agentic-praxis-grimoire

APG40 APG39 matryer/is and Nix Integration Exit

Phase ID: APG40

Status: Complete — APG39 Nix profile integrated provisionally, matryer/is deferred, and ADR 0027 rejected

Date: 2026-07-26

Scope and outcome

APG40 verified and delivered the exact immutable APG39 authoring branch, preserved its sole-parent relationship to APG38, and based one forward Codex integration commit directly on it.

nix-test-profile is retained-provisional after exact-source and rights review, forty corrected public-safe scenario families, one coherent source-fact correction pass, bounded structural-corpus review, and fresh non-author corrected-state acceptance.

matryer-is-test-profile is deferred-material-defect. Canonical source and runtime probes corrected APG39-IS-07 so the nil-like operand renders <nil> and the non-nil operand renders with type and value in both operand orders. Fresh corrected-state review then found that the candidate still misstated the release’s final equality mechanism. The one-cycle rule prohibited another behavior correction, and all current candidate surfaces were forward removed.

ADR 0027 is Rejected. ADR 0026 remains Accepted and controlling for the directly selectable native Go and go-cmp owners. ADR 0025 remains Rejected. go-testing-stack remains absent.

Resulting state

Development contains:

Public and active v0.3.0 remain 19/19/19. No retained candidate is promoted beyond provisional.

Evidence and validation

Exact-version matryer/is probes covered nil orders, modes, assertion families, helper registration, nested, mixed, generated or cross-package boundaries, wrong instances, all-captured fallback, filename exclusion, subtests, parallel-order limits, source diagnostics, missing source, and protected output. The complete temporary root was removed.

Nix evidence was source-only and covered exact platform sandbox defaults, fallback and configuration precedence, relaxed and derivation-kind behavior, network, sandbox paths, builder and cache boundaries, flake output-specific validation, package phases, executable-platform gating, package-associated derivations, helpers, containers, virtual machines, matrices, and structural ownership.

Final validation passed the retained fixture and focused contract, owner graph and no-stack checks, skill-library and projection checks, project lifecycle, router and release policies, strict inventory, 80/80 unit and integration components, Python and shell syntax, command help, Markdown and local links, public/private independence, privacy, provenance, rights, record identity, whitespace, and a disposable current-development v0.4.0 candidate build/check.

No Nix parse, evaluation, build, flake check, store/cache operation, container, virtual machine, activation, deployment, or external operation ran. The combined, readiness, smoke, release, publication, deployment, target-repository tests, and broad Go race, fuzz, benchmark, and coverage suites were not run.

Trial assessment

The third Claude/Codex trial is supported with revised guardrails. APG39 closed the APG38 terminal defects and reduced redesign work, but exact-version profiles still require executable diagnostic probes and independent source/structural review. Immutable author history and forward corrections remain useful. This bounded result is not a permanent universal organization policy.

Preservation and stop boundary

APG39 was not rewritten, no Claude branch moved, and every APG40 correction is forward-only. Public, active, reference, RepoMap, and personal baselines remain unchanged. The target repository’s commit remained unchanged; unrelated uncommitted test work appeared there concurrently during APG40, was separated by repository and timestamp, and was not read, modified, tested, staged, or discarded by this phase. No successor phase is authorized; readiness, smoke, release, publication, deployment, and any later matryer/is redesign require explicit maintainer authority.

Subsequent readiness

APG41 later retained the complete 28-skill surface for an exact v0.4.0 candidate and dispositioned nix-test-profile as ready-provisional-with-limitation from source-only dogfood. No Nix operation ran, no maturity changed, and this subsequent evidence does not rewrite APG40’s original outcome or authorization boundary.