Report: clock (grandfather clock)

Files: dogfood/clock.aix (97 lines), dogfood/clock_reimport.aix, outputs in out/clock/ (sheet, slices, steps, poses, anim_tick.png, model.glb 5.8 MB, model.stl 9.2 MB, model.obj, report.md) and out/clock_beauty/beauty.png (1024, pose show). 15 renders of the model, about 10 checks.

1. The model

A grandfather clock 1.94 × 5.85 × 1.16 units, front on +z, walnut throughout, 41 named steps:

• Plinth: base_mould + plinth + plinth_cap, three rounded boxes stacked and overlapped.
• Waist: waist_solid (1.3 × 3 × 0.8) → shell(0.08) → - door_cut through the front wall only; a door_frame (an extruded rect-minus-rect, "z"); glass = a 0.045 pane sunk into the front wall, material("glass", transmit=0.95). The inner walls get a decal(box…, "tan") so the interior is not black in the beauty render.
• Pendulum: rod (r 0.035, reaching 0.05 into the waist's top wall) + bob (cylinder r 0.27 turned to face +z) + bob_boss, wrapped in joint(..., "pendulum", 0, 3.86, 0).
• Hood: hood_solid - hood_recess (a 0.06 recess in the front face), two turned columns (mirror), cornice_1, cornice_2, pediment, brass finial.
• Dial: dial_plate brass cylinder r 0.5 facing +z, sunk 0.045 into the recess floor; chapter_region (ring 0.31..0.49) decal in ivory, tick_region (60 boxes via ring) and numeral_region (twelve Roman numerals, each text(...) extruded "z", moved to radius 0.355 and rotate(z=-30*i), so they read radially) decals in black. bezel torus, hub cylinder bridging both hands.
• Hands: hour_hand and minute_hand are extruded polygons pointing to 12 at rest, each a joint at the dial's centre (hour at z 0.47, minute at z 0.565, separated by a 0.04 air gap and joined through the hub).
• Maker's plate: brass plate 1.0 × 0.2 under the dial, maker_text "FABLE" in serif raised 0.05.
• Poses: rest, three (hour −90°), ten_ten (hour +55°, minute −60°), swing (+8°), swing_back (−8°), show (10:10 with the pendulum at −8°); animation tick swing→swing_back→swing over 2 s.

Verified in pictures: out/clock_three/front.png (hour on III, minute on XII), out/clock_dial/sheet.png (10:10 at cell 0.006, numerals legible, ticks and chapter ring right), out/clock_maker/sheet.png (FABLE clean, close-up watertight yes at 0.005), out/clock/slices.png (hollow waist, rod and bob inside, the pane in the front wall), the 1024 beauty (dial and plate read, pendulum visible through the glass).

Final check: no warnings. Sizes: base 1.9 × 0.18 × 1.12 … hood 1.94 × 2 × 1.12 (y 3.85..5.85), dial 1 × 1 × 0.08 at z 0.38..0.46, hands 0.09 × 0.41 × 0.06 and 0.08 × 0.55 × 0.05, clock 1.94 × 5.85 × 1.12, grid 256 → cell 0.0229.

2. Print readiness and re-import

From out/clock/report.md (rest pose, the exported one):

Row	Value
Watertight	yes
Pieces	1
Stands	yes, centre of mass 0.555 inside the footprint (18-sided hull)
Overhangs	4.5% of the surface faces down more than 45°
Cavities	volume 2.079 at (0, 2.405, 0.005) — the glazed waist, sealed by design (not drained)
Volume	5.512, triangles 184 596

Re-import (dogfood/clock_reimport.aix, import("../out/clock/model.glb", resolution=160), 17.4 s to sample in check and again in render):

	source	re-import
Size	1.94 × 5.85 × 1.16	1.941 × 5.85 × 1.16 (z −0.56..0.6 in check, padded)
Volume	5.512	5.491
Pieces	1	1
Cavities	2.079	2.082
Stands	yes	yes (0.552 inside)
Overhangs	4.5%	3.7%
Watertight	yes	no: 21 edges, "mostly 4 at (−0.031, 4.689, 0.532)" — all at the hub and hands, which are 0.05–0.06 thick and the import's cell is 0.037

What survived: every wooden part, bezel, dial disc, the raised FABLE, the hands (as bumps), the finial. Lost, as expected from positions-and-triangles-only: all materials, so the numerals, ticks and chapter ring (decals) and the glass are gone; the door reads as a recessed solid panel.

