Mechanism fix ledger, mechanisms-finalize

Three review agents (geometry verifier / mechanical loads / assembly walk) audited commit fa13f64 and converged on ~20 verified defects. Four fix stages run sequentially (they all edit src/build.py); stage 4 is an independent re-verification. Status legend: [x] fixed+probed, [~] in progress, [ ] queued, (!) accepted-as-is with reason.

Stage 1, chassis + pan stack [DONE, all probes pass]

• [x] No chassis deck (worst defect: race/balls/platform floated; whole head load hung on the pan motor's 5 mm plastic shaft). Cavity now stops deck_t=20 below the top (5 mm of material under the race seat); seat floor z=37 carries the ring.
• [x] Race grooves too deep + oversize (platform sank 3.0 mm onto the ring; ring groove floor was 0.1 mm). Grooves engage 1.8 mm of ball each, oversize 0.2; stack derived in one helper (_pan_stack); plate 7.6 thick, top flush at z=52; tangency probe error 0.0000.
• [x] No uplift retention, 3 L-clips at 120° screwed to deck pockets, engaging a platform rim REBATE below deck level (top-side clips would be sheared by the neck sweep, measured r≤62.9).
• [x] Pan D-bore jammed on the round shaft section, D-profile now only over the flat zone (z 44.5..52), Ø5.3 round counterbore below; hub stops 0.3 above the motor's Ø9.1 boss.
• [x] Platform overconstrained (located by shaft AND race): D-bore arcs opened +0.27/side, flats snug: race locates, shaft only drives. Hub grub deleted.
• [x] Pan motor pad/ear mismatch (can sank 5.5 into the pad, ears floated 12.4 above pilots), pedestal top at the ear plane, Ø29 can relief, Ø2.5 pilots, wiring-box relief.
• [x] Neck bolts broke the platform rim + all pilots were Ø3.5 clearance (nothing bites), bolt circle r16→12, neck pilots Ø2.5×12, underside counterbores for the heads.
• [x] JST won't pass Ø12 bores, platform + neck passes are 16×8 obrounds.
• (!) Cable exit relocated to (12,−24): the spec position (0,−38) punches the fixed race ring.
• (!) Marginal, unfixed: outer neck-bolt counterbores nick the groove outer flank ~0.9 mm (outside ball contact); neck bottom coplanar with deck top (zero-gap rub when panning).

Stage 2, tilt drive + neck + Pi + axle [DONE, all probes pass]

• [x] 695 bearing seats sealed inside the cheeks, open flush to the inner faces (x=±18), bore Ø12.85. Probe: Ø12.7 test plug vs neck = 0.000 mm³ both sides (insertable).
• [x] Worm wheel freewheels / rubs / unconstrained axially, width 7 centered on x=0, grub-keyed hub (M3 pilot), Ø8 spacer tubes to both 695 inner races. Probe: wheel spans x −18.00..+18.00 (races at ±18), wheel-vs-neck = 0.000 mm³.
• [x] Tilt speed 16 s per 60°, worm_wheel_teeth 24→12; center distance factored into worm_cd() (one helper, both former call sites). Probe: teeth=12, cd=11.50.
• [x] Worm undriveable, full-depth double-D bore (vol 845 vs 942 unbored), worm_len 16, Ø6 tail stub riding a Ø6.4 bushing post off the bracket. Probe: worm-vs-neck = 0.000 mm³.
• [x] Tilt motor buried + ear holes 7.9 mm off, Ø29 can pocket + wiring-box and ear-bar reliefs; motor ROLLED about its shaft so the 7.875 offset points UP (can hangs below the worm axis, clear of the back-wall sweep; ears horizontal at the CAN axis); gear face on the plate back face. Probe: motor-vs-neck = 0.000 mm³.
• [x] Pi 5 inside the mechanism volume, Pi now PORTRAIT (85 edge vertical) and offset to x −84..−28: everything on the neck stays inside |x|≤26, so no tilt angle can reach it. Standoffs moved to the new 49(X)×58(Z) pattern, 3 mm long, all four rooted outside the slot; pivot bosses shortened 24→12 so the −X one clears the board. Probe: pi vs neck/motor/worm/bezel/back/axle/screen = 0.000 mm³ at −30/0/+30 (and every 5° between).
• [x] Axle end-grip topology, Ø7 clamp-boss torque tubes from x=±27 out to the side walls, Ø2.5 grub pilots at x=±30 (driven from below through the slot), snug Ø5.1 bore at the bosses, far-wall bores demoted to loose Ø5.3. Probe: boss material 489 mm³/side; a Ø5.24 plug bites the boss 4.97 mm³ (grips) and the wall 0.00 (loose slip fit).
• [x] Tilt envelope −11°..+19° vs ±30° spec, full ±30 achieved, no ±25 fallback. Neck slot narrowed to 62 wide (cheeks end at |x|=26; Pi bosses keep wall roots) but raised to z=191; the bottom-CENTER bezel↔back post swept into the neck column, replaced by a pair at x=±40. Probe: sweep −30..+30 at 5°, 14 fixed-vs-head pairs, all 0.00 mm³.
• [x] Tilt ULN2003 had no mount, 4× Ø6 standoffs + M3 pilots on the column back face (35×32 pattern centered z=110). Probe: 746 mm³ standoff material behind the back face.

