Skip to content

Latest commit

 

History

History
118 lines (95 loc) · 7.12 KB

verification.md

File metadata and controls

118 lines (95 loc) · 7.12 KB

Brief-Spec 0.5.0 verification truth boundary

Updated: 2026-08-13 Source basis: 4adf20412028aa858a982c2149c3622327efa11a Canonical repository: github.com/luanmorenommaciel/brief-spec

This document separates implementation, validation, host evidence, and publication. A local or globally installed candidate is not a public release.

Current state

Boundary Evidence Status
Source candidate Version 0.5.0 derived from 4adf20412028aa858a982c2149c3622327efa11a Locally validated release candidate from an uncommitted worktree
Repository private canonical repository Renamed; visibility is still private
Latest GitHub release v0.2.0 Published historical release
Local gates 469 tests, 403 source checks, 459 wheel checks, 86.25% coverage on macOS / Python 3.13.9 Lint, formatting, release verification, tests, and coverage pass locally
Hosted CI run 31708342030 for 4adf20412028aa858a982c2149c3622327efa11a blocked-before-execution; steps executed: false
Live harness evidence Sanitized disposable-repository results retained locally Codex 8/8, Claude 8/8, OMP 4/4, Grok 4/4, and Kimi 4/4 passed
Live release authorization release/live-e2e-evidence.json; source fingerprint 975647c437867b419979db657413ce7e377ce1e19ad678eaeb18a3328781c6ec authorized by none
GitHub v0.5.0 Release API and workflow state not-published
PyPI core/PDF/audio PyPI project/file lookup not-published / not-published / not-published
Canonical schemas GitHub v0.5.0 release assets plus brief-spec-schemas.bundle.json built locally; not published

The hosted failure is infrastructure evidence, not repository test evidence: GitHub rejected the jobs before any step ran because of the recorded GitHub Actions billing or spending limit condition. The five required local live-host gates are authorized; public release remains blocked on the exact-SHA hosted and account-owned publication gates.

Harness evidence tiers

Harness Current claim
codex live-verified
claude live-verified
omp live-verified
grok live-verified
kimi live-verified
copilot experimental
cursor experimental
goose experimental

live-verified means the deterministic installation controls and the required authenticated live matrix passed for this local candidate. hold is a visible unresolved condition. experimental is best effort and non-blocking.

Retained live-host matrix

Harness Passed/required Local sanitized summary
codex 8/8 .briefspec/live-e2e/ultimate-final-codex/summary.json
claude 8/8 .briefspec/live-e2e/ultimate-final-claude/summary.json
omp 4/4 .briefspec/live-e2e/ultimate-final-omp/summary.json
grok 4/4 .briefspec/live-e2e/ultimate-final-grok-verified/summary.json
kimi 4/4 .briefspec/live-e2e/ultimate-final-kimi/summary.json

Every passing scenario required the host hook, bounded contract, strict validation, equivalent deterministic downloads, bundle and receipt verification, and an authorized-change check. The summaries retain metadata and hashes, not credentials or raw transcripts. Grok's native gate uses passive classification hooks, one bounded Stop repair, and a native read/edit allowlist in a disposable repository; shell, web, memory, and subagents remain disabled.

The committed evidence file contains only sanitized decisions, artifact hashes, categorical failure reasons, and hashed opaque session references. CI recomputes its release-source fingerprint and the release authorization fails closed unless all required hosts pass.

Candidate artifacts and rollout

Boundary Retained evidence Status
Distribution set .briefspec/ultimate-release-candidate-dist; manifest 0aedc4e61157632cfdb17f220a11093d7d12c8e4dd7d91382eb1e7fa66dd431f Six wheel/sdist files pass Twine, SHA-256, resource projection, and wheel/sdist clean-room installation
Core wheel SHA-256 77305f5d5c08504baf3b91eeb5264566aeff74d5e7eb71bbf0dbed0921dff2c2 Installed globally from these exact bytes
Offline schemas SHA-256 ae182e3db46c8c9b124edad4131f2a13cd8412414ee47988132b47784772c6c6 Built and validated locally; not published
Global tool Core plus PDF and audio renderers Core, PDF, and audio 0.5.0 installed from the exact local wheel set
Global doctors User and detected project scopes Codex, Claude, OMP, Grok, Kimi, and Cursor PASS; absent Copilot and lifecycle-limited Goose WARN
Intentional overrides Receipt-owned paths only Converge Claude and Nexo Codex PASS at 0.5.0
Rollback Byte-for-byte manifest verification Pre-upgrade user and receipt-owned Converge/Nexo snapshots restored and the final candidate reapplied; unrelated dirty-status digests remained byte-identical

The artifact manifest intentionally has no hosted source authorization: these bytes came from an uncommitted local worktree. Hosted CI must rebuild once from the exact committed revision and the release workflow must reuse those hosted bytes unchanged.

Browser and optional-renderer gates

Gate Local result
Self-contained HTML PASS at desktop and mobile widths with zero external requests
Playwright PDF PASS for selectable text, fonts, metadata, geometry, page rendering, and deterministic bundle bytes
Offline macOS audio PASS for MP3 decoding, duration, disclosure, source-script hash, and deterministic bundle bytes
Explicit-consent OpenAI audio Not run because OPENAI_API_KEY is absent; mocked coverage passed

Candidate trust corrections validated locally

  • honest classification with negation masking, abstention, decision records, and reported-marker downgrade;
  • zero-network and no-plugin verification defaults, SSRF-safe public-address policy, bounded files and archives, and no command execution;
  • receipt-hash ownership with preservation of locally modified skills;
  • deterministic runtime precedence and hermetic no-CLI/fake-CLI test lanes;
  • status-first human renderings, portable release-hosted schemas, and an offline schema bundle;
  • exact-SHA CI authorization that reuses tested bytes and blocks tag-time rebuilding;
  • evidence-based harness maturity, with all five required hosts live-verified.

Account-owned release gates

  • Restore hosted CI execution
  • Choose public visibility at release time
  • Register three PyPI Trusted Publishers
  • Approve the v0.5.0 tag and public release

The repository-owned implementation may prepare these gates but cannot truthfully mark them complete. The release workflow must consume the exact candidate artifact produced by successful CI, publish identical bytes through Trusted Publishing, verify PyPI/GitHub digests, then finalize GitHub.

Historical v0.1.0 evidence remains in verification-v0.1.0.md and is never promoted to current evidence.