# Evidence for job #3618: finding #2043 on `research/corner-measurement.md` Integrates return **#595** (`accepted`, `patch_status: pending integration`, so the served file does not yet carry it) with the three corrections finding #2043 names, and nothing else. | what | sha256 | bytes | |---|---|---| | served text (base of this revision) | `74e781d01217ebc809bee8c3f0305ea9b56ed4bd238a8e50fb123481031f364f` | 19859 | | **revised text (this return)** | `f5ceb46f31259ad80a280ea4c036ef420eb85e0ecb3169ffa3f8140acbb56110` | 23873 | Three edits: #595's hunk 1 (the §5 float-cutoff bullet), #595's hunk 2 (§6's exact band counts, applied verbatim), and #595's hunk 3 (the two new §6 bullets) with items (2) and (3) applied. The note's ledger block is byte-identical, and no other line moves (`diff` is the whole change). ## "It still runs and its stdout reproduces byte for byte" This artifact is a note, not a script: it has no stdout of its own, and §1 binds its numbers to `corner-measurement.js`'s OUTPUT block ("the custody-bound copy is the script's OUTPUT block"). The producer is untouched here, and no embedded 64-hex hash exists in the note, so nothing needs re-embedding. The check that matters was run instead: `check_against_script.py` recomputes the right band `(Z, rpHi]` from first principles — exact integer floor of a real power of two, then exact integer division, no floating point — and compares it with **every** row the served producer's OUTPUT block records for the two variants that carry the actual cutoffs: ``` rows compared with the served script's OUTPUT block: 34, mismatches: 0 ``` so the band model behind every number below is the model the served script actually ran. ## 1. The float-cutoff bullet (item 1): scoped `E_1 = floor(2^(j*(1 - 1/20 - 2/100)))` in double precision uses the exponent `j*0.9299999999999999`, and `Tmax = floor((x - 2)/(max(Z, E_1) + 1))`. Recomputed exactly (`Decimal` at 420 digits, floor of a real power of two; the float side is the script's own expressions run under node): | claim | measured | |---|---| | `Z = floor(2^(j/20))` agrees with its exact value | at every `j <= 700` | | `E_1` first disagrees | `j=54`, by **-1** (`1311373436501308` float vs `1311373436501309` exact) | | `Tmax` disagrees for `j <= 491` | only at `j = 100, 200, 300, 400`, each **+1** | | those four moves | at `2^7, 2^14, 2^21, 2^28` - powers of two, hence composite, so the prime set is unchanged | | first move changing the prime set | `j=655`, adding the **prime** `63419863243463`; no `j <= 654` changes it in either direction | So the float and exact bands are the same set of primes at every `j <= 491` - the whole range the run uses - and are not the same at `j=655`. #595's own explanation of the `j=54` move ("E_1 passes 2^53 and the double mantissa runs out") is wrong and is replaced: `E_1` there is about `2^50.2`, where a double still resolves quarters, so the move comes from the exponent's own rounding. Beyond 491 the moves are not confined to multiples of 100 (`j=492` is +1 at `23306284880`, even and not a power of two, still composite; `j=500` is a move at a power of two; from `j=588` they are frequent). ## 2. The non-monotonicity bullet (item 2): `k=4`, first at `j=61` Band prime counts, exact, `j=20..300`: ``` first j with count >= k : k=1 -> 23, 2 -> 41, 3 -> 53, 4 -> 61, 5 -> 65, 6 -> 71 permanent from j : k=1 -> 34, 2 -> 50, 3 -> 59, 4 -> 61, 5 -> 65, 6 -> 71 ``` `k=4` is therefore the smallest `k` for which the first `j` holding `k` is also the last `j` holding fewer than `k`; for `k <= 3` it is not (the band is empty again at `j=32,33`, falls back to 1 at `j=47,48,49`, and to 2 at `j=57,58`). The bullet now says exactly that. ## 3. The Bertrand bullet (item 3): a prime in `(Z, 2Z)`, `Tmax >= 2Z - 1` Bertrand's postulate gives a prime strictly between `Z` and `2Z` for `Z >= 2`, so the band `(Z, Tmax]` is non-empty as soon as `Tmax >= 2Z - 1`. The `(m, 2m]` form used in #595's text asks for `Tmax >= 2Z` and is not equivalent; it fails in a *populated* case - exactly the counterexample finding #2043 gives: ``` eta_0 = 1/40, j = 20 : Z = 2, E_1 = 262144, Tmax = floor((2^20-2)/262145) = 3 band (2,3] = {3} Tmax >= 2Z-1 : True Tmax >= 2Z : False ``` The bullet's empty-band data was re-verified and stands: largest empty `j` is `33, 10, 6, 3` for `eta_0 = 1/100, 1/40, 1/20, 1/8` (checked over `j <= 200, 200, 120, 60`), each strictly below `1/(2 eta_0) = 50, 20, 10, 4`, and the `>=2 needs eta_0 >= 1/71` and `>=5 needs eta_0 >= 1/31` figures at `j=36` reproduce exactly (`count = 2` at `1/71`, `5` at `1/31`). ## Unresolved obligations 1. This revision fixes what #2043 names. The moves past `j=491` are not enumerated in the note beyond the first that changes the band; the full measured list to `j=700` (78 moves) is in this run's `work/moves.txt`. 2. `research/QUESTIONS.md` reads this note's ledger block, which is untouched, so the registry is unaffected by this revision. 3. 20 of @Benjaminsen's returns wait for a verdict; nothing for the person to do.