3. Friction

1. Line 23 (now 28), glass — warning: 'glass' (line 23) is only 0.03 units thin, 1.128 of the 0.027 cell: it may be missing or broken in the mesh. Thicken it or raise the grid (set grid 488). Expected a 0.03 pane to be fine at 1.1 cells; the threshold is 1.2. Thickened to 0.045. Clear message, one iteration.
2. Line 79 (now 84), maker_text — warning: 'maker_text' (line 79) has gaps (a letter's counters, the space between letters, a slot) only 0.005 wide, 0.175 of the 0.027 cell: they close up in the mesh. Use a larger size, a lighter weight or more spacing=, or raise the grid (set grid 512). The message does not say which gap. I probed: spacing=0.07 left it at 0.022 (so a counter, not spacing); weight=0.026 cleared it but gave has a wall, tube (at its thin end, if tapered) or stroke only 0.026 thick, 0.978 of the 0.027 cell; size 0.13/weight 0.03 gave 1.128 of the cell. Five checks and three name/size changes ("FABLE & CO" → "FABLE", 0.07 → 0.14) to land on size 0.14, weight 0.032, which passes. The plate had to grow to 1.0 × 0.2 to hold it.
3. Glass is opaque on every diagnostic picture. sheet.png, poses.png and anim_tick.png (8 frames) show a grey pane and nothing behind it; the animation strip shows a clock in which nothing moves. Only slices.png (Z = 0 cut) and the beauty render show the pendulum. Expected the strip to show the swing. I verified the swing from the beauty close-up instead (--focus glass --beauty --pose show), two extra renders.
4. Wrong steps blamed for non-watertight edges. out/clock_dial/report.md (--focus bezel --pose ten_ten): Watertight | no: 4 edges ... 1 at (0.386, 4.946, 0.537) in 'dial_plate', 'hood_solid'. z 0.537 is neither the dial (front 0.46) nor the hood (front 0.475); it is the minute hand's pointed tip. Later, --pose show at grid 220: 1 at (-0.195, 4.06, 0.49) in 'hood_solid', 'hood_recess'; 1 at (-0.196, 4.054, 0.504) in 'plate', 'maker_text' — both are the A's counter in maker_text. The blame sent me looking at the hood recess first. Fixed by blunting the hand tips (polygon tips 0.032 wide, line 77–78) and grid 256.
5. A pointed polygon tip is not warned about. check warns for tubes, strokes and shells under a cell but said nothing about the hands' 0-width tips (polygon(..., 0,0.47, ...), original line 73), which produced the edges in item 4. Expected the same "thin end" warning tubes get.
6. Two hands at rest both pointing to 12 — line 77–78. First version had hour 0.462..0.507 and minute 0.513..0.557 with the dial front at 0.46: 1 at (-0.064, 4.723, 0.513) in 'dial_plate', 'hour_hand'. The guide's "overlap by a cell" advice cannot apply to two thin plates that must stay separate parts of a rig; I had to leave a 0.04 gap and bridge it with a hub. Worth a line in the rigs section.
7. Interior lighting. First beauty (out/clock_qb, 320 px): the door was a black rectangle. Three beauty renders (quick, 640, --focus glass) to arrive at set light_azimuth -20, set light_elevation 30, set ambient 1.8, a tan decal on the inner walls (line 26) and swinging the show pose to −8° so the bob sits on the lit side. The bob is still in shadow in the final picture. Nothing in the docs says how deep a key light at elevation 55 reaches into an opening.
8. --quick is of little use on this model: warning: quick pass: 20 steps are thinner than this pass's 0.091 cell ... (plinth_cap, waist_shell, door_frame, glass, rod, bob, ...). Half the model is thinner than the quick cell, so every judgement needed the full render (12 s at grid 220, ~20 s at 256). Fine, but the quick loop the skill promises did not apply.
9. poses.png thumbnails cannot show a 0.05-thick hand (out/clock/poses.png): six identical-looking clocks. Documented ("meshed coarsely"), but a --focus per pose was the only way to judge the times: two more renders.
10. Re-import watertight "no" (item in section 2): 21 edges ... in 'back' with no way to tell from the report that they are at the hands rather than a real export defect; I inferred it from the coordinates (y 4.69, z 0.53).
11. Minor: report.md lists the material I named blued by its variable name (Materials | walnut, tan, brass, steel, glass*, ivory, black, blued), which is nice, but the * on glass is unexplained.

4. What worked well

check is fast and exact, and the size table caught every placement before a render. Decals for the chapter ring, sixty ticks and twelve numerals cost nothing in geometry, survive into the atlas, and read at 1024. joint/pose was trivial: hands and pendulum right first time, and --focus bezel --pose three proved the 3:00 pose at cell 0.006 with its own watertight row. shell + one cutter gave the glazed waist; the slices proved it. The beauty render's glass refracts and shows the pendulum, and the report's stands/pieces/cavities rows answered print-readiness directly.

5. Three changes that would have helped most

1. Name the gap in the lettering warning: "counter of 'A' 0.022 wide" versus "gap between F and A", and offer a concrete pair that clears it ("at weight 0.032 size ≥ 0.14"). Item 2 cost five checks.
2. Draw transmit materials see-through (or as an outline) on the sheet, poses.png and the animation strips, or at least say on the strip that a step is behind glass. Item 3 made the animation strip and pose sheet useless for the one moving part.
3. Blame the step whose surface actually holds a non-watertight edge, and warn in check about a polygon or profile tip thinner than a cell the way tubes and strokes are warned. Items 4 and 5 were the only misdirections in the whole loop.