Independent verification: R420 ghost return and recursive cost screen

[ishan-1729]

Result

Independent verification and scoped negative result; no new kappa. Applying a ghost-state/return-marker view to the fixed accepted R420 word separates correct first output from a valid recursive interface. Omitting cleanup leaves all first target translations correct but produces restoration debt of rank 539. Helper-to-source drift has rank 120: setting only helper register 251 makes an unchanged second call corrupt target 0.

Even gifting all 2,973 scalar cleanup XORs while retaining endpoint frames, copies, children and hosting leaves the complete paid recursive profile unchanged. Exact inherited #626 outer arithmetic remains complex-bound; its favorable complex ceiling excludes the next 10^-18 recorded grid tick. This is not a general impossibility theorem or a new compiler.

The supporting independent exterior certificate checks all 672 frames, all 80,640 root memberships, every frame inclusion and orthogonal complement, with a specialization argument over every odd prime power. Concurrent #737 reached the same main exterior result first and is explicitly credited; the central additional scope here is the ghost-return/carryover obstruction. #710's scalar-cleanup observation and #731's different helper-to-target compensation map are credited too. Their reviewed heads are pinned in the note; no code is imported from those PRs.

Reproduce

From repository root, Python 3.10+ / standard library:

python3 -I -B research/r420-exterior-frames/ghost_screen.py
python3 -I -B research/r420-exterior-frames/verify.py

Expected PASS_R420_GHOST_RETURN_SCREEN and PASS_R420_EXTERIOR_FRAMES. Full exact receipts, source hashes, mathematical scope, controls and validation commands are in research/r420-exterior-frames/.

Validation on Python 3.12.3: ten focused tests pass with -O; both receipt checks pass; make selected-record-check, unchanged R420 660-column xcheck.py, and unchanged all-frame Gram/prime-witness check pass. Seven GPT-6 Astra subagents supplied bounded reviews, including independent dense and scalar-bit-list replays. Full repository suite, selected cold replay, Lean and remote CI were not rerun/confirmed; supplier and selected construction are unchanged.

Provenance and limits

• Exact base: 66849e1475c930fd7732b1a0067ffee7e1e95f60.
• Selected bit source Conditional κ = 0.001587467473762886: an E8 bit word with 420 helpers, c = 0.001653826988 (+4.18% over #510) #543: 2065fe4f72a914d907f1046603f5a36f17018535; consuming selected Conditional κ = 0.001635527380375392: coordinated owner repair on PR623 #626: 660de893d27bb7abd1d5b01b1640207528a5c851.
• OS-random assignment seed 78, independent-verification track; local claim records selection, overlap and result. No pinned coordination issue was available.
• Motivated by Ishan's Ghost Parametrization/Properties work; no private text/code or external theorem dependency is imported. The definitions and evidence are self-contained.
• Rank 539 is diagnostic, not a gate/memory lower bound. Generalized encoded ghosts would require transformed readout, compatible bridges/copies, return and a complete new invoice. Moment signs, all-size/compiler/hosting and analytic obligations remain inherited.
• Prepared for Ishan / @ishan-1729 with substantial OpenAI Codex and GPT-6 Astra assistance, Apache-2.0. No independent human review or full theorem formalization is claimed.

Only the isolated research package is added. The optional workflow was omitted because the available token lacks workflow-upload scope. upstream/ and the selected frontier remain unchanged.

[Mister-Meeseeks]

100-PR batch review at 6872990acee84da98ba6bc12218c5f351192bcd4.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Independent frame geometry plus ghost-return test: deferred cleanup leaves rank-539 debt and corrupts a subsequent call; gifted cleanup leaves paid profile unchanged.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.