9 waiting. Watch, then tell any agent to approve or reject by number. Everything already deployed, buried or from the first 48 hours is filed below.
The starburst mosaic again, now 48 cells not 24. Same script and same move as the 31 July film that is live on the page, so the shot is proven; only the cell count changed, because selection moved off the stale has_mesh column onto meshes_clean. New out= name on purpose: it must not touch the live 24 cell render or its _frames folder.
Full rainbow 2 mm fur, the look from the 4 Sept stills Amy picked. Beats 120/240 checked 6 Sept. Lights ride with camera.
v2 of job 19: area lights were reflecting in the glossy floor as a pale rectangle behind the brain for part of the turn. Lights now hidden from glossy rays; beats 120 and 200 checked clean 6 Sept. Existing frames moved aside by the queue.
Golden dot hologram, camera level with brain centre, 1.4x back (longer lens). Height sweep 0-100% checked as stills 5 Sept. ~14 s/frame.
Pial surface + 2.4M direction-coloured hair strands, Cycles, 360 turntable 15 s at 24 fps. Beats 100 and 250 checked at 1920x1080/128 samples: gyri legible, colours right, shadow catcher floor. 16 s/frame measured.
Retimed: opening hold 3.4s -> 1.0s and the second dead hold removed, measured frame difference was under 0.005 in both. Alpha switched BLEND -> HASHED: the white DN was dropping to zero visible pixels on single frames between 15-17s because EEVEE sorts blended surfaces per object. Labels come from shotB_beats.json which the render itself writes.
PAUSED at 388/576 frames by Amy 31 Jul 09:25. PNG sequence, resumes exactly where it stopped. Set status back to queued to restart.
camera ends tilted up, aiming mid column not below it. NO colour jitter, Amy does not want it: pure depth ramp.
Corrected: cells now fire on the binned reported preferred direction, so hue and compass agree. Ghost raised so the mosaic stays visible between sweeps.
Byte-for-byte identical to a file currently sitting in a project repo. Verified by hash, not by filename.
Full sequence: five recording fields, recoloured to cell type, non-selective cells fall away, then the bar sweeps. 937 frames, 39s. Each stage labelled with the key true of it. The 324 opening and closing frames are dissolves in linear light, not renders: nothing moves in this film except colour.
live at https://amyleesterling.github.io/retina/ as ds_mosaic.mp4
Cell-type orbit with the burned-in legend: every type with its swatch, full name and count, plus 'colour is identity, not depth'. Live on the cortex page.
live at https://amyleesterling.github.io/microns/ as cortex_by_type.mp4
24 starburst amacrine cells laid in one at a time, then rotated edge on to show they all sit in one thin band
Published to the retina page
live at https://amyleesterling.github.io/retina/ as sac_mosaic.mp4
Here lie the renders that did not make it. They are kept because a wrong version you can still watch is worth more than one you deleted, and because most of these were only wrong in a way nobody could see until it was rendered.
All 364 cells, all 106 DS, with a fade in and out of the whole population. 757 frames, 31.5 s. Intro and outro are cross-dissolves in linear light, not rendered frames: the camera and geometry never move in this film so they are pixel-identical to rendering, at no GPU cost.
cause of death superseded: only had a plain all-cells fade, before the cluster and cell-type stages existed
Cells coloured by predicted cell type instead of depth. 36 of 38 typed: 11 x 23P, 8 x 5P-IT, 7 x 4P, 7 x 5P-ET, 1 x 6P-IT, 1 Martinotti, 1 oligodendrocyte. The lamination still appears even though colour is now identity, not position.
cause of death unlabelled: colour meant cell type and nothing on screen said so
All 364 cells, all 106 DS, on the rebuilt meshes_clean library. 43.1M faces. Test still at frame 86 checked: dendrites continuous, no beading. SLOW: 55-60 s/frame plus a 29 min import, about 10 hours. PNG sequence, resumable.
cause of death no overlay: the queue rendered it but never ran the overlay step, so it shipped with no colour key, compass or readout
ghost wireframe; each cell takes colour only at its own best direction. readout, compass and key.
cause of death hue and compass did not agree: 29 of 64 cells lit in a colour more than a full 45 degree step from the direction the compass read
202 calcium-imaged cells, CA3 palette, somas to camera, with readout overlays
cause of death first version
180 calcium-imaged cells, hue = measured preferred direction, brightness = own tuning curve. For a 10am meeting.
cause of death natural causes, superseded
first pass: rainbow palette, somas facing away. superseded
cause of death old version
Everything made before 31 July 2026, when this whole pipeline was two days old.
orbit 72 deg then descend, per-area decimated
block to cell to thorn to synapse
feedforward inhibition, 7 beats
sixth of a turn across the convergence gradient
widescreen master, the synapse sequence
widescreen master, populations in circuit order
six mossy fibres, one fails and six together succeed
56 partner segments in purple