J94/bit-vector-tensor-control-policy

1# API2 3Clean product one-liner: the conversational API contract lives here.4 5Why: the user should meet one front door over graph state, not raw internal subsystems.6 7## Local turn runner8 9```bash10./bin/bvtctl "summarise the current runtime"11./bin/bvtctl ask "run the demo manifest" runtime/examples/demo_manifest.json12./bin/bvtctl chat13./bin/bvtctl context14./bin/bvtctl bootstrap-context15./bin/bvtctl inference16./api/run_turn.sh "summarise the current runtime"17./api/run_turn.sh "run the demo manifest" runtime/examples/demo_manifest.json18./api/build_system_context.sh19./api/write_session_bootstrap.sh20```21 22Why: `bin/bvtctl` is the operator-facing front door; `api/run_turn.sh`, `api/build_system_context.sh`, and `api/write_session_bootstrap.sh` keep the underlying single API honest and reusable across sessions.23 24## Inference backend25 26`inference.yaml` at the repo root is the inference routing surface.27 28Why: the turn kernel should use one local config to decide whether graph-first reasoning is backed by Codex CLI, so the repo can use the signed-in ChatGPT subscription instead of a separate API key path.29 30## Self-improve entry31 32```bash33./bin/bvtctl self-improve "tighten one bounded product slice"34./bin/bvtctl self-improve-apply "add one missing operator doc for the self-improve loop"35```36 37Why: self-improvement should enter through the same CLI surface and still end in manifests, receipts, and benchmarks rather than free-form repo mutation.38