MIG-5: nothing left behind — the mechanical check and the reduced contract

[snkmcb]

MIG-5, "nothing left behind": the first two of its four items. Two commits: the check (code), then the contract (docs), which describes the check, so they go together.

1. The mechanical check (scripts/check_cmake_boundaries.py)

"No generic motion source file remains here, checked mechanically."

The audit already refused a usd-motion-plugins identity built here, but only under that repository's current package names. None of the six identities that went to motion-connectors could fail it, and neither could motionRuntime, vrmRetarget or a departed suite such as execMotion_blend_tests.

It now holds every identity WORKSPACE.md §9.1 sends elsewhere, under every name it has had (here and there), and fails on:

• a member directory under one of those names (tools/motionRetarget is the one exception: it holds motion_retarget);
• a target named for one, or prefixed by one (<id>_...);
• any file under adapters/ or profiles/motion/;
• a source that opens one of their namespaces (openstrata, motion, vrmAdapter*, …). Using them (openstrata::motion::MotionPose, namespace om = ...) passes.

2. The reduced contract (docs)

"WORKSPACE.md §1 and §2 describe the reduced tree, §9 becomes a record, and PACKAGE_CONTRACT.md drops the packages that left."

• WORKSPACE.md (−711 / +169):
  • The intro says the moves are done. It points at git show v0.9.0:docs/architecture/WORKSPACE.md, the last release that built the departed identities, for the reasoning this document held about them.
  • §1 lists only what this workspace builds (execMotion row gone) and names what it consumes. adapters/ blockquotes gone.
  • §2's allowed edges are the declared ones. Forbidden list rewritten without the adapter / liveTransport / osc / motionTracking / BVH-chain rules. New: nothing here depends on motion-connectors, because a live session arrives as a file. The graph count is re-measured on ost 0.23.6: 5 bundle(s), 3 bundle edge(s), 2 libraries, 13 library edge(s), 1 tool(s), valid.
  • §5 drops execMotion from the artifact list (embedded in execVrm's package) and keeps the product-membership rule (no product name, nothing that opens a transport) without the adapter and BVH history.
  • §8's rows 6a/6b/8 note when each identity left.
  • §9 opens as a record. It gains the missing motionCore / motionRuntime rows, and every row is dated.
• PACKAGE_CONTRACT.md: §4 had already dropped every departed package. §3 rules 4 and 5 were still stated in the present tense over liveTransport, osc and the adapters, and are now in the past tense with the rule kept.
• roadmap/current.md: the migration section said execMotion could not yet be consumed. It now says MIG-0..MIG-4 are done here and lists MIG-5's four items. The P0 identity list is the current one.
• roadmap/README.md: the priority bullet and the NPZ bullet are updated.
• motion-foundation-split.md: status line, MIG-1/MIG-2 ✅, MIG-5 🚧 with two of four items ✅. MIG-1's "the .vrma stage does not change" is marked held: no usdVrmaFileFormat golden changed from 2026-09-18 to 2026-09-24.

Verification

• --selftest: 28/28, including 10 new cases. Two of them are negatives: using the namespaces, and motion_retarget / oscillator, which are not departed names.
• The real tree passes (9 members).
• Mutation on the real tree: a file under adapters/, and a source opening namespace openstrata, each fail with exit 1.
• ost plugin test --workspace --graph-only is valid, with the counts above.
• check_docs.py is OK.

Left in MIG-5

• The aggregate product installs and opens a .vrm and a .vrma from release artifacts, with the shared core as a dependency. This is the next release's dry run.
• The cross-repository VRMA → MotionClip → VRM test runs by default.

🤖 Generated with Claude Code