Stage-2 side effects for Stage 3: the Pi is portrait at (x −56, y −16, z 192.5) → re-aim the back-cover I/O slot for ports exiting the board's ±X/±Z edges there; bezel↔back fastening is now 7 posts (M3×35 ×7, nuts +1); pivot bosses are 12 long (they still protrude past the outer wall, the Stage 3 trim item stands).

Stage 3, head/screen/camera + tracks [DONE: head/camera 3a, tracks/drive 3b]

Head/camera fixed by the Stage-3a agent (attempt 2). Combined screen+Pi model swap (user design change) done FIRST; several fixes below were re-scoped by what the new mesh + official CM3 data proved.

• [x] Screen+Pi combined reference model, screen_ref_stl → ..._-_Pins_Out_v8.stl (display WITH the Pi on its own 58×49 back standoffs; bbox 192.96 × 38.01 × 110.76, 13.02 deeper on the back). load_screen() now anchors the GLASS FACE at local y=12.494 instead of the bbox centroid (which the deeper back dragged 6.51 forward). Probe: glass world y old 30.9940 / new 30.9940, delta 0.0000; all 4 scr_mount_pts vertex clouds delta 0.00; W/H bounds delta 0. The separate pi5 placeholder part and the stage-2 Pi standoffs in head_back are REMOVED (obsolete).
• [x] Screen retention → REAR standoffs (re-scoped from "shorten bezel bosses"): the factory 126.2×65.65 holes open BACKWARD; the old bezel bosses ran forward from the tab plane (25.03) through the GLASS (front 30.99; 69/50k screen samples inside them). Now 4× Ø9 standoffs on head_back (inner wall −27 → tab plane 25.03, at x=±63 they clear the Pi stack |x|≤48.1, tubes, louvres, both slots), each with a Ø6.5 driver channel through the back wall + 8 mm M3 seat (stock M3×12). Probe: material beyond the tab plane 0.000 ×4, 215.6 mm³ seat each.
• [x] Camera aperture per OFFICIAL CM3 (ledger's Ø10→Ø18 guess superseded by RP-008153-DS): Ø6.3 through-bore, countersink to Ø8.0 at the outer face at 45°/side, cam_lens_dz kept 2.47. Probe: 75° diagonal FoV cone (Ø2.63 pupil 3 mm behind the face) vs bezel = 0.0000 mm³.
• [x] CM3 does not fit the old forehead, bay redesigned (proven, not styled): the stack needs ~14 mm between the window top (229.9) and the interior ceiling; 243 top gave 9.1, so the board + connector punched the top wall 347.6 mm³, all 4 M2 bosses floated in the pocket void and punched the display panel band (measured y 25.03..30.99), and the barrel crossed the module's top edge. Fix: body_z_top 243 → 251 (+8 forehead), cam_lens_z 233.5 → 237.0, bosses moved onto a ceiling-hung PIER behind the panel top strip (cam_pier_*, 0.53 clear of the panel; AF-housing cutout; 1 mm bosses, blind Ø1.7 pilots). Probe: cam|bezel/back/cover = 0.0000, cover|bezel = 0.0000, board top 246.46 vs ceiling 247.0, barrel bottom 234.12 vs module top 233.4, screen-samples inside cam/cover = 0, FoV cone 0.0000.
• [x] cam_cover fastening, 2× M2 through the cover into the diagonal boss pilots (cam_m2_clear_r now used) + ribbon pinch slot; cover trimmed to board height (a +2 skirt punched the new ceiling). Probe: pilot voids 0.00 (fully drilled), self-tap annulus present, cover watertight.
• [x] Glass-lip interference (found by sampling, not in the original ledger): glass front 30.99 sits FLUSH with the face (31), so the 0.5-deep retaining lip interpenetrated it 0.49 mm, the module could not seat on its factory-screw mounts. The pocket now pierces the face (front 31.1): flush edge-to-edge glass, pocket side walls locate, factory screws retain. Probe: the lip-band screen-in-bezel sample cells are gone.
• [x] Pi I/O slot re-aimed at the REAL Pi (combined mesh, landscape on the display standoffs, stack x −37.7..48.1, y −7.0..+5.5, z 151..207.5): ETH+USB short edge faces +X → slot through the RIGHT side wall at y −8.5..6.5, z 191.5..208.5 (the band 165..191 belongs to the pivot boss). USB-C + 2× HDMI exit the BOTTOM long edge into the open interior → route out the existing bottom-rear neck slot/cable port. Old left-wall + top-wall slots removed (nothing exits there; GPIO pins point −Y). Probe: residual wall in slot path bezel/back = 0.0000.
• [x] Bezel side posts pulled in 5 mm (stage-2 agent had done it; verified) + layout reworked: the top-CENTER post's M3 shank ran through the CM3 board (x=0, z=238.4) and the raised ceiling left it unrooted → top posts are now a PAIR at x=±40 fused to the new ceiling; side posts at 0.75·hh (z≈223), above the new I/O slot. 8 posts total (M3×35 ×8, nuts ×8). Probe: bezel x-bounds exactly ±102.50.
• [x] Pivot bosses trimmed flush (stage-2 shortening already left nothing proud; verified). Probe: bezel/back material outside |x|=102.5 = 0.0000 both sides; bounds ±102.50.
• [x] Regression: head parts tilt sweep −30/0/+30 vs neck = 0.0000 all angles; bezel/back/cover watertight; EXPORT=1 rewrites all STLs (bezel 93818 mm³, back 178880 mm³, cover 1317 mm³).

