# Recipe — job #656, cross-lane synthesis A reviewer reproduces every figure in the report with two commands and checks the served file's binding with one. Nothing needs more than ~5 GB of RAM and 40 s of wall time. ## 0. Inputs From `/projects/twin-primes/docs/research/`: | file | raw sha256 as served | |---|---| | `attack-foldL-03-transport.js` | `edc8e4ef4a19bc146487c451080f977c06765f044b869a6817901b973b95a586` | | `Lgrowth.js` | `471c76e8c19568f39ad6c1c1cedbf02e05746b9b2f994640bd1972735e3a2da2` | | `U-FRAME.md` §11 | `0941103e8d2fcf1f08b6635c159256d1d61afc13182f2fa5ebe51296245fa7fb` | | `qc/tailfmt.js` (only for step 1) | — | The uploaded files are `ver656.mjs`, `t29check.mjs`, `synthesis656.md` and this recipe; fetch them with `GET /files/`. ## 1. Check the served producer is the copy #159 read (10 s) ```sh python3 - <<'PY' import io, hashlib raw = io.open('attack-foldL-03-transport.js', encoding='utf-8', newline='').read() lines = raw.split('\n') head = '\n'.join(lines[:818]) # bytes above the OUTPUT banner print(hashlib.sha256(head.encode()).hexdigest()) PY ``` Expected: `74e291517b4fbb63ab1c2af8d68ca43b8b6ea1dcd056035ebd16339df0ebc0cf` — the `code-sha256` the file's own fingerprint records and the value return #159's pre-registration quotes. The boundary `lines[:818]` is `tailfmt.js`'s `headText`: the tail starts at the line `// OUTPUT — EMBEDDED, do not hand-edit.` (line 819, 1-indexed), so the head is everything above it. Then, optionally, the pasted-output check (17 s here): ```sh node attack-foldL-03-transport.js 23 > tct23.out 2> tct23.err ``` Its stdout reproduces the block embedded in the served file from `PART 0.3` onward, 422 of 423 aligned lines identical, the one difference a trailing banner line, with elapsed-time tokens (`[1.6s]`) the only scrubbed field. ## 2. The identification, the two certificate profiles, and the depth (8 s) ```sh node ver656.mjs ``` Expected stdout sha256: `570e2cf37696ac65541fc5d99b773168d2fd2c6eb23ce9a78c6120eaae8858ca`. It prints: * **the custody ladder**: `D = 3, 15, 135, 1485, 22275, 378675, 7952175` and `G2 = 12, 30, 42, 66, 108, 150, 204` for `T_5 … T_23` — every entry `MATCH`; * **the fold table of claim 2**: for folds 7 → 11, 11 → 13, 13 → 17, 17 → 19, 19 → 23, 23 → 29, 23 → 31, 23 → 37 the columns `trueL`, `M_loose`, `M_alt`, the `L` values contributing at `M_loose`, the largest `L` contributing in each form, and the window counts. Expected rows, in order: `1 42 42 1 1 1 15/15`, `2 66 66 1 2 2 141/141`, `2 108 108 2 2 2 1557/1557`, `2 150 150 1,2 2 2 23363/23363`, `3 204 204 3 3 3 390693/390521`, `2 270 258 3 3 2 8196279/8195991`, `3 270 258 3 3 3 8200797/8200509`, `2 240 240 2,3 3 2 8048135/8048071`. The `M_loose`/`M_alt` pairs, the `M_loose` deciding-L column, and the 23 → 31 and 23 → 37 window counts are return #159's own printed C output; the first five `M` values and the deciding-L column `1 / 1 / 2 / 1,2 / 3` are the served script's part 4 table; * **the identity sweep**: 163 entries T_13 × 17..1009, 40 entries T_17 × 19..211, 17 entries T_19 × 23..97, 3 entries T_23 × 29..37, plus the 7 diagonal entries — all `disagreements 0`, and the diagonal reads `2, 1, 2, 2, 2, 3, 2`. ## 3. The depth boundary #161 names (31 s, ~2.6 GB) ```sh node --max-old-space-size=6144 t29check.mjs ``` Expected stdout sha256: `3abed9698434fe4f5ec230783ab4c3f6a264d236ab5974239976163b4e6741a3`. It builds `T_29` by iterated fold and prints `D = 214708725 W = 6469693230` (the values #161 stores and #162's run prints), then `L(T_29, p)` at `p = 29, 31, 37, 113, 127, 131, 137, 1009`: **`3, 4, 3, 2, 1, 1, 1, 1` — 8 of 8 equal to return #161's T29 column**, including its 2 → 1 step at `p = 127`. ## 4. What a reviewer is being asked to check 1. That the two statistics really are one: the sweep in step 2, and the one convention difference stated in the report (`runFor` scans one copy linearly with the 2-set anchored at the run's first pair; `trueL` walks the `q` copies with the per-copy translate). 2. That the alternation-refined sum's depth equals `trueL` on the eight folds of step 2 — and that the loose form does not (it reaches `trueL + 1` at 23 → 29 and 23 → 37). 3. The falsifier: any entry with `runFor ≠ trueL`; a fold with max contributing `L ≠ trueL`; or a fold `T_29 → p`, `p ≥ 127`, with `M_loose ≠ M_alt` or deciding `L ≥ 2` — equivalently `L(T_29, 127) = 2`. ## Note on the two hashes The `Lgrowth.js` row above carries the sha256 of the bytes fetched on this run; the value is also in the uploaded files list and is what step 2's identity sweep was run against. Both served scripts are plain UTF-8 with LF endings; if a fetch converts line endings the raw hash changes while every figure in this recipe does not (the reads are byte-for-byte over LF).