Epic: deterministic-substrate — the council-ranked substrate programme

[tcballard]

Execution tracking (epic) for the deterministic-substrate roadmap (ADR-093: intent lives in the corpus, execution here).

Intent (corpus): rac/roadmaps/deterministic-substrate.md — the council-ranked programme advancing rac-core as the deterministic knowledge layer underneath AI coding agents. The scoring method, averages, and the review constraints (recorded as Risks) live in the artifact.

Tracking model: each item first graduates from rac/roadmaps/future/ into its own scoped roadmap (intent captured), then gets its own execution epic when it is taken into build. Until an item is scheduled for build it is tracked as a checkbox here and its roadmap's ## Related Tickets bridges to this epic — a graduated-but-unscheduled item is tracked, not missing an epic.

Tranche A — core differentiators and free wins:

• decision-to-code-proximity — declared code-scope references + deterministic path→decisions lookup (CLI + MCP); precondition for the drift gate — delivered (epic Epic: decision-to-code-proximity — decisions at the point of edit #273, closed; PRs feat(relationships): add decision code-scope Applies To vocabulary [roadmap:decision-to-code-proximity] #284/feat(resolve): add path-to-decisions lookup over CLI and MCP [roadmap:decision-to-code-proximity] #285/feat(explorer): surface decisions governing a code path [roadmap:decision-to-code-proximity] #286; also closed traceability gap 7 on Epic: relationship-vocabulary — close the traceability gaps #236)
• freshness-and-drift-detection (phase 1) — git-derived staleness fields + advisory drift finding in doctor/review — phase 1 delivered (PRs freshness-and-drift-detection: read surfaces — surface git-derived staleness on find/search and MCP results #278/freshness-and-drift-detection: drift finding — advisory suspect-artifact finding in doctor/review #279 merged). The later-phase CI-gate form and freshness-biased retrieval remain fenced and unscheduled (no requirement minted yet; each needs its own decision), tracked in Epic: freshness-and-drift-detection — loud staleness and suspect links (phase 1) #277
• integration-recipe-factory — harness recipe + first verified examples (zero engine diff) — delivered across all three initiatives (PRs docs(recipes): integration recipe factory + first three harness recipes #309/docs(recipes): harness backlog + opencode recipe #310): the recipe authoring template + contract, the verification gate, and the prioritised harness backlog, proven with four recipes (Zed, Windsurf, Cline, opencode); roadmap Achieved (docs(roadmap): mark Tranche A roadmaps Achieved (housekeeping) #313)
• lean-context-delivery — measured, regression-checked MCP context footprint; CLI-first documented — delivered across all three initiatives (PR feat(mcp): lean-context-delivery — measured surface budget, selective retrieval, CLI-first path #311): the measured MCP surface budget (~915 tokens under a 1000 budget), the selective-retrieval guarantee, and the CLI-first delivery docs; roadmap Achieved (docs(roadmap): mark Tranche A roadmaps Achieved (housekeeping) #313)

Tranche B — reach:

• retrieval-diagnostics — explain-miss + floor-ratio bounded-boost gate, golden-pinned — taken into build (epic Epic: retrieval-diagnostics — explain-miss and the floor-ratio boost gate #314, sub-issues retrieval-diagnostics: explain-miss — deterministic "why not X" trace #315/retrieval-diagnostics: floor-ratio boost gate — the bounded-boost invariant as a tested gate #316), scheduled for the July release train (2026.07.3–2026.07.4)
• corpus-export-to-rag-backends — shipped (v0.25.0 projections, ADR-074; rac-connectors companion, ADR-095, incl. self-hostable neo4j); residual docs deliverable absorbed into Epic: corpus-sync — the versioned consumption plane #255/corpus-sync: export contract schemas — publish and CI-validate the three projections #256; artifact Achieved. The consumption-plane evolution continues as the corpus-sync programme (Epic: corpus-sync — the versioned consumption plane #255)

Tranche C — extensibility, in order:

• artifact-family-factory — family-creation contract proven on the built-in Risk pilot — graduated to its own scoped roadmap (rac/roadmaps/artifact-family-factory.md, Planned); unscheduled, tracked here until taken into build
• Third-party artifact types (ADR-083) — entry-point registry: inert seam first, discovery + generic validation second, the hardcoded OKF type map closed in the same acceptance; public invitation stays behind GATE-2

Deferred until their stated triggers:

• lore-at-team-scale — the 50+ developer signal; mandatory audit-on and cache-coherency guarantees are entry conditions — trigger met (organisation-scale rollout) and delivered across all four initiatives (epic Epic: lore-at-team-scale — shared serving at organisation scale #262; HTTP transport ADR-098, derived-index cache ADR-099, shared-server audit identity)
• skill-trust-and-surfacing — third-party skills becoming real (still deferred; rac/roadmaps/future/skill-trust-and-surfacing.md, Planned)

Tranche A status: cleared — all four items delivered, save the fenced later-phases of freshness-and-drift-detection. Tranche B's retrieval-diagnostics is now in build (#314); the next unscheduled items are Tranche C's artifact-family-factory and the ADR-083 entry-point registry.

Ordering and task state live here; the roadmap remains the durable record of what and why.