Still Hunt

• Replays mrlygame's own quests at fixed seeds and scores every frame for a ring-like spectrum, then tests whether the peak ring sits on the seed comb, in the mask's first negative lobe, or on both.
• The replay follows mrlygame::quest draw for draw through the public pieces (variations::board, variations::mask, sequence::rulebook, music::score, frames::secondaries, pivot_options) and steps each chapter with its own count, shift-and-add or an FFT convolution chosen by cost, with mrlymath::life::animate's fate rules; the stepper is checked against mrlymath::life::next_grid on 24 random cases and the whole replay against mrlygame::quest cell for cell on the cheapest quests, those whose naive count stays under a printed cost.
• The engine's naive count is the cost the replay avoids, and no timing figure is claimed or printed: the run prints counts, not seconds.
• A frame is cropped by mrlymath::life::crop to the quest's lived-in square, its mean removed, zero padded to the next power of two at or above the crop and the widest mask, and read on integer rings of the full Fourier square; the peak ring is the ring of highest mean power among rings holding at least 8 bins at or below half the field, and the crate's ring cut (mrlynum::fft::radial_profile, peak_ring) is read beside it, with peaks past half the field counted as corner peaks.
• The ring share is the peak ring's power over the power off DC; the baseline is the same share at the same ring of a density-matched Bernoulli field on the same crop, four samples under a printed seed, cached per crop, field and fill bucket of a twentieth.
• A frame is Chladni-like when its share is at least three times the baseline and its wavelength (field over peak ring) is at most twice the mask side; frames under 8 live cells are skipped.
• That cut is a contrast floor, not a resolution test, so the run prints the envelope beside the total: how many hits past generation 0 peak at ring 1 or 2 of their field, how many read a wavelength longer than the crop, how many are resolved (crop at least four mask sides), and the ring 1 or 2 count per path.
• Every quest prints its own hits past generation 0 as a histogram over peak rings with the plurality ring named, so a quest whose hits sit on one ring is read apart from one that spreads.
• Every law table prints the chapters and quests its rows come from, the rows on one-tile boards and the comb spacing spread, since frames inside a chapter are consecutive states of one board and not independent trials.
• The seed comb is field over the tile's period (the tile side unless the tile repeats inside itself), the mask lobe is the first run of negative rings of the ring mean of the real part of the mask's DFT embedded in the same field, and the band is the negative run around the most negative ring; laws are tabulated over drawn masks (basic and copy paths), resolved drawn masks (crop at least four mask sides), each path, side-3 masks, and the settled stills, each with its chance level for the comb.
• Stills are the last frame of a chapter whose fate is alive; a chapter's generation 0 is its seed board or the previous chapter's pivot frame and is excluded from the laws.

RUN

• bash scripts/cargo.sh cargo run --release -p still-hunt from mrlyprod/
• Domain: outer seeds 1 to 64 under mrlygame::Config::default() (canvas cap 256, mask cap 64, 64 generations, 8 chapters, 64 attempts), then seeds 65 to 68 under canvas cap 512 and mask cap 128 within a printed budget; baseline seed 1729; prints only, exits nonzero if the stepper check fails.

WITNESSES

• These are the numbers this study prints for the atlas section of automata.md; that section carries none of them yet, so nothing below quotes a page row.
• automata.md The atlas: 64 quests, 4626 frames, 609 Chladni-like, 47 of them at generation 0 and 3 among the 64 settled stills, all three on side-3 masks; by path 426 of 1043 basic, 76 of 92 copy, 107 of 3491 simple.
• automata.md The atlas: the copy path is one quest, seed 46, whose three copy chapters draw a mask the size of its own 45-cell canvas on a 64 field, so every copy-path count in this study is that one quest's frames.
• automata.md The atlas: the envelope of the cut is 149 of the 562 hits past generation 0 peaking at ring 1 or 2 of their field, 34 reading a wavelength longer than the crop, and 262 resolved; 66 of the 149 are seed 46's 73 copy-path hits, 34 at ring 1 and 32 at ring 2.
• automata.md The atlas: the 176 resolved drawn-mask rows are 15 chapters of 13 quests, correlated frames rather than independent trials, comb spacing 1.08 to 5.69 with median 5.22 and 22 rows on one-tile boards; on them the peak ring sits in the mask's first negative lobe in 52, within half a ring of a multiple of the seed comb in 77 against 52.9 by chance, on both in 22; witness seed 3, chapter 1, generation 1, mask side 23, peak ring 11 against lobe 13 to 22. (Refuted.)
• automata.md The atlas: at canvas cap 512 and mask cap 128 the copy path yields 40 Chladni-like frames of 51 and the one drawn-mask still, seed 66, copy mask side 75 on a board of 5 by 5 tiles, peaks at ring 10 of field 512, wavelength 51 cells, inside the lobe 8 to 13.
• automata.md The atlas: seed 67's copy mask of side 121 on a board of 3 by 3 tiles peaks at the tile fundamental ring 4 in 11 of its 27 hits past generation 0, a plurality and not the rule: ring 13 in 5, ring 8 in 3, and one each at 17, 21, 30, 34, 47, 55, 63 and 93.
• automata.md The atlas: the replay reproduces mrlygame::quest cell for cell on seeds 1, 4, 7 and 9, the four cheapest quests, each a single attempt at canvas climb 1 with no copy path; the other 60 rest on the same draw order and the stepper's 24-case agreement with mrlymath::life::next_grid, and the full-square peak leaves the ring cut for a corner in 376 of 4561 read frames.