{"n": 1, "what": "looked at the archive dog (kit/archive-dog, 50 frames, 384x288) as the material for a photographed ARAP dog", "saw": "the dog is ~130 px long, blurred and grey; its SAM mask merges the hind legs and drops the tail; at film size it reads as mush. The footage does contain a real turn (profile left -> front -> profile right)", "changed": "chose the other half of the lineage's theorem: an engraving. Drew the dog in code as ink lines on paper so ARAP's preservation of local structure is visible in the hatching; kept the archive only as proportion numbers"}
{"n": 2, "what": "engraving v1: per-part hatch, lambert tone from a sqrt-distance height field", "saw": "uniform line width (the height field was too flat to give any light), concentric 'fingerprint' rings on contour-hatched shoulders and haunches, U-shaped rings at leg tops", "changed": "puffy height h = sqrt(2*70*d); limb hatching uses the unclamped perpendicular distance (no rings at caps); shoulders, thighs, haunches, chest hatched by direction instead of contour"}
{"n": 3, "what": "mesh: Poisson-disk outline + interior samples, Delaunay, triangles kept only when centroid and 6 edge points are inside the drawing", "saw": "legs separated cleanly, ears separate; P_body 1,086 vertices / 1,653 triangles, P_head 265 / 420", "changed": "nothing"}
{"n": 4, "what": "ARAP pose tests (40 iterations, cold): rest, head low, cower, lie, chin rest, look in, stride", "saw": "0.05 s per solve; head low and cower read; lying had 11-19 inverted triangles in the folded hind legs and forelegs", "changed": "added knee handles placed by two-bone IK"}
{"n": 5, "what": "first full-scene stills: blue dusk light multiplied over the dog", "saw": "the dog and the dusk sky had the same value; the dog vanished into the railing", "changed": "separate light for figures (pale cool) and background (dark dusk); moved the porch post off the dog's watching place"}
{"n": 6, "what": "lying inversions traced", "saw": "wrist/ankle patches rotated the wrong way (+1.45 rad instead of clockwise); paw and ankle patches touched each other so triangles between two rigid patches could not be fixed by any solve", "changed": "joint patches now turn with the bones measured from the IK joints; paw and ankle patches separated; lying pose: 0 inversions"}
{"n": 7, "what": "first full render (warm-started, 8 iterations/frame) with an inversion repair that stiffened flipped triangles x3 forever", "saw": "62-140 inverted triangles in cowering and refusing frames: the stiffening compounded frame after frame and the pose (knee patches fixed at angle 0) was the cause, not the solver", "changed": "stopped the render; joint patches rotate with the bones; stiffening is capped and decays (x0.93 a frame); a small one-ring flip guard. Residual: single triangles in ~7% of frames"}
{"n": 8, "what": "full render 2 (joint patches turn with bones, capped decaying repair, flip guard)", "saw": "the look-away head at 3.54-3.63 was a crumpled 'beak': the reflected head solver started from an unreflected shape; energy spikes 0.8 in the walk from over-long hind steps; lie-down onset measured +0.58 s late because the rising head defines the top of the dog", "changed": "reflected head cold-starts from the rest head placed rigidly; step trigger/target 12/13 cm -> 9/9 cm; lie onset measured on the rendered shoulder patch; door beam brighter on the boards; the look-in stretches the neck 14%"}
{"n": 9, "what": "full render 3 viewed: contact sheet at 1 s, every 2nd frame of both turns, crops of walk and sill", "saw": "on the sill the back humped (shoulder and hip patches squeezed the spine by 50 mm) and the tail hung below the floor line; end card mesh not legible as triangles", "changed": "lying patches no longer squeeze; tail lies back along the boards; end card redrawn as wireframe with handles. Re-rendered 3.67-7.58 s and 23.2-32 s. Measured: all onsets within 0.125 s, inversions in 38/768 frames (49 triangles), head below the back in 100% of the approach, 0.56 of the lying dog inside the door frame; checker ALL OK"}
{"n": 10, "what": "critic round 1: DOES NOT PASS (look-in was a profile lift toward the woman; folded hind leg overlapping; view swaps were cuts; llama/deer silhouette; dotted fringe and moire)", "saw": "the five repair conditions", "changed": "new L_head (three-quarter from behind) cross-faded in on the score's look; lying is its own rest mesh P_lie (no fold; 0 inversions in all lying frames); view changes foreshorten, overlap two frames and arrive ARAP-widened; shorter neck, deeper chest; hatching 1.6x coarser, no cross-hatch, no scene wireframe, 3x supersampling, world cross-hatch removed. Full re-render: checker ALL OK, onsets unchanged, inversions 107 single triangles in 56 standing/walking frames, 0 lying"}
{"n": 11, "what": "critic round 2: DOES NOT PASS narrowly (ghosted crossfades at view changes; beak-like over-shoulder head; thin rod muzzle in the look-in; looping hind legs in the cower and the refusal crouch)", "saw": "the four repair conditions", "changed": "every frame one drawing: hard swaps at the narrowest foreshortened frame (no dissolve), lie/look swaps on the score's beat; no mirrored head in the turn away, the look-back head at a fixed level angle; L_head muzzle redrawn with mass and nose, aimed up into the room; cower hips -75 mm with dropped hocks, deep crouch hips -170 mm matching the lying mesh. Full re-render + 76-104; checked frames extracted from the encoded film; checker ALL OK; refuse onset +0.208"}
{"n": 12, "what": "critic round 3: DOES NOT PASS very narrowly (doubled rump/tail/thigh at 3.85-4.5; hind-leg knot at 7.6-7.8 rising)", "saw": "one layer per frame already; the doubling came from the near thigh's own inner outline and the tucked tail lying along the thigh; the knot from the standing mesh being driven into the start of the lying fold before the swap", "changed": "no inner outline on thigh and shoulder masses; tucked tail passes under behind the thigh; standing mesh clipped to the deep crouch, deep crouch shallower; heavier outline, lighter railing. Full re-render; crops of the encoded film at 3.88/4.00/4.25/4.50/7.62/7.75 checked; checker ALL OK"}