Prove intrinsic P104 translated-progress bound and audit canonical adapters

[Clear1nsight]

This closes a specific hypothesis left by #3954 and #3962: how much persistent progress can a translated P104 call claim? For the unchanged intrinsic endpoints, matched Clifford translations have total progress at most 2312 under the role-local, common-chart grammar. The proof uses the actual native width phi(M)=rank_F2(B) and the conjugacy-invariant upper bound rank(M-I); it does not assume that translation preserves the paid profile.

There is a source-bound qualification. At actual port X0 (q=386), the bare canonical endpoint has a legal matched Clifford translation that raises native width from 8 to 9. Its old rank-zero intrinsic adapter then has width 2; including that adapter restores the intrinsic width bound 8. The 208 intrinsic monomial adapters are distinct from the later 208 rank-one side-correction transports. A whole canonical-word control reprices every one of the 1410 transports: mass 2248, progress 2066, excess 182, compared with the original excess 56. This is a transport-only control, not a complete corrected-call invoice.

The theorem is endpoint-dependent: if a verified changed word leaves all Y roles at F9, the corresponding sharp cap becomes 2416. The package explicitly leaves the changed word's logical return, paid corrections, physical wrappers and global caller bill open. It does not import an unverified sibling saving, claim a universal P104 barrier, or change the selected frontier.

All changes are in research/p104-translated-progress/. Canonical main remains 20ecf2891d80b5acf18d980b2164fb44ea7faa90, priority revision 6. The coordinated overlap pass covered all 676 open PR bodies including drafts at 07:38:11 UTC, with later targeted refreshes; the scope claim distinguishes this endpoint/rank audit from the parallel correction, coding and changed-consumer work.

Validation on published head c6ea4442a62975244ca74b320587d7a3fe791905:

• Source-bound checks authenticate nine canonical inputs, reconstruct 14400 root pairings and 104 physical ports, and check all 280 endpoint cocycles and 1410 original native tariff uses.
• A separate checker imports neither the primary verifier nor the producer. It checks all 518400 Sp(4,2) products, 2160 matched translations, six exact Gaussian-rational operator actions, and the actual counterexample/adapter matrices.
• Six optimized test groups include seven actual CLI receipt/source mutations. Every mutation is preceded by a passing pristine private-fixture run, so rejection cannot be attributed to an already broken fixture.
• Registered package replay, committed-byte/source-history preflight, official export check and all three optimized commands in a read-only Git-free export pass. The frozen export has 20 files including its receipt; their bytes and read-only modes remain unchanged.
• Export SHA256: a3211c2cfeb54e86ab1328db964f5ada50d2c1f723991a7c6fa8915131700a70.

Unsigned symplectic rank is not a phase certificate, a general scalar-operation lower bound, or all-size admission. Coupled operator-valued charts fall outside the proved potential. The next admissible experiment is a real changed endpoint or coupled interface with its full source-to-return accounting.

Credit: Rohan Arun/#1972 and Adam Gall/#1940 for the scalar/frame word; #1991 prepared for Chafik Boukhalfa and its credited native compiler/physical-label predecessors, including icekylinx; Douglas Colkitt and the residual/caller notes; #3882/#3894/#3940 for the correction/caller stream; #3954/#3962 for the conditional gate; #3961/#3965 for distinct boundary/lifetime obligations, and ToikY/#3966 for phase-uniform gauges. Exact replay inputs, compiler notice and license remain pinned. Clear1nsight with substantial OpenAI Codex multi-agent assistance; no independent human or proof-assistant review claimed.

Reproduce: python3.12 -B scripts/research_package.py --package research/p104-translated-progress. Hosted CI passes at this exact head: planning, focused package replay and committed-input preservation succeed.