[x] RESOLVED (Stage 2R): combined screen+Pi stack vs the TILT MECHANISM. User call: keep the Pi on the display back, rework the tilt geometry. The tilt axle moved BACK 18 mm to (y −18, z 178) (tilt_axis_y new param; tilt_axis_z unchanged; screen anchored separately via screen_cz, it did not move). Everything keyed off the axle moved with it: cheeks, 695 seats, clamp tubes/hubs/grubs, wheel, axle, worm group. Consequential rework, all probe-driven: cheek boxes now stop AT the axle + a Ø19 bearing hoop (kills the pre-existing 260-sample cheek-top overshoot into the display back; worst cheektop→display gap now 3.12 mm @ −30°); worm group sits 4 mm further behind the axle (face offset 4→8) with the bracket plate shortened 46→36 tall (old top clipped the head back wall at +10..+25°); the outboard bushing post, which both crossed the −30° swept stack and clipped the wheel-tooth circle, became an open-top cradle (arm+riser+pad hugging the measured swept-stack profile; worm separation force presses the Ø5 tail stub down into the groove, so open-top is load-correct); gusset front pulled to y −20; column front-top notch (y −21..−13, z 94..112) clears the head bottom-wall arc at −30°. Probes (12k stack surface samples + manifold booleans): stack vs {neck, worm, motor, axle, wheel, head shell} intersection 0.000 mm³ at every tilt −30..+30 step 5; min gaps stack→neck 1.62 mm @ −30°, stack→worm 2.71, stack→motor 16.8, stack→axle 8.5, stack→wheel 11.0 (all ≥ the 1.0 mm target); real bezel+back sweep vs neck+worm+motor 0.000 at all 13 angles; worm cd = 11.500 exact; wheel|cheeks 0.000; Ø12.85 seat plugs 0.000 both sides; neck_clevis/head_back/head_bezel watertight, EXPORT=1 clean. Tracks/drive fixed by the Stage-3b agent. Measured-param corrections applied first (audit corr. 1/3/4/5 + reference/tt-motor-1079893/NOTES.md): sprocket_outer_d 42.0 → 37.6, track_wheel_r 19.0 → 19.32 (exact 12T × 10.0 polygon), pin bores Ø2.0 (new track_pin_bore_d), track_pitch comment fixed (reference pitch is 9.65; our 10.0 is a re-model by design).

