# Verification recipe, return for job #67 (paper, slug tailcount-transport) A reviewer checks the manuscript against the record and reruns the two cheap checks. The deep run (fold 41) is optional and priced separately. ## 0. Fetch ``` /files/ # tailcount-transport.md /projects/twin-primes/docs/paper/proposals/prop-tailcount-transport.md /projects/twin-primes/docs/research/history/staging/attack-foldL-03-transport.md /projects/twin-primes/docs/research/history/staging/verify-tailcount-transport.md /projects/twin-primes/docs/research/history/staging/frontier37.md /projects/twin-primes/docs/research/attack-foldL-03-transport.js /projects/twin-primes/docs/research/attack-frontier37-02-transport.js /projects/twin-primes/docs/research/exact-g2-ladder.js ``` (`/files/` needs the bearer token as of 2026-09-11.) ## 1. The two proofs that are new (no compute) - §3 Lemma 2: non-emptiness of $A_L(i)$ is equivalent to the interior class word being a legal two-state walk. Check the induction in both directions. - §4 Theorem 2: every window with alternation-legal interior sits inside a realised maximal dead run of the big tile at the copy $k$ with $-(s_i + kW) \equiv a \pmod q$, so $G_2(T_q) \ge M_{\mathrm{alt}}$; with Corollary 1 this is the identity. Check that the run is finite (the big tile has $D(q-2)$ live positions) and that wrapping the old period is handled by the convention that one turn adds $W$. ## 2. Mechanism check of Theorem 2 (2.5 s) ``` node job67-alt-exact-check.js > job67-alt-exact-check.log sha256 job67-alt-exact-check.js = 8d0eb8f6dcd3b83e94d852dab03e4eae8999457fab28037ec0533f88017c9148 sha256 job67-alt-exact-check.log = a6c39cedea18600e87f5d8b860161f5aa71cd44a94617bbc4227cdec62db66d0 ``` Expected: six `ok` lines for folds 7 to 23, `violations=0`, `nu-identity failures=0`, legal window counts 5, 15, 141, 1557, 23363, 390521, containments checked 8, 30, 276, 3042, 45638, 769282, and the final line `ALL OK`. The log has no timings, so the hash reproduces byte for byte. ## 3. The served origin producer (14 to 40 s) ``` node research/attack-foldL-03-transport.js 23 > foldL03-23.log ``` Expected: the certificate table 42, 66, 108, 150, 204, 270 against truths 42, 66, 108, 150, 204, 258; `max N_new/RHS` 1.0000, 1.0000, 0.8881, 0.8975, 0.9180 with 0 violations; the part-4B chain table 108/180/240/330 at every fold; `all checks passed`. The log carries timings, so compare figures, not the hash (my log: sha256 38ff6580620237abf5c102290aaba94ae1ec95a31f806c496c401e96b37de116). ## 4. The served fold-37 producer (recorded 661.9 s; not rerun here) Compare the manuscript's §4 and §5 tables against the embedded OUTPUT block of `research/attack-frontier37-02-transport.js` (or rerun with `node --max-old-space-size=4096 research/attack-frontier37-02-transport.js full`). ## 5. The parallel port and the fold-41 leg (optional; 9 threads) `job67-transport-parallel.js` (sha256 dedb112923e0baab0cd0825b4153651be2b230f8ebb607bd86ce5591dc11db7b) is a worker-thread port of the served fold-37 engine, with the stream partitioned into blocks that carry pre- and post-rolls. Its calibration reproduces the served producer's numbers at folds 11 to 37, including the run count 12,338,231,614 at fold 37: ``` node --max-old-space-size=8192 job67-transport-parallel.js calib > job67-calib.log # seconds node --max-old-space-size=8192 job67-transport-parallel.js fold31 > job67-fold31.log # 4 s wall on 9 threads node --max-old-space-size=8192 job67-transport-parallel.js fold37 > job67-fold37.log # about 2 min wall on 9 threads node --max-old-space-size=8192 job67-transport-parallel.js fold41 > job67-fold41.log # 3607.8 s wall on 9 threads, 28346 s CPU ``` Expected at fold 37: `D(new) counted = 217929355875 ... EXACT`, `G2(new) = M_full = 528`, `M_loose = 528 M_alt = 528`, `violations ... = 0` both, `max N_new/RHS = 0.9477 at theta = 48`, windows scored `6341904311` and `6341472275`, `ALL CHECKS PASS`. Logs carry timings; compare figures. Expected at fold 41 (`job67-fold41.log`, my log sha256 9ab74d4720718664c53223e730bb699336f03b50a93536aeea899cca1d7a0099, timings inside): `D(old) = 217929355875 ... EXACT`, `D(new) counted = 8499244879125 ... EXACT`, `L(direct) = 3`, `maximal kill runs enumerated = 434169935510`, `best merged gap by run length 1..7: 540, 540, 546, 0, ...`, `G2(new) = M_full = 546 expected 546 AGREE`, `M_loose = 546 M_alt = 546`, `91 theta values`, `violations (loose Q_L) = 0 violations (alt-refined Q_L) = 0`, `max N_new/RHS = 0.9551 at theta = 72`, `N_new = 4 SUM Q_L = 4 (alt 4) N(theta) = 0`, `windows scored: loose 219618074383, alternation-legal 219618073707`, `longest qualifying gap run = 3`, `ALL CHECKS PASS`. Memory: each worker holds T_23 only (about 100 MB of arrays); no tile above T_23 is ever stored (peak RSS was not measured). Threads: the script hard-codes 9 workers; change `NWORKERS` for another machine. ## 6. Time to check Items 1 to 3: under 15 minutes of reading plus one minute of compute. Item 5 fold 37: 2 minutes. Item 5 fold 41: one hour on 9 threads, about 8 CPU hours single-threaded.