PR #187 - TheHalfMoon/commandF

Purpose

This PR adds a V3.2 master-plan redesign as planning candidates only. The plan comes from a live audit of main at f82565cca917d119e1c774b2c470e2ac20e0d6dd (tree 7b8a1394…) on 2026-10-08.

V2 remains execution authority. This PR does not change:

any existing Spec Kit;
any AF-02 authority path (.github/, donors/, specs/016-*, tools/af02-verifier/, Cargo.*, oracle_model.rs);
product source, workflows, or rulesets.

No issue is closed.

Files (11; additions only, plus an appended index section)

File	Content
`docs/COMMAND_F_V3_2_REPOSITORY_AUDIT_2026-10-08.md`	Capability inventory, discoveries, PR triage, debt
`docs/COMMAND_F_V3_2_SOURCE_QUALIFICATION_MATRIX_2026-10-08.md`	External and founder sources, with identities, rights, and modes
`docs/COMMAND_F_MASTER_ARCHITECTURE_V3_2_CANDIDATE.md`	Identities, aggregation algebra, contracts, trust boundaries
`docs/COMMAND_F_V3_2_GAP_RECONCILIATION_LEDGER.md`	G01–G50 owners; new G51–G54
`docs/COMMAND_F_V3_2_D1_QUALIFICATION_PLAN.md`	Optional local decision advisor: license, security, CPU matrix, go/no-go
`docs/COMMAND_F_V3_2_PRODUCT_AND_UX_PLAN.md`	`review` journey, Studio, pilots
`research/COMMANDFBENCH_V3_2_PLAN.md`	Builds on sp-1 and sap-1; rights-qualified, independently labeled corpus
`docs/COMMAND_F_V3_2_EXECUTION_ROADMAP.md`	Phases P0–P9, founder decisions FD-1 to FD-10
`docs/COMMAND_F_V3_2_CANONICAL_MIGRATION_PROPOSAL.md`	AF-02 amendment-record bootstrap; durable evidence refs for #100
`docs/COMMAND_F_V3_2_INDEPENDENT_CRITICAL_REVIEW.md`	Disclosed author self-critique (not an independent gate)
`docs/COMMAND_F_PLAN_INDEX.md`	Section K pointer, appended

Key findings recorded

af02-base-verifier is not a ruleset-required check, yet it rejects every Cargo.lock, workflow, and donor change. It fails on Dependabot PRs #95, #97, and #98. PR #101 was merged with it failing. G51 and FD-1 propose a non-circular amendment-record path.
CF-17 (specs 025–046, about 37% of Rust source lines) has no CLI surface (G52).
AF-03 planning is not blocked by AF-02 under the Assurance Program; the serial chain in specs 047 and 048 comes from the V3 playbook candidate.
Windows: af02_a1_corpus_bijection fails with LNK1104 at a 261-character path (MAX_PATH) and passes from a short path.
HL7 core 7.0.0 is out, and upstream #2554 is still open. Dependabot #93 (6.10.4) must not be merged as a pin change.
LiquidAI d1-3B: LFM v1.0 licenses commercial use only below USD 10M revenue, per downstream licensee. llama.cpp d1 support is master-only (after v0.6.0), and the hardening PR #30027 was closed unmerged.

Review evidence (exact head)

Alibaba OpenCodeReview v1.12.11, ocr delegate preview --from main --to HEAD: 0 reviewable / 11 total, all unsupported_ext. Not reviewed by OCR.
Jev: not run. It requires a hosted TypeSafe API key → BLOCKED_COST_NOT_AUTHORIZED. Not a PASS.
Local cargo test --workspace --all-features --locked on Windows (before this docs-only change): 143 passed, 1 failed (path length, see finding 4).
Required checks (rust, assurance-proof, scorecard): see this PR's checks.
The author is an AI agent. Its self-critique is not independent review. A human Code Owner review is required.

Founder decisions requested

FD-1 (gate amendment rule), FD-2 (make the gate required after the bootstrap), FD-3 (#100 terminal state), FD-4 (oracle repin target, deferred), FD-5 (ig-registry data rights), FD-6 (repository license), FD-7 (model-pack distribution), FD-8 (Jev spending), FD-9 (thresholds), FD-10 (CF-14/CF-16 deferral).

🤖 Generated with Claude Code

Summary by cubic

Adds the V3.2 master-plan redesign as planning candidates only, produced from a live audit of main. V2 remains execution authority: no Spec Kit, AF-02 authority path, product source, workflow, or ruleset is changed, and no issue is closed. Requests founder decisions FD-1 to FD-10.

The plan requires exact-SHA founder authorization for the one-time AF-02 exception, keeps issue #100 unresolved pending validated forward evidence, and documents the embedded advisor's egress limits.

Key findings

af02-base-verifier rejects every Cargo.lock, workflow, and donor change and fails on Dependabot PRs #95, #97, and #98; PR #101 merged with it failing.
CF-17 (about 37% of Rust source lines) has no CLI surface.
Windows verification fails at a 261-character path limit.
HL7 core 7.0.0 is out; Dependabot #93 (6.10.4) must not be merged as a pin change.
LiquidAI d1-3B licensing limits commercial use below USD 10M revenue.

A human Code Owner review is required; the included critical review is a disclosed self-critique, not independent review.

Architecture review addendum — 2026-10-08 (commit `7a084324238868645d820c84f0ce4d1e18145ceb`)

An additional planning-only review corrected the candidate without changing product code, AF-02 authority paths, workflows, rulesets, or historical proofs:

Retained evidence: withdrew the inaccurate claim that refs/evidence/** objects are available in every clone. Prefer bounded reviewed evidence bytes on a canonical branch, with explicit fetch and offline-recovery verification and a separate retention strategy for larger/restricted artifacts.
Issue #100: original historical artifact bytes remain unavailable; the issue stays open pending implementation, recovery tests, and a separate truthful disposition decision. A forward archive design does not reconstruct the past.
FD-1: an admin-bypass capability is not an approval; any one-time exception requires a separate founder authorization tied to exact base/head commits, paths/blob identities, and independent review. None has been given by this PR.
Local D1: listening on loopback does not itself prove no outbound egress. The qualification plan now requires observed egress/containment evidence.

Review status: this addendum is an AI architecture review, not an independent human Code Owner approval, Alibaba OCR review of Markdown, or a Jev pass. All CI and approval states must be reassessed at the new exact head; pre-update green checks do not carry forward. V2 remains execution authority.