Independent T37 schedule and targeted payload/binding verification (same κ as #942)

[tekuuu]

This independently developed T37 owner schedule reproduces the now-selected conditional κ = 0.001661960547713947 through a different physical witness. Jonathan Capalbo's concurrent PR942, published during this work's verification, reached the same profile and κ and has since been integrated. This PR does not improve the current selected value or claim priority. Its contribution is the alternative ten-edit schedule, full reproducible reconstruction, and additional package/payload boundary controls.

The construction combines pinned T21 and T16 schedules with merge keys 16*i and 21*j-528, preserving both source orders. Ten owner edits, including six relocations, realize the trade (44,3,2) → (48,1). PR942 uses an unshifted merge and eight edits. Both yield idle profile {48:11086,1:6} with 333 invocations and 167832 whole helper roles, and both pay complex payload1024 for prefix779 and row17495 < 20161. PR942's all-T proof is credited but is not independently validated or claimed here.

The package lives at research/complex-t37-independent-seed39/ to avoid colliding with PR942. Its strict replay passed all 20 stages, including a fresh U8 reconstruction, source-event windows and charts, full emitted controller/invoice, both retained exact moment engines and all 47 outer inequalities. New checks include all 30 inherited/current omitted owner edits and both omitted delays, 47 package rejection controls, two positive gzip recompressions, byte-for-byte runtime-adapter reconstruction, and five payload mutants that must fail at their specified assertions. The replay also runs 69 admission, 12 bit and 15 controller controls.

Independence is scoped: the owner schedule and new package/control logic were developed independently from PR890 and PR870; much of the verified supplier/controller machinery and both exact moment engines are shared pinned predecessor code. This is not a fully code-disjoint audit of PR942. The final overlap refresh through #984 also found Joel Pulikkan's PR962, head 50c6ae8151493af21d284afe34fb35f68f6d4934, with another eight-edit construction at the same κ. That work is credited too; this submission adds the shifted witness and the specific runtime-byte, receipt-binding, recompression and assertion-targeted controls, not a new packing mechanism relative to those concurrent results.

A standalone exact check also resolves the bootstrap allowance difference: our retained runtime literal is512, while the concurrent packages use1024. All14 recorded cutoffs satisfy the stronger1024 reserve and are identical under both formulas, because gap <= 9/514 makes 36/gap^2 dominate 2*(4+1024)/gap. Run python3 -B research/complex-t37-validation/check_bootstrap_reserve.py; expect PASS_FOURTEEN_UNCHANGED_BOOTSTRAP_CUTOFFS_AT_PAYLOAD1024 with four negative controls. This check imports no inherited verifier and gives byte-identical output under -O.

Reproduce from the repository root with Python 3.12.14/zlib 1.3.2, NumPy/SciPy/SymPy, C++17 and Boost headers (set BOOST_INCLUDE for a nonstandard header location):

python3.12 -B research/complex-t37-independent-seed39/verify.py --work /tmp/t37-cold

Expected: PASS_STRICT_U8_T37_PR589_COMPOSITION 0.001661960547713947. The default rebuilds the full PR589/PR741/PR799 ancestry; the README also documents checked ancestry reuse and separate package controls. The recorded run freshly rebuilt that complete ancestry, then explicitly reused it for the strict T37 replay. The final relocated package manifest is 2bde79cc87f02fa6c4c11d38ae306015cd2d4f57710d15d7ee50233a51c3440f.

All 18 targets of the base Makefile's verify recipe completed in documented stages, including all 87 isolated test modules and patch application checks. The main invocation stopped when a historical compiler command missed the Boost path; after configuring CPLUS_INCLUDE_PATH, the remaining four targets passed. No checker was changed to obtain this coverage. This is not a claim that a single uninterrupted make verify exited successfully. Exact commands, environment, failures, source/file hashes and execution receipts are in research/complex-t37-validation/.

Base: f567884aeb8851b6e794af1184119a3bfbe1e631; assignment seed39, complex supplier track. Principal source pins are PR890 61b71bcf5a34db70220da81fa978d5b0c7bbb008 and PR870 b734850dc348800942f0ffd1f0723fdec7539b2b, with full transitive pins and source notices retained. PR890's later head 4b3bd8cbe8e3e81c44161017c9b72c905a8040a8 changes only its CI timeout; the exact diff is included. Selected main was refreshed at bbc600db2fc15007b7c914f4f2b2ed55fa909c87, and the submission inventory was inspected through #961, followed by an explicit PR942 refresh. No pinned sprint issue exists, so the claim is recorded locally.

Credit Jonathan Capalbo (concurrent T37), Chafik Boukhalfa (T21), Hao Yan (T16), NewBArch and icekylinx (U8 lineage), Jeremy Kothe (PR799), DreamingOfClouds (PR589), and the other retained predecessor notices. Prepared by Tekalign (@tekuuu) with substantial OpenAI Codex assistance; new work is Apache-2.0 and predecessor licenses are preserved.

Inherited all-size/compiler, common-label/mutable-source realization, whole-call hosting, precision/tape, analytic-transfer and bit H1–H5 assumptions remain. Finite certificates are not a formal proof of the full multiplication theorem. No independent human review or hosted CI is claimed for this branch. Neither the selected frontier nor upstream/ is changed.