreading unresolved §12 open points, not this diff breaking anything deployed today.

Overview
Records the 2026-11-02 operator ruling (D45) that resolves a doc contradiction: tendcf-agent is Android-only (Kotlin), while Linux and macOS run CFEngine plus tendcf-validate — a non-daemon validate-and-project command bin/projector.py is promoted to — before cf-agent reads host_specific.json.

Architecture corpus is updated end-to-end: new D45 and an amended D18 (JSONL capture everywhere; SQLite index owner on Linux/macOS still open), plus §1/§8/§9/§12/§13 and a long §12 open-questions list (trigger, validation scope, TUF/baseline/digest ownership, device-trust agent entry binding, Python/rfc8785 trust boundary, etc.). Reconciliation and guide/paper docs get dated amendment notes (strikethrough on old “inside tendcf-agent on every host” wording) without reopening the core projection decisions.

CI/lint narrative now states the §13 golden gate is the normative projector on Linux/macOS and the Kotlin port must match the same bytes (R22). Schema descriptions only (agent_map, report-row) note D45; wire ids and fixtures are unchanged.

No new binaries, no projector logic changes — documentation and comments only.

^{Reviewed by Cursor Bugbot for commit 2ea90b3. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: efe04ede-4a1d-48a5-8492-c01e243d5e75)

[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 2026-11-02T20:39:53.565421Z	9681ef3	PR opened
🔒 Security Review	✅ Completed 2026-11-02T20:40:54.104452Z	9681ef3	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.

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: bcf71677-0fea-47b0-8f8a-c6ef5fb91895)

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 5438ac70-53c9-4e24-a290-e1665a54ac3e)