spec(t27f): reel_particles -- the t27-basics reel drawn by a .t27 particle engine on glossy black #10305

Boundary

specs/t27f/reel_particles.t27 (new), specs/t27f/reel_particles_mux.t27 (new), specs/automation/owner_voice_reel.t27 (restored). No other path.

Scenarios

• The owner asked for the t27-basics reel on a glossy black background, without the honeycomb cells, with the picture generated by a physical particle engine (2026-10-18).
• The engine spec lowers to Zig through the compiler's gen; the binary writes 450 raw RGB24 frames of 1080x1920 to stderr, which ffmpeg reads as rawvideo.
• The mux spec prints the ffmpeg filter graph (captions, voice, music, limiter) from the numbers of owner_voice_reel.t27.
• specs/automation/owner_voice_reel.t27 (merged in spec(automation): owner_voice_reel -- the 15 s owner-voice reel recipe as data (Closes #10279) #10281) was deleted by the unrelated PR spec(claw): preview -- the T27F preview deployment of a PR as law (Closes #10282) #10287; the mux spec reads it, so it comes back.

Requirements

• Every pixel is computed in .t27: 27 lesson particles on damped 3D springs (semi-implicit Euler) into a 9x3 grid, lit or draft from reel_t27_basics.STATUS; 72 dust particles on softened Kepler orbits (velocity Verlet) with the particle_field event hues; a glossy floor with mirrored reflections, a warm key light, vignette, tone roll-off and LCG grain.
• Deterministic: fixed DT = 1/60, 2 steps per frame, seeded LCG, no clock.
• No code in any other language: no JS, Python or shell logic; ffmpeg runs with plain flags.

Criteria

• The test report passes on both new specs and on the restored one, with 0 vacuous passes.
• The rendered reel is 15.000 s, 450 frames, 1080x1920, -14 LUFS integrated; the frame corner is near #0A0A0A.