[Matarener]

Requested by Mark · project thread

Change and mathematical scope

This is an independent audit, not a new bound. It is a non-author review of the G20 ordinary BIT chain PR5679 (.003507410738409) and PR5708 (.003507719059749). Revision-8 priorities asked for this review before another G20 search. The package research/g20-independent-review adds four checkers written from scratch. None of them imports code from either package or its predecessors.

Result: no discrepancy found. Both packages pass every check below:

• Exact local word check. All 636 frames and all 2,367 query kinds are checked by exact rank and subspace inclusion. All 685 columns get both a GF(2) endpoint check and a full operator endpoint check with exact 8×8 rational matrices, with no moduli and no probe points.
• Global caller replay over 443,238 and 317,004 events. Paid histogram is recounted from the QUERY events. All 18 calls are checked to equal the local word under one consistent injective wire map. Target inventory and rank surplus are checked.
• Admission arithmetic recomputed with stdlib Decimal at 60 digits: raw crossing, tested margin, α/γ/δ tails and H0/H1.
• The PR5679→PR5708 delta is +36H1−72H2+36H3, with targets unchanged.

Finding. PR5708 seeds its five finite ordinary levels with PR5679's unadmitted value, which in turn seeds from its own predecessor. Rerun with the same recursion, seeding from the selected admitted ordinary backend .0034363538396 reaches exactly .003507719059749 in six levels. The uniform tail guards hold on [.0034363538396, .00351]. If the PR4814 levels accept any completed ordinary backend of this type as a seed, PR5708's component could therefore be admitted without first admitting PR5641…PR5679. This reading needs maintainer confirmation.

Remaining obligations:

• the five framework assumptions;
• FRAMES/controller inputs and the 3,552 MAIN-bank target ranks, which are taken as given;
• the finite fee, stock and compiler schema, beyond the H0/H1 recomputation;
• the conversion from ordinary to admitted BIT and its use in the outer rows;
• PR5782's modulus guard, which is not rederived.

Task and overlap

The base is main 19fdb086e. The sources are PR5679 head 715dda75f and PR5708 head 829d01fc0, both pinned in SOURCES.json. Card: bit-helper-contract, scope refined. The card's first experiment, reading #5679/#5708 as pending proposals, is carried out as an independent review rather than a new split search.

The nearest overlap is PR5782 (local native audit) and PR5798 (global caller audit). Both are by the chain's author. This package adds non-author coverage with independent code, plus the reseeding finding. No other review of this chain was found in PRs up to #5943.

Validation

python3 scripts/research_package.py --package research/g20-independent-review --check-only
python3.12 scripts/research_package.py --package research/g20-independent-review

Both commands pass on Linux with CPython 3.12.3 and the standard library only. The full run takes 84 s and ends with PASS_G20_INDEPENDENT_REVIEW_5679_5708. RESULT.json records every compared value.

There are twelve negative controls: wrong query rank, dropped or duplicated ADD, wrong ADD frame, free move of a payload, erased original, wrong exit frame, global rank change, dropped global ADD, lost target debit and understated invoice. All are rejected. A legal over-payment is accepted and shows up in the bill. These are finite checks of the exact frozen objects, not proofs of the framework premises.

Credit

All G20 constructions and packages belong to Ali Romman, Dugongue, ai-nanachi and the retained lineage named in those packages. The review and checkers are by Matarener, with substantial Anthropic Claude assistance. Apache-2.0.

Open next questions

Both are listed in the package README and TASK.json:

1. Can the PR4814 levels be reseeded from the selected backend? This is blocked on a maintainer reading.
2. Derive the 3,552 MAIN-bank target ranks from CONTROLLER.json.

🤖 Generated with Claude Code

https://claude.ai/code/session_011VhYmBuQgh8SCgbwhCq637