• [x] Track loop below ground, wheel centres raised to _track_zc() = R + 4.5 + 1.5 = 25.32 so the bottom-run grouser face IS the ground line; road wheels + idler ride the knuckle crowns (r 15.82 about the pins) with 0.12/0.10 running clearance, since the sprocket needs the central ±4.9 channel open and the pad inner web (radial 21.12) is above the crowns. Probe: loop min z 0.0000 both pods, sprocket min z 6.52, roadwheels/idler vs links 0.0000.
• [x] Sprocket bore → TT double-D (_sprocket, not dbore_neg): socket Ø5.65 / 3.85 flat gap (print clearance on the measured Ø5.40/3.70), 8.0 deep = the flat length; tooth boxes truncated to the exact 18.80 tip circle; inboard hub OD12 reaches x=±58.3 through the wall web; Ø6 free bore + Ø9 × 1.7 counterbore on the outer face for the M2 retaining screw + washer into the shaft tip's Ø2 axial hole. Probe: tip r 18.800, sprocket|links 0.0000, sprocket watertight.
• [x] TT motor rebuilt measured + real wall mount, motor_tt() now: gearbox 36.80 × 22.40 × 18.64 + Ø22.4 collar (flat on the shaft side) + can Ø20/14.99 AF + Ø9.9 × 2.2 boss + shaft Ø5.40/3.70 flats 8.8 proud, PERPENDICULAR to the body, 11.5 behind the front face. Motor sits INSIDE the chassis (gearbox face 0.1 off the wall inner face), shaft axis = sprocket axis. Chassis mount per NOTES.md: Ø8 wall pass (clears the Ø7.2 boss), outer Ø17 recess → 3 mm web, 2× Ø3.2 M3 through gearbox + wall at 17.5 c-c (nuts in the pod gap), Ø4.2 × 2.2 nub pocket (y −49), tab pocket + Ø2.8 hole in the rear wall (y −74), deck pocket to z 36.8 (seat floor 37, race ring r34..46 clear at r≥50.2), cavity-corner relief; ULN standoffs shifted +20 Y and vent i=−3 dropped (both collided with the motor envelope). Probe: motor|chassis 0.0000, motor|pod 0.0000, shaft-vs-sprocket axis delta 0.0000 both sides.
• [x] Links re-modeled articulated (_track_link): pad web + grouser + interleaved knuckle combs (near A ±9.4..14 / far B ±4.9..8.9, X-disjoint at the shared pin so articulation can never bind), Ø2.0 pin bores for Ø1.75 filament, 45° inner-face draft chamfers at the web ends. Probe: all 36 adjacent pairs 0.0000 mm³ both pods (worst 0.0000), link watertight, track_L/R export as 52 watertight bodies each.
• [x] Idler tension slot + F688ZZ seat, idler wheel r 15.7 × 18 wide with Ø15.95 press seat through, Ø18.5 × 1.0 flange recess on the inboard face; chassis grows a wall arm + Ø28 slotted plate inside the front loop arc (everything radial < 15.7 so the wrapping links clear), obround Ø8.2 slot with ±idler_slot/2 = ±2 Y-travel for the Ø8 stub axle (hardware; M3 set-screw lock). Probe: seat r 7.9750 / recess r 9.250 exact, idler|links 0.0000, chassis|pod 0.0000.

Notes for stage 4 / buy list: - The tt-motor reference CAD is SINGLE-shaft; purchased TT motors are often dual-shaft. If the real unit has a rear shaft, the far pod side needs a Ø7.5+ clearance pocket mirrored about the gearbox (currently nothing is there, the pods have no outboard wall, so it only matters if a pod wall/cover is added). Verify the actual motor before closing that side. - body_z_top grew to 251 in stage 3a (head +8, CoM higher): size the chassis ballast for the taller head when the buy list is done, keep the mass in the chassis floor, wheelbase is 120.

