Tracking: SOTA Roadmap

[Delqhi]

Umbrella tracking issue for the SOTA roadmap (see docs/plans/sota-roadmap.md, added in #7). Larger architectural workstreams; each will be broken into smaller issues when picked up.

Workstreams (suggested priority order)

• WS4 — Security scanning (ADW/new): secret scanning, SCA dependency-vuln, SAST, license checks; surface via ADW + Bundle CLI. (production blocker, low risk)
• WS5 — Eval harness expansion (Oracle): curated task suite + scoring, regression evals in CI, pass-rate deltas between versions.
• WS1 — Compiler/LSP correctness oracle (SCKG + Oracle): integrate pyright/tsserver/gopls behind an adapter; LSP diagnostics as first-class signal; SCKG consumes LSP definition/references.
• WS2 — Budget-aware context selection (SCKG): relevance ranking (centrality + recency + task affinity) + select-context --budget N.
• WS3 — Behavioral trace diffing (IBD + Oracle): capture before/after traces, produce behavioral diff alongside AST/intent diff.
• WS6 — Incremental graph updates (SCKG): file-watcher/changed-files incremental updates; persist + invalidate per-file subgraphs.
• WS7 — Polyglot parity (SCKG + IBD): bring JS/TS (+1 language) to parity; language-capability matrix in docs.
• WS8 — Persistent cross-task learning (Bundle): store task outcomes keyed by repo + task signature; expose retrieval as an MCP tool.
• WS9 — Semantic merge for parallel agents (new): symbol-level conflict-aware merging reusing IBD AST diff. (research-y)

[Delqhi]

✅ COMPLETE — SOTA Roadmap delivered

All planned subsystems implemented:

• SCKG: Semantic knowledge graph (55 tests)
• IBD: Intent-based diffing (56 tests)
• POC: Proof of correctness (49 tests)
• EFSM: Ephemeral mocking (42 tests)
• ADW: Debt watchdogs (35 tests)
• Oracle: Verification (29 tests)
• Orchestration: Multi-agent (32 tests)
• Review-Interface: Code review (35 tests)

Plus SIN-Brain memory cortex (26 tests)
Plus Simone-MCP write layer (113 tests)

Total: 472+ tests across 17 repos