v0.4.0 full documentation audit

Status: completed directory-wide reconciliation of the accepted HOSTED GREEN release baseline.

Audience: maintainers, contributors, documentation readers and coding agents.

Audit date: 2026-09-10.

Read this as a dated audit of v0.4.0, not a review of today's documentation. It preserves the file inventory and evidence from that snapshot; newer pages and code need their own checks. The current changelog archive page is a short index; its original bytes remain at the immutable Git snapshot linked there. The dated audit's preservation claim refers to its own snapshot.

Scope and authoritative subject

The audit accounts for all 485 pre-existing tracked files under docs/: 394 Markdown pages and 91 other files. The per-file inventory records the disposition of every file, including unchanged pages, compiler-embedded guidance, generated catalogs, retained evidence, historical decisions and non-Markdown assets. These files were inspected from the exact tracked-source snapshot 3d34cff7aa984bbb0e1de02d3140d1368399b575; all 3,271 repository blob identities in that snapshot were checked against its tracked tree before the audit.

The implementation subject remains release v0.4.0, commit dfc15e2ddc818fa97744b5a9d69fd6108dd6a321. The release baseline records the maintainer-accepted HOSTED GREEN classification. The release-note publication problem is not an outstanding implementation or conformance task. Updating documentation does not create a compiler version or supply evidence for later code changes.

The audit reconciled concurrent documentation changes. It retained or refined current Status/Audience metadata and the already-pushed Agent, generic, collection, library/I/O and roadmap corrections within the same release scope. It changes no application code, tests, executable examples, frozen protocol formats or target admission.

Corrections closed

Implementation versus authoring history. Current references no longer use obsolete writing-session statements such as unexecuted regression cases or pending first hosted validation to describe implemented release behavior. Headers and body paragraphs agree on the current baseline. Exact older run IDs, measurements and local case counts retain their historical subjects. Release acceptance is not substituted for an unexecuted comparative model trial, a newly claimed physical device, or an explicitly unselected aggregate.

Semantic workspaces and candidate tooling. Candidate constructors, recovery and archive stores, typed holes and drafts, schema/client discovery, analysis and coverage reports, environment/consumer declarations, source review, semantic deltas, rebase/merge, publication and session integration now state their implemented scope and evidence consistently. A tested report that says runtime or external-consumer evidence is not inspected still means exactly that. Compiler regression evidence does not turn its output into a live-world observation or an approval.

Current language and runtime. The algebraic-data RFC and programme ledger now account for admitted owned Result propagation, nonidentity forwarding, argument inference, nested reconstruction, multi-owner records, authored generic variants, compiler collections, callable values, scalar captures and iterator operations. Their additive specifications remain authoritative; older wire/layout/cleanup profiles have not been widened by rewriting an overview. Linked Agent migration and durable recovery are implemented additions, while native/Wasm Agent-stage execution, general providers and wider public support remain separate.

Library and installed toolchain. The library inventory agrees with the 34 bundled packages: nine core, sixteen portable, three alloc, three hosted, one agent and two test packages. The complete required module set is not relabeled as implemented. Installation and release guidance now reflect the shared doctor dispatch in both CLI binaries while preserving fail-closed unavailable production-profile acquisition. Private Rust packaging and full-host publication boundaries remain distinct.

Caches and future work. Authenticated cross-process checked-HIR reuse is already implemented by Persistent Semantic Cache v1. It is no longer described as entirely missing. General incremental checking, broader cache compatibility, measured performance and complete session recovery remain their own requirements. Current roadmap and completion descriptions start from the released mechanisms instead of repeating obsolete first steps.

Navigation and implementation ownership. Repository links, chapter anchors, source/test ownership references and metadata were checked across the whole set. Stale test-harness paths were corrected to the actual projection, Wasm nesting, native-package and collector owners. Body links to root-level Markdown were made explicit repository links rather than nonexistent book chapters; the mdBook catalog retains local source paths for its included chapters. The source-verified tour and installed guidance retain their exact committed text and examples.

Preservation and evidence rules

The audit preserves every pre-existing non-Markdown file, all retained evidence under docs/evidence/, historical decision and execution records, the changelog archive, and compiler-embedded guidance/catalog bytes. Every existing fenced example remains byte-identical to the inspected snapshot. No test, expected diagnostic, fixture, schema identity or known-answer artifact is weakened to make documentation validation succeed.

Historical Phase 0 evidence v2 remains the reviewed 86-row aggregate at its original subject. The separately specified v3 aggregate has no newly fabricated execution bundle. The reserved external Zero comparison lane and unobserved model trials remain unexecuted. The provisioned Linux doctor record retains its actual earlier failed attempts and host requirements. These are scoped evidence records, not an unexecuted backlog for the accepted v0.4.0 implementation.

Private profiles stay private, generated packages stay unpublished unless their own publication decision changes, simulator coverage is not physical-device coverage, and source/HIR admission is not runtime execution. Future checks for changed code remain mandatory. The full product objective remains governed by the completion matrix, not by the number of pages edited.

Reproducible checks

Run the read-only consistency and preservation check from the repository:

python3 docs/audit/verify_v040.py --baseline 3d34cff7aa984bbb0e1de02d3140d1368399b575

The checker validates H1 titles, Status/Audience in the first twelve lines, exactly one catalog entry per page, local repository targets, chapter anchors, all sixteen source-cited language-tour excerpts, unchanged fenced examples, and byte preservation of the protected baseline files. It is standard-library Python and creates no compiler, runtime, network or publication authority. External website availability and every possible host/runtime environment are not inferred from repository link checks.

The publication check additionally executes the four original source-only documentation tests from tests/documentation.rs without changing their assertions, and builds the documentation with the repository-pinned mdBook 0.5.4. Those checks validate documentation; they are not a rerun of the full compiler/runtime test suite. The validation record below identifies the actual hosted audit execution rather than reusing a historical compiler CI run.

Hosted validation record

Audit run 34500651681 executed the checks against the final audited documentation prepared from bb156e1602b88e2812762c9aea2c6c13b4d8b423.

The directory check passed for 395 Markdown pages, 2892 local links, 95 chapter anchors, and all 16 source-cited tour excerpts. Metadata and catalog coverage passed; 108 protected files and all pre-existing fenced examples remained unchanged.

The four original source-only documentation tests passed, and the complete mdBook 0.5.4 build passed. The final report was then included in a repeated check and book build. The full compiler/runtime suite was not rerun by this documentation-only audit; its accepted v0.4.0 HOSTED GREEN baseline remains separate.