Commit Graph

5 Commits

Author SHA1 Message Date
jml
65065f3d8f feat(aphoria): implement community corpus with wiki import and pattern aggregation
Implements Phase 4 (A4) - Community corpus as first-class citizens:

- **Community Corpus Builder** - Queries StemeDB pattern aggregates
- **Wiki Import** - Bootstrap corpus from markdown docs (aphoria corpus import wiki)
- **Pattern Aggregation** - Automatic learning from local scans (--sync flag)
- **Storage Layer** - StemeDBPatternStore with content-addressed deduplication
- **Promotion Logic** - Multi-tier thresholds (95%/80%/50% adoption rates)
- **Corpus Build** - Unified registry for RFC/OWASP/Vendor/Community sources
- **Trust Packs** - Export corpus as signed, distributable artifacts
- **Documentation** - bootstrap-corpus.md guide + CLI reference updates

Technical details:
- Pattern aggregates stored as assertions with predicate "pattern_aggregate"
- Content-addressed subjects via BLAKE3(subject:predicate:value)
- PatternAggregator handles write path (observations → patterns)
- StemeDBPatternStore handles read path (pattern queries)
- Integration tests + fixtures in tests/wiki_import_test.rs

Deleted hardcoded.rs (368 lines) - corpus now fully emergent from StemeDB.
Deleted enriched-corpus-patterns.md (677 lines) - feature shipped.

Closes VG-026 (community corpus), part of A4 milestone.

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-02-09 00:12:31 +00:00
jml
cce54358d2 feat(aphoria): add git commit tracking + comprehensive documentation
**Git Commit Tracking**
- Automatically capture git commit hash when claims/observations are ingested
- Store in assertion metadata for temporal context and audit trails
- Graceful degradation in non-git environments
- Solves double-commit problem by capturing hash at ingestion time

**Implementation**
- walker/git.rs: get_current_commit_hash() utility function
- bridge.rs: Accept optional git_commit parameter in all conversion functions
- episteme/local: Store project_root, capture git hash during ingestion
- 5 new tests for git hash tracking + metadata validation
- All 1162 aphoria tests passing

**Documentation Overhaul**
- README: Added Observations vs Claims distinction, git tracking, dashboard
- CLI Reference: New sections for git integration and ignore/exclusion system
- Comprehensive ignore documentation: .aphoriaignore, inline comments, 4 methods
- Enhanced verification engine docs with matching capabilities
- DOCUMENTATION_UPDATES.md: Complete audit summary

**Dashboard Separation**
- Moved Aphoria-specific UI from stemedb-dashboard to aphoria-dashboard
- Clean separation of concerns: StemeDB for core, Aphoria for security
- Added dashboard documentation and setup guides

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-02-08 18:36:46 +00:00
jml
ef2c8c5940 fix(aphoria): fix 3 critical verification engine bugs
Fixed 3 bugs in Aphoria's claim verification engine that were causing
false positives in Maxwell validation testing:

**Bug 1: Path matching + predicate filtering**
- Added predicate filtering to prevent cross-predicate matches
- Added path prefix matching to respect crate boundaries
- Prevents core/imports/serde from matching hypervisor/vsock/imports/serde

**Bug 2: Value-specific absent checks**
- Absent mode now checks for specific forbidden value, not any observation
- Example: "Clone absent" + "Debug present" = PASS (not CONFLICT)
- Only conflicts when the exact forbidden value is found

**Bug 3: Wildcard pattern support**
- Wildcard patterns like message/*/derives now match multiple paths
- Enhanced wildcard_matches() to support prefix/*/suffix patterns
- Correctly strips full scheme+language from observation paths

**Test coverage:**
- All 39 existing tests passing
- 3 new tests added for bug fixes
- 2 tests updated to use correct predicates
- Zero clippy warnings

**Maxwell validation:**
- maxwell-core-no-serde-001: CONFLICT → PASS (respects path boundaries)
- maxwell-singleton-no-clone-001: CONFLICT → PASS (value-specific absent)
- 5 claims now correctly show as MISSING (expose predicate mismatches)

The fixes successfully eliminate false positives while exposing pre-existing
issues where claims used incorrect predicates.

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-02-08 15:13:10 +00:00
jml
6430ff0fd6 fix(aphoria): move claims.toml to project root and fix verify integration
## Root Cause
Claims file was in applications/aphoria/.aphoria/ but all commands looked
for .aphoria/claims.toml relative to project root. Additionally, .aphoria/
was fully gitignored, preventing version control of claims.

## Changes

### Path Fixes
- Move claims.toml from applications/aphoria/.aphoria/ to .aphoria/ at project root
- Update .gitignore: .aphoria/ → .aphoria/* with !.aphoria/claims.toml exception
- Now claims can be version controlled while keys remain secret

### Verify Integration (Scanner)
- scanner.rs: Load claims from ClaimsFile and call verify_claims()
- ScanResult: Add verify field with VerifyReport
- Report formatters: Add claim verification sections showing PASS/CONFLICT/MISSING

### Clippy Fix
- report/json.rs: Replace filter().map().expect() with filter_map()

## Verification
- aphoria scan . → Shows claim verification with verdicts
- aphoria verify run → Per-claim verification results
- aphoria verify map → Extractor coverage mapping (7/10 claims = 70%)
- aphoria claims list → Reads from project root
- aphoria claims create → Writes to project root
- All tests pass (1120+ aphoria tests)
- clippy --workspace passes

## Impact
Both primary use cases now work:
1. Day-to-day (commit-time): Skills can read/create claims via CLI
2. Audit (scan-time): Scanner verifies code against authored claims

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-02-08 11:09:57 +00:00
jml
3b5f88b4f0 feat(aphoria): implement claims architecture (A1-A5) with verify engine, corpus, coverage, and explain
Complete Aphoria claims system overhaul:
- A1: Rename ExtractedClaim to Observation (extractors produce observations, not claims)
- A2: Add AuthoredClaim with full provenance, invariants, and authority tiers
- A3: Verify engine comparing observations against authored claims, CLI + formatters
- A4: Corpus as first-class assertions with predicate indexing, authority lens, trust packs
- A5: Coverage analysis, explain/docs generation, self-audit extractor, claim suggester skill

Also includes: 42 extractors updated for Observation type, verifiable_predicates trait,
conflict detection with comparison modes, claims TOML persistence, Grafana dashboard,
backup/restore scripts, and comprehensive test coverage.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-08 09:11:47 +00:00