Judging report

What works. The strongest runtime and slices. One clock (60 s cycle, 72 lub-dub beats, 12 breaths) drives everything. Pulses leave the heart on each stroke, travel down the vessels, and click the gear trains forward on arrival. Every one of 1,000 slices shows tissue, a vessel or a gear: none is 97% or more one colour. The two brief slices split the engine cap exactly at the seam. The world is built once per seed as plain data, separate from drawing, and validation names its reasons. Seeds visibly change palette and structure.

To fix. 1) Render into a new canvas (the probe's first 1000x1000 frame) -> it took 1,073 ms, and a cold 2000x2000 render took 2,038 ms -> expected well under 200 ms; the module-level caches (worlds, per-viewport still layers, per-scale gear and heart art) hide this when warm, but add state round 2 must respect. 2) View the whole world at 1000x1000 -> up to 40 similar-sized gear trains along the branches read as scattered, with little hierarchy below the heart -> expected fewer, larger organs that read at full view. 3) Try to view exactly 731 ms -> #time is a range slider and the arrow keys step one beat, so exact times are hard to reach -> expected a numeric field for exact milliseconds. 4) Play -> about 21% of pixels change every 200 ms, much of it ground glow -> consider a calmer ground so the pulses carry the rhythm.

What works. A real signal system: pulses run along the conduits at 0.6 units/ms, and each organ's piston, bellows or ring responds after its conduit's travel time, so motion links the heart to the limbs. Gear pairs mesh at matching pitch and speed ratios. The creature fills the whole 2000x2000 world, so only 16 of 1,000 slices are near-empty. Zoomed slices are drawn as vectors and stay sharp at 5x. The viewer is small, and Play tracks real time.

To fix. 1) View the default seed whole at 1000x1000 -> parts sit in dark teal on near-black, with tan gears and line work and orange only in the heart, so the organs read dim -> expected a wider value range so the organism reads at full view. 2) Apply seeds 0x11..., 0xab..., 0x00... and 0xff... at 2137 ms -> the same layout each time, only the tint changes (six fixed anchors jittered about 40 units in src/machine.ts) -> expected the seed to change structure (organ kinds, limb routing) while staying the same machine. 3) Render slice {x:0,y:0,width:50,height:80} -> a plain dark panel with a faint grid; 305 of 1,000 slices are 90% or more one colour -> expected more articulated detail in the margins. 4) Read src/draw.ts -> one 316-line function full of literal coordinates -> expected heart, organs and arteries as separate functions with named dimensions, so round 2 can extend them.

What works. The most polished poster: a dark body holding a breathing orange chamber, four flywheels with slider-cranks, looping tubes, gauges and labelled callouts, in a strong palette. One 12-second drive phase moves every part at fixed ratios, so the motion is calm and connected. The brief's two slices form one pinwheel across the seam. The browser test suite is thorough and traps Math.random and Date.now.

To fix. 1) Press Play -> each frame renders the whole world and the 250x400 detail; the detail rasterises a 4000x4000 world (about 400 ms per render in our headless probe), so we estimate playback at roughly 2 frames per second -> expected smooth playback: draw the detail at canvas resolution with a transform instead of upscaling a world raster. 2) Render the 1,000 round-2 slices -> 189 are 97% or more one colour (blank cream) and 558 are 90% or more; the corner and bottom-right edge slices are empty paper -> expected most slices to hold part of the machine. 3) Apply six seeds -> same layout; only the accent colour and phase change -> expected structural variation. 4) Read src/machine.ts -> layout and drawing are mixed in literal Path2D strings, and every call renders the whole world before cropping -> expected a seed-built world model drawn per viewport. Also, `let window` in src/viewer.ts shadows the global.

What works. A clean, well-composed engine: a turbine heart, crescent carapace and three satellite turbines, linked by belts and conduits in a considered palette. One 12-second shaft drives every part at fixed ratios and lags, so the motion is calm and legible, and Play tracks real time. The tiled renderer keeps slices matching the whole and sharp at 5x. The code splits cleanly into seed data, primitives, composition, runtime and viewer, with browser and server tests.

To fix. 1) Render the 1,000 round-2 slices -> 623 are 97% or more one colour (median dominant 99.8%); the corner, bottom-right edge and a random slice at {x:1503,y:377} are flat paper with 2 colours -> expected most slices to show part of the machine, so a buyer does not get a blank card. 2) Apply six seeds at 2137 ms -> nearly identical pictures; the seed only sets phase, rib count, aperture count and paper flecks -> expected visible variation in structure or palette. 3) View the whole world -> wide empty paper inside and around the orbit rings, and it reads as a clockwork diagram more than an organism -> expected the body to spread through the world. 4) Read src/machine.ts -> every part's coordinates are literals, and the seed data is thin -> expected a seed-built world model round 2 can vary.