Human representation: consumers and evidence

MHR-through-SOMA supplies the nominal body used by Inkmap and simulation. A nominal body is not a measured robot target: the robot’s compile (tatbot ros compile, ros/README.md section 4.2) accepts plane charts only, and body placement stays out of scope until a measured body-to-surface address mapping exists.

Maintained core and frozen experiments

Capability	Consumer	Evidence type and limit	Unresolved dependency	Next experiment
Pinned body assets and contracts	Body cache audit, asset exporter, typed readers	Software/cache integrity; no physical geometry acceptance	Reviewed bytes, visual review and device parity	Compare retained poses with the pinned provider
Nominal body and placement	Inkmap browser and simulation bundles	Browser/geometry checks; nominal only	Measured mapping to a physical target	Map one retained placement onto a fixed body-shaped phantom
Artwork and material compilation	Inkmap, simulation and tatbot_ink	Deterministic software and synthetic coverage tests	Measured placement and deposition error	Compare one retained design with marks on a stationary plane fixture
Execution lowering	Simulation factory references	Shared Cartesian compiler, C++ joint planner and FK evidence	Current measured inputs and existing native qualification	Measure placement, scale and trajectory error on that fixture
Registration diagnostics	Offline body diagnostics	Synthetic diagnostics or explicitly supplied measured observations; unqualified	Instrument repeatability, independent held-out landmarks and a body mapping reader	Measure phantom registration error and uncertainty
Proposal training and differentiable patch optimization — frozen	Tests and historical evidence only	Synthetic proposal/gradient checks	Named consumer and measured baseline deficiency	Reopen only for a specific consumer problem and comparison
Phantom mechanics — frozen	Tests and historical evidence only	Synthetic fits; no accepted physical model	Qualified instrument and held-out force/displacement measurements	Reopen only when measured rigid-model residuals justify compliance
Coupled mechanics and anatomy — frozen	Tests and historical evidence only	Synthetic interfaces; no admitted anatomy or coupled model	Measured coupling failure or reviewed source objects and a consumer	Reopen only for a measured failure of the simpler model
Release review	Export/deployment gates and review packets	Explicit reviewer-bound approvals; report generation grants none	Evidence for each independently scoped action	Review the exact candidate revision and its evidence

Frozen means maintenance and reproduction only: no new features, runtime consumers, dependencies, dataset campaigns or recurring evidence jobs. A deliberate scope change must identify the consumer, measured baseline deficiency, experiment and acceptance criteria. Existing tests remain; a checked import boundary prevents maintained runtime code from acquiring a frozen research dependency. scripts/check body-single-path runs that stdlib check even without the simulator.

One retained-design lowering path

Simulation lowers a retained design through human_rep.ink_program, pen_path and the C++ path_plan_check planner. The separate body ExecutionProgram sample compiler and its standalone evidence generator have been retired; their schema and example records remain readable for historical inspection only. Body placement must not be converted to a plane or cylinder by discarding its address or synthesizing a registration. Offline plans grant no hardware authority.

Evidence reporting

Report consumer, evidence type, unresolved dependency and next experiment. review_evidence --evidence-root reads named capability directories and writes capability-inventory.json and independent approval packets. Each new input manifest.json uses tatbot.capability-evidence/1, names its capability, evidence_type (software, synthetic or physical), status, base_git_sha and dirty_state. Only an exact clean revision match is reported as current; a reported result still does not establish acceptance. Missing evidence is missing, and old numbered packets are historical regardless of their pass label. The body audit and synthetic registration suite emit named capability manifests. The legacy root flag remains an alias for reading those immutable records.

The next physical result is one retained design on a measured stationary plane fixture, with independent held-out landmarks and observed-versus-planned mark error, scale, shape and uncertainty. Set acceptance limits from task requirements and instrument repeatability before collection. A plane result does not qualify the body bridge; that needs a separate body-shaped phantom mapping experiment.