Symmetric animation - simplified: the code and data behind the page A plain JavaScript port of Vladimir Bulatov's SymSim, and the numbers measured on it. It is not his GPU code; his code is the SymmHub repository, https://github.com/SymmHub/SymmHub, read at commit 1fbb615. The page's demos run this port. Measured Sun 1 Nov 2026 (week 44). Node 20 or later, no other dependency. Files symsim-core.mjs the port: his 17 groups, reverse pixel lookup, nine-point stencil, Gray-Scott and complex Ginzburg-Landau steps symsim-core.test.mjs 30 tests: node --test symsim-core.test.mjs One reads a state he saved and is skipped without his repository; set SYMMHUB to your copy of it to run that one. symsim2d_measure.mjs the measuring script: node symsim2d_measure.mjs list, then node symsim2d_measure.mjs run "" [out.json] SMOKE=1 shrinks a run, to test the script only. Job "p" reads his presets, and needs SYMMHUB too. symsim2d.json every table in section 6, merged from those jobs (see below) symsim2d-perf-browser.json cost in headless Chrome, with CPU throttling standing in for a slower machine symsim2d-perf.json cost in node symsim-presets.json what his shipped presets set: steps per symmetrization, iterations, symMix, domain size in pixels mathematica/ the page's Mathematica tabs, one .wl per tab: paste it into a notebook in Wolfram 15.0 and evaluate. Each is self-contained. The tables in symsim2d.json summary.a value and slope jump across the domain's sides, 17 groups summary.b the seam against steps after the first symmetrization summary.c K = 1, 10, 100, 1000: what a symmetrization would change, and how far the domain moves from the K = 1 run summary.d how far the tiled pattern is from solving his scheme summary.e symmetrization off: how long the symmetry lasts summary.i few lookup iterations against a full lookup summary.k the seam against K and the margin summary.presets his 65 saved wallpaper presets read through the port raw the series behind c, e and b Each table carries its units. "Matching" on the page is "compatible" here: the grid's wrap-around is one of the group's moves. What was not tested His GPU. Nothing here runs his WebGL; float32 behaviour is this port's, by rounding on store only. The only contact with his engine is reading his saved states. A real phone. Chrome's CPU throttling only. A triangular mesh. It would make the hexagonal groups match the grid. Hexagonal stability on an exactly invariant subspace is unmeasured. Wave speeds. "K below margin over speed" was read from one seed of summary.k, not from a measured speed. Interpolation apart from stencil anisotropy in the small error far from a hexagonal domain. Hexagonal "distance from K = 1": read as a different pattern being selected, from its size and flatness across K, not from comparing the patterns. One saved 632 state that reads 0.39. Unexplained. Seams only at a = b = 1, one parameter set per group. symMix only at K = 1. Iterations only on a = 0.25 xx and a = 0.3 333. Whether the lookup can cycle: it did not on 3000 points per tested parameter set; not proved. His starting noise: simplex noise is replaced by Pearson's square patches for Gray-Scott and this generator for Ginzburg-Landau. His slide steps: a slide's "SymSim step" is a frame of several Euler steps, and its length in those slides is not known. Part 2, measured Mon 2 Nov 2026 (week 45) glue-film.mjs the delayed lookup on the Brusselator: the page's section 11 and 13. Plain JavaScript, no dependency twisted-average.mjs simulate, slide back, average: the page's Run button twisted_average.py the same as a small NumPy program: python3 twisted_average.py [seed] part2-numbers.json every number part 2 shows, with the results file and key it was read from part2-measure.json what the page's own JavaScript does: reach from noise, the switch, T held wrong, the formula's three turns, the fold's costs bakeoff.json the algorithms of section 16, one row per variant patch-a-phase-twist.diff SymmHub patch A: one time cost per side, a phase factor, 16 lines patch-b-delayed-lookup.diff SymmHub patch B, A included: a ring of past frames, a Brusselator worker, 199 lines patch-b-on-top-of-a.diff B alone, for reading patch-README.md how to apply and run them Part 2 was checked by agents who wrote their own code from the mathematics before reading the authors': the dynamics, the groups and the geometry. part2-numbers.json names each. Not checked: a real phone, and anything in part2-numbers.json marked "its author". Part 3, find_film line by line: the numbers its cards cite, and the scripts that measured them. twisted_average.py is the program under study. It and twisted-average.mjs run as published. The find_film_* scripts are the record of how each number was measured. They read and write this project's own folders, so run from this folder most of them stop with FileNotFoundError or ModuleNotFoundError: read them, or give them those folders. find-film-runs.json what each line does, measured over many seeds: removing or changing it, k of n find-film-modes.json the exact frame, the spectrum of the one-third-period map, the formulas and their tests find-film-extra-lab.json the first line, and the run of seed 1 sweep by sweep, for the page's own demos find-film-extra-cardsA.json numbers for the cards on the frame, the noise, the guess and the turn find-film-extra-cardsAstar.json seeds with the first line removed: which land on the star find-film-extra-cardsB.json numbers for the cards on the loop, the free period, simulate and the frames find-film-extra-groupc.json numbers for the cards on the slide, the lag, the stop test, the return and what follows it find-film-extra-D.json numbers for the cards on speed, rate, the fit, the 3, the clip and T -= find-film-extra-Dscan.json seeds with no bound on T's step: film k of n, and the first failures find-film-extra-E.json numbers for the cards on the average, the filter, the arrows, the waves, the sweep's factor and the raise find-film-extra-fix3.json std(u) round the film, and the formula for the frozen run's distance find-film-extra-fix4.json the free stretch at other lengths, and seed 1's lag against speed find_film_runs.py writes find-film-runs.json: the program with a switch for each experiment find_film_modes.py writes find-film-modes.json find_film_lab_extra.py, find_film_extra_cardsA.py, find_film_extra_cardsAstar.mjs, find_film_extra_cardsB.py, find_film_extra_groupc.py, find_film_extra_D.py, find_film_extra_D_scan.mjs, find_film_extra_E.py, find_film_extra_fix4.py each writes the file of the same name