DONE: physical component precision and an interior-bit witness for the P104 dual

[Arham-Tahir64]

DONE / RELEASED / 2026-10-15 17:00 UTC / priority revision 5

Established, author-checked: a sufficient exact physical component grid and magnitude envelope at every boundary of PR2586's unchanged 5,398-event canonical word and every positive payload tensor depth. An integer product-delta input requires ceil(9 lambda/2)+2 fractional bits internally, exactly one more than its complete final state. At one layer D18[0]=(-1+i)/128 after event 2143; final fields need at most six bits.

Public artifact: PR2674, exact commit aeb5551b162b78e33c0b42681abaff92eb608d70. Package research/p104-physical-dual-grid. Registered replay, direct array execution, reverse checks, four behavioral groups, corruption controls and repository suite pass (three repository skips). No independent human/maintainer review, hosted CI or native compiler regeneration claimed.

Next useful work: an actual child rounding/disposal contract, or relative-chart cancellation to reduce the conservative format. Exact atomic children, canonical charts and ideal private erasure remain premises. No hosting, complete supplier or new kappa. PR2662 fusion and intrinsic bridge work have different word/boundary contracts and are not admitted here.

Claim released; original claim and scope history preserved below. Arham Tahir, substantial OpenAI Codex assistance; predecessor credit retained in the artifact.

---

CLAIM / 2026-10-15 16:44 UTC / priority revision 5

Question: what sufficient dyadic component grid and addresswise magnitude envelope covers every instruction boundary of PR2586's unchanged 5,398-event canonical physical inverse-transpose at every positive payload tensor depth?

Base: current main b4560d9. Exact pending dependency: PR2586 at 7953c9a, credited to ywz65 with its full correction/native lineage retained. Card independent-moment, changed scope recorded locally: independently reconstruct the literal frame/invoice ledger as a control, then certify the unpriced component-grid implication.

Method/domain: separate standard-library rational interpreter, all public dirty inputs and all 5,398 events including 416 endpoint conversions; exact signed rows and an independent reverse check; canonical frame denominator counts and a written all-tensor-depth bound. Meaningful chart/scalar/grid/source mutations. Expected artifact: research/p104-physical-dual-grid with exact word/source pins and opt-in replay, or a precise falsification/handoff. About 60 minutes, reserving checking time.

Nearest overlap: PR2549 treats completed-child errors in logical Hilbert norm for the older 4,982-event clean dual; PR2586 checks physical native identities; PR2603 treats a different QQ16 component-grid contract; Discussion2619 optimizes endpoint fusion. This claim changes neither native word nor hosting. Exact atomic children and canonical chart semantics will remain explicit premises. Child-internal precision, approximate errors, strict rounded erasure, physical routing and a supplier moment remain open. No new kappa.

Evidence status: LOCAL-ONLY / no artifact yet. Active claim expires 2026-10-15T18:44:00Z. Prepared for Arham Tahir with substantial OpenAI Codex assistance; no independent human review.