Stage 4, independent re-verification [DONE: NOT READY, 2 blocking + 2 minor defects]

Independent probes (own scripts, scratchpad/stage4_*.py), nothing reused from stages 1-3b. Method note that invalidates several earlier "0.000" claims: the screen reference mesh is NOT watertight, and trimesh.boolean with engine=manifold RAISES on it. The earlier probes' ivol() caught that exception and returned 0.0, so every prior "stack vs X = 0.000 mm3" boolean was vacuous (their ProximityQuery gap numbers were real). Stage 4 re-tested all screen pairs by surface-sample containment (30k-400k samples) against the watertight partner.

Category results

DEFECTS

• D1 [FIXED, stage 5] (BLOCKER, stage 3a) - screen rear standoffs punch the display's own mount bosses. The Pins-Out reference mesh has a raised boss/tab structure around EACH factory hole spanning y 22.53..25.03 (2.5 mm), annulus r 1.5..8.1 about the hole axis. The Ø9 head_back standoffs run solid to y=25.03 (the hole plane, which is the FRONT of that boss, not its base), so at all 4 mounts an annular band r 1.75..4.49 x 2.5 mm deep interpenetrates (386/6.2k local surface samples inside head_back, max depth 1.37 mm, identical at all four sites). The screen cannot seat; it would stand 2.5 mm proud. Fix direction: stop the standoffs at y=22.53 or counterbore them to swallow the display boss. Stage 3a's "material beyond the tab plane 0.000" probe only checked y>25.03 and missed this.
• D2 [FIXED, stage 5] (BLOCKER, stage 2R) - tilt worm interpenetrates the open-top cradle: 97.9 mm3, bounds x -5..+5, y -24.5..-14.5, z 161.25..166.5. The cradle arm/riser/pad band (y -23..-11, groove r 2.7) was sized for the Ø5 tail stub, but the worm carries full-radius threads (r 5.34) all the way to y=-13.5; the bare stub emerges only for the last 0.5 mm (worm ends y=-13.0). About 9.5 of the 12 mm cradle band sits under the thread envelope; the worm cannot rotate. Root cause: the stub (y -21..-13) is modeled INSIDE the threaded length; stub and cradle must both move past the thread end (or the worm shortens).
• D3 [FIXED, stage 5] (minor, stage 2) - worm double-D bore clocked 90 deg to the motor shaft flats: 17.4 mm3, bounds x -1.5..+1.5, y -30.5..-26.25 (the shaft's round shoulders buried in worm core). After the motor's two rotations its flats face +-X, but db.apply_transform(R(TAU/4,(0,1,0))) leaves the bore flats facing +-Z (the in-code comment claims +-Z is correct; it is not). One-line clocking fix; the placeholder encodes assembly intent so it should be right.
• D4 [FIXED, stage 5] (minor, stage 2) - neck_clevis is 5 disjoint bodies. The 4 tilt-ULN standoffs (Ø6x8 at x +-17.5, z 94/126) span y -69..-61 and are only FACE-TANGENT to the column back face (y=-61); uni() does not fuse tangent solids. 186.6 mm3 each, floating in the STL. Push them 1-2 mm into the column.

Notes (no geometry change required)

• N1: pan D-flat engagement is 3.5 mm effective, not 7.5: the Ø5.3 round counterbore is cut d_bot - hub_bot + 4 long and overshoots 2 mm into the flat zone (to z 46.5; shaft flats span z 44..50). Works, but half of what stage 1 claimed; trim the +4 if more bite is wanted.
• N2: the cable corridor from the neck-channel mouth (z=150) into the head necks down to a ~Ø5 free disc (max inscribed radius 2.5 mm at z=160), squeezing between the tilt-motor can (blocks x<~14 up to z~173) and the +X cheek (blocks x>18). The documented Ø3.6 power pair passes with a jog; nothing thicker does. A straight Ø7 bundle does NOT fit (768 mm3 vs the motor can).
• N3: chassis cavity furniture at (x 8.5..14.7, y -24..-20.5, z<=30.25) sits 1.7 mm below the deck cable-pass mouth (straight Ø7 rod clips it 18.8 mm3); a flexible bundle routes around.

