bit-vector-tensor-control-policy / docs /canonical_thin_shell_v0.md
J94's picture
Initial Space upload
3436bdd verified

A newer version of the Gradio SDK is available: 6.14.0

Upgrade

Canonical Thin Shell v0

Clean product one-liner: keep one authored system file, compile the operator-facing contracts, and let the existing kernel/runtime keep doing the work.

Why

The fast build path is not a rewrite.

It is:

canon/system.yml -> compile -> current shell kernel

That reduces duplicate authority without throwing away the working harness.

Authored Surfaces

Surface Role
canon/system.yml canonical source for API, runtime, control, and inference state
scripts/compile_system.sh deterministic compiler from canon into live surfaces
api/run_turn.sh current executable shell kernel
benchmarks/run_standalone_control_bench.sh current proof surface

Generated Surfaces

Surface Generated from canon
api/conversational_api_v0.json yes
runtime/work_manifest_v0.json yes
policy/control_language_v0.json yes
inference.yaml yes
build/system/compiled_system.json yes
build/system/user_governance.json yes, from data_prep_stage reduction surfaces
build/system/continuity_provider.json yes

Operator Path

./bin/bvtctl compile
./bin/bvtctl governance
./bin/bvtctl benchmark

Why in plain English: edit one file, compile, then measure the result through the same harness you already trust.

User-Governed Next Moves

The compile path also pulls reduced user-data doctrine from data_prep_stage:

  • operator brain stable rules
  • motif-family reduction
  • primitive trigger map
  • interaction primitive status

Why: the next moves should come from repeated user patterns, not only from local repo taste.