Audit: independent replay confirms #681's 1,484-operation raw factory saving (no new kappa)

[Rohansguliani]

Independent audit of #681 by Alluri (@alluri-ritual), pinned at head e536f6fc8aad65adef8ac4682d13625598a5c3f9. It covers #681's own contribution, the 87→34-gate replacement in all 14 raw PR622 coordinate-total windows. The bundled #696 lowering was audited in #783. Credit for the construction belongs to #681.

Verdict: the 1,484-operation saving is confirmed exactly, for the raw constructor word. It does not reach the selected providers and does not change κ.

Exact base: 66849e1475c930fd7732b1a0067ffee7e1e95f60 (current main, also #681's merge base). Only research/community-pr681-audit/ is added.

Reproduce

python3 -B research/community-pr681-audit/replay.py

Expected:

PASS_PR681_INDEPENDENT_REPLAY windows=14 scalar_saved=1484 fwd_primitives_old=170488->160296 fwd_primitives_pr696=128696->123768 kappa=51110230636731/31250000000000000

It uses the standard library, runs offline, takes about 30 s and imports no #681 code. The two templates are transcribed from #681's CIRCUIT.json (sha256 f64a9e88…9f71).

Findings

The replay runs the unchanged main constructor directly and finds the 87-gate template by injective relabeling. It finds exactly 14 windows on disjoint registers, without using BLOCKS.json. It splices the witness and rebuilds the cleanup by the literal inverse rule, after checking that rule against the old cleanup. Exact linear forms of all 716 registers match after A, A+B and A+B+cleanup. The inherited native audit, emitter and local auditor from main pass on both words.

Quantity	Before	After	Saved
Raw scalar events	6,340	4,856	1,484 = 2·14·(87−34)
Native scalar operations	16,972	15,488	1,484
Fwd / bwd primitives, inherited lowering	170,488 / 170,536	160,296 / 160,344	10,192 each (5.98%)
Fwd / bwd primitives, #696 lowering	128,696 / 128,712	123,768 / 123,784	4,928 each (3.83%)

• Beyond Research: lower selected complex scalar primitives by 22–25% (no new kappa) #696: once Research: lower selected complex scalar primitives by 22–25% (no new kappa) #696 applies, Consolidate six circuit PRs: 31.72% fewer bit ADDs and 22–25% fewer complex primitives #681 adds 4,928 primitives per orientation on the raw word, not 10,192. The consolidation's 170,488 → 123,768 is composed correctly: 41,792 from Research: lower selected complex scalar primitives by 22–25% (no new kappa) #696 plus 4,928 from Consolidate six circuit PRs: 31.72% fewer bit ADDs and 22–25% fewer complex primitives #681.
• Paid profiles and phase/adapter/center events are unchanged. 98 frame climbs inside the windows are reordered; each register's climb path is unchanged. A randomized dirty-input replay of the emitted scalar programs gives identical final states. Mutating one witness coefficient makes the replay fail.
• κ: Consolidate six circuit PRs: 31.72% fewer bit ADDs and 22–25% fewer complex primitives #681 leaves certificates/selected-result.json and the selected CK/JP8 provider words unchanged, and it does not transport the edit into selected recipes. The 1,484 are raw-constructor operations, not selected-factory operations. Scalar primitives enter only the finite invoice (Audit: independent recount confirms #696's 22–25% primitive saving (no new kappa) #783). κ stays 51110230636731/31250000000000000.

Limits

make verify, make frontier626-verify, Lean and #681's hosted replay were not run. #681's local verify.py and the consolidation's local verify.py both pass here. The five bundled R420 bit-supplier packages were not audited. The prefix bounds come from the inherited auditor and were not re-derived.

Rohan Guliani (@Rohansguliani), with substantial AI assistance. New files are under the repository's Apache-2.0 license.

[Mister-Meeseeks]

100-PR batch review at 0957b70fc2ed1ac7592467e7a24d6d0788ff6758.

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

Reviewed scope: Independent PR681 raw-factory replay and 1,484-macro saving; unrelated bundled bit work not validated by this command.

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.