Docs drift (CLAUDE.md; found, not fixed)

1. "9 watertight per-part STLs" (l.11) and the "9 printed parts" list (l.178): now 10; pan_clips missing from both lists.
2. Worm wheel "24T" at l.68, l.121, l.144: build uses worm_wheel_teeth = 12.
3. Key numbers (l.114-117): screen "measured ... 193.0 x 25.0 x 110.8" + "Loaded live from ..._v12.stl": the build loads ..._-_Pins_Out_v8.stl (combined display+Pi, 192.96 x 38.01 x 110.76). PARAMS["screen_d"]=25.0 no longer matches the loaded mesh either.
4. "Overall assembly bbox = 221 x 208 x 248" (l.117): measured 209 x 170.5 x 251 at neutral.
5. "bezel<->back: 6 perimeter posts" (l.183): now 8 posts (stage 3a).
6. "screen: 4x M3 bosses on the bezel" (l.184): retention moved to 4 rear standoffs on head_back with back-wall driver channels (stage 3a).
7. "Pi 5: 4x M2.5 standoffs on the back cover (58x49)" (l.186): removed; the Pi rides the display's own standoffs inside the combined reference mesh.
8. The stage-2R tilt-axle relocation to y=-18 (tilt_axis_y, screen_cz decoupling) is absent from Mechanical intent / Key numbers (only z=178 is stated).
9. Stage-1 claim above, "D-profile now only over the flat zone (z 44.5..52)": effective D zone is 46.5..52 (see N1).

Verdict

NOT READY TO COMMIT. Blockers: D1 (screen seating), D2 (worm/cradle). D3 + D4 are small but should ride along; docs drift items 1-8 belong in the same pass.

Accepted findings (no geometry change)

• Worm lead angle 9.46°: self-locks dry, backdrives greased, but the 28BYJ's internal 64:1 train never backdrives, so the de-energized hold stands regardless. (Honest-worm option: worm_od 14.)
• Hollow-axle cable routing at Ø2.5 is fiction for a 5 A pair, the real route is the bottom-rear slot beside the clevis (tilt is only ±30°); the bore stays as weight relief. Docs updated.
• 18 uncaged BBs: grease-stick at assembly; a printed cage ring is a nice-to-have part.
• Axle: buy Ø5 brass/steel tube (K&S 300 mm stock); a printed axle is unusable (5.9 mm sag).

Fastener BOM (from the assembly review; re-check after stage 3)

M3×35 ×8 (bezel↔back, stage-3a layout) · M3 nut ×11 · M3×12 ×4 (screen, from behind through the Ø6.5 back-wall channels into the display's threaded pan holes) · M3×16 ×3 (platform→neck) · M3×8 ×2 (pan ears) · M3 grub ×4-5 · M4×10+nut ×2 (tilt ears) · M2.5×8+nut ×4 (Pi) · M2×8 self-tap ×2+2 (camera + cover) · 695-2RS ×2 (owned) · F688ZZ flanged bearing ×2 (8×16×5, flange Ø18) + Ø8 stub axle ×2 + M2×16 self-tap ×2 (sprocket retain) · 6 mm BB ×18 · Ø1.75 filament pins ×72 · Ø5×209 tube ×1 · 28BYJ-48 ×2 + ULN2003 ×2 (owned) · TT gearmotor ×2 (own 1, buy 1).

Stage 5, fix pass (D1-D4) + neck-on-platform [DONE]

• [x] Neck base footprint r62.9 overhangs the spinning platform (r48; solid top r45 inside the clip rebate) and rides the FIXED deck with zero gap. Move the column in (neck_y -38 -> ~-14) so the whole footprint sits on the platform at r<=44. The tilt axle (y=-18, z=178) and the head DO NOT move (stage-2R world-coord parameterization). Move with it: 3x M3 bolt circle, platform cable slot, deck cable pass alignment; deepen the column front notch for the head chin arc at -30 deg. Re-probe: pan sweep vs deck/clips, tilt sweep, cable route, watertight.

Stage 5 results (fresh probes, scratchpad st5_*.py; screen checked by surface samples,

booleans raise on failