[petalas]

Summary

The frozen PR5466 bounded Word RAM entry rejects optimized Python because certificate assertions would disappear. This opt-in adapter retains all 55 assertion sites and executes the complete checklist under -O, with frozen source bytes unchanged.

# Each authenticated assertion becomes an optimizer-resistant branch.
if not predicate:
    raise certificate_assertion_error(message)

Claim type: scoped implementation-preservation proof with finite conformance. The new result is interpreter portability, with no new arithmetic theorem, storage saving, runtime speedup, supplier admission or kappa. The complete scientific output equals a fresh original normal execution after removing only the clock field. Optimization stops at this first verified improvement.

Produced by Nick Petalas. Credit David Ganz / Dugongue for PR5466's kernel and complete checklist, with its unchanged tcslib, FastPoly, CatCrypt, mpfx and arithmetic/multiplication lineage. Python assertion semantics and AST lowering are standard techniques. PR5886's address-liveness allocation is separate context.

Evidence

Before: the original -O command exits with Run without Python -O: certificate assertions must be enabled. After: both complete adapted modes print PASS_WORDRAM_ASSERTION_PORTABILITY, with every assertion lowered to an explicit failure branch.

The complete original checklist is retained: eight NTT test groups, 65,537 modular-multiply inputs, 5,096 butterflies, 2,197 circuit fixtures, 277 complete programs and 20 approximate-backend pairs per precision. Those are predecessor domains preserved here, not new mathematical results.

New evidence includes 24 predicate/message/exception comparisons at compiler levels 0, 1 and 2, all 55 enumerated source sites, five boundary rejections and four actual-component result faults. Normal/optimized receipts agree except for requested flags. Individual complete commands took 10.46 and 10.10 seconds locally, each including its fresh original reference. These durations are check-budget evidence, not a performance improvement.

From the repository root, with standard-library CPython 3.12:

/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -B research/wordram-assertion-portability/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -O -W error -B research/wordram-assertion-portability/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/research_package.py --package research/wordram-assertion-portability
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/check_research_commit.py --package research/wordram-assertion-portability --check-source-history

All commands pass locally. Metadata/pins, exact committed bytes, full declared source ancestry, official export integrity/input coverage and read-only Git-free normal/optimized replays pass. Both export replays leave all payload bytes and modes unchanged. Another CPython 3.12 path may replace the displayed interpreter. Sources are entirely available at main 7c49fe32c55de02c2c873655c804b16e4c3fef0a; no pending Git objects are required.

Limitations and unrun obligations

The proof concerns the fixed source checklist and ordinary assertion effects. It excludes traceback/debugger/coverage formatting, frame introspection, code-object identity and general source transformation. A syntactic assertion inventory and four component faults do not establish a separate falsifying fixture for every site. -OO has synthetic assertion-semantic checks only, and native Windows is unrun.

No selected cold replay, native/compiler/physical admission, Lean or Docker run, independent human acceptance or formal proof-kernel review is claimed. The selected frontier and five framework assumptions remain unchanged. This preservation work is not a scientific audit of PR5466.

Open next questions

• Add per-site falsifying fixtures without changing the scientific domain. Begin with a visit trace that preserves lazy predicate/message evaluation; current component faults do not cover each site separately.
• Run both complete commands on native Windows CPython 3.12 from the committed export. Actual subprocess/path behavior remains untested there.

Related open work

No other open PR studying this bounded-kernel optimized-replay seam was found. The completion refresh inspected all four other account PRs created since the advance CLAIM: #6429, #6430, #6431 and #6432. They concern rational-gate temporaries, P112 compatibility replay and mathematical bounds for different objects. #6430 supplies an implementation-preservation result for a different replay; no shared benchmark or instance is computed. PR5553 remains a historical manifest/planner search hit studying P104 handoffs. No extra computation checked another contribution.

CI notes

Both hosted checks, plan and replay (research/wordram-assertion-portability), passed at exact head dcbde65a7df2c88cc0717d7cb46018301601c453 in run 41538606733. The completed checks were read once; separate run metadata confirms completed success and the matching head. Hosted replay executes both registered normal and optimized commands. Local normal/optimized checks, committed-byte/source-history checks and read-only Git-free export replays also pass. CI certifies the registered finite replay and source bindings, not independent human acceptance or a formal proof-kernel theorem. No failed or waived check remains. Advance claim.

Merge Danger

Door: two-way

Blast Radius: research

Seven new files in one opt-in package. Shared code, predecessor bytes and the selected result are unchanged.