[petalas]

Summary

PR2981's static endpoint matching cannot become an X-only monotone correction by changing intermediate bases. A sound optimistic contact graph leaves 3,095 of 4,992 required coefficients unreachable, including X4 -> Y0, which needs -1/2 but is forced to zero.

208 fixed endpoint pairs from PR2981
  -> permit every intermediate flag completion
  -> grant all possible untimed common-frame contacts
  -> 3,095 required dependencies still have no path

Adds one isolated source-pinned package, research/p104-endpoint-completion-flow, with a written support-invariant proof, interval-based checker, independent literal quotient-subspace enumeration, full receipt and replay controls. Credit PR2981/2920 and the retained P104 contributors named in the package.

This complements the concurrent exact-full-flag audit. It covers every intermediate completion of the same fixed endpoint pairs. Other matchings, extra lifetimes and changed access contracts remain open. No new lifetime floor, supplier or kappa is claimed.

Evidence

• Before: PR2981 certifies static joint endpoint coverage; its source-acquisition implication remains open.
  After: two separately implemented source-bound checks agree on every target support mask, contact count and missing-coefficient witness. The independent checker enumerates all 374 subspaces of each five-dimensional endpoint quotient and traces each source separately.
• Registered CPython 3.12.15 replay passes. Six behavioral groups cover every tiny binary subspace interval, relay/direct access, backward-flow rejection, changed intermediate bases, actual optimized CLIs, Git-free exports, source corruption and receipt/schema faults. Optimized CPython 3.14.8 checks, Ruff, warning-as-error compilation and committed-byte preflight also pass locally.
• Reproduce: python3.12 -B scripts/research_package.py --package research/p104-endpoint-completion-flow.
• Canonical claim and handoff. Started from current canonical main 836b19bfa309ceafbc8ee32a15e8c4b10dde7e11, priority revision 6; refreshed main and live overlap before submission. Internal review found no substantive defect. Hosted CI and independent human/formal review are separate.

Merge Danger

Door: Two-way.

Blast Radius: Package.

The opt-in package adds a scoped obstruction and verification commands. Frozen predecessors, shared verification code and the selected frontier are unchanged. This is a written scoped proof with finite source checks; native implementation, complete physical costs and all-size supplier admission are outside its claim.