ContextAtlas v1.0 ships as a public Claude Code MCP server with
LSP-grounded atlas substrate fusing structural precision +
architectural intent + git history + test associations into
single-call context bundles.
Substantive launch substrate per Stream A + Stream B + Stream C +
Stream E v0.9+B cycle work:
Stream A (Ruby adapter; v0.9.0 tagged):
- Fourth supported language via ruby-lsp + ruby-lsp-rails (ADR-21)
- 116 Ruby-specific tests + adapter implementation (~1300 LOC)
- 5 ADR-21 amendments capturing cycle empirical findings
Stream B (repo cleanup + community substrate):
- MIT license + CONTRIBUTING + CODE_OF_CONDUCT + SECURITY + SUPPORT
- Cycle docs migration to docs/cycles/v0_X/ convention
- CHANGELOG + .github/ CI substrate
- Language adapter guide + npm metadata finalization
Stream C C.1 (README launch positioning; 6 substeps):
- Marketing top + two-path framing + differentiation elevation
- 6-surface substrate-currency repair
- Structural restructure + docs/release-history.md cycle narrative
- Quality-defense framing + config-awareness + cross-repo deferral
- Beyond Tokens design-alignment case study
- Blog post substrate (external; not committed)
Item (4) doc-substrate refresh:
- README first-use ADR expansion
- ROADMAP v1.1+ priorities section + [SUPERSEDED] block compression
- DESIGN.md 14 surgical patches (status header / BM25 §Two-layer
gating / atlas schema 1.3→1.4 / Language Adapter §Shipped Ruby
v0.9 / Scope Gates updates / two-paths extraction architecture
subsection / Performance + Benchmark Methodology / Versioning /
Extraction Pipeline reasoning effort)
Stream E execution:
- V-2: dynamic version resolution (V-2-α runtime read from
package.json; introduction-class substrate-currency repair
preventing V-1-class fixture currency gap recurrence)
- Version bump: 0.9.0 → 1.0.0 + lock regeneration (license metadata
sync UNLICENSED → MIT bonus discovery)
- V-3: dogfood extraction refresh (atlas.json 526 → 844 claims +60%;
V-2-α empirical payoff validated at extraction surface; $8.52 cost)
Benchmark substrate (carries forward from Phase 8-10):
- 45-72% token reduction with zero quality regression across
benchmark axes on hono / httpx / cobra benchmark suite
- Paired-t at N=27 per axis; factual_correctness CLEAN;
hallucination/actionability borderline; completeness not
distinguishable
v1.1 candidate cluster surfaced through Stream E.3:
- ROADMAP/CHANGELOG-in-extraction (discrimination principle revisit)
- Commit-filter discipline calibration to project conventions
- Depth-floor calibration refinement for research-note docs
- Unresolved candidate ratio investigation
- Content-shape discrimination (current-tense vs past-tense)
- release-history.md content-shape reclassification
Cycle discipline observations preserved for v1.1 inheritance:
- 22 Pattern 7 axis 5b instances at empirical-verification-before-
classification surfaces (config discrimination + commit filter +
depth floor + STEP-PLAN scope catches + recall drift catches)
- 21 consecutive push-with-commit discipline substeps
- Two-stage fresh-context process for load-bearing restructure
- Substrate-currency-repair cluster pattern (discovery vs
introduction class)
- Cross-role verification: advisor framing → dev empirical
verification → corrected scope ships (canonical sub-pattern)
Closes substantive v1.0 launch substrate-record at the v0.9+B cycle
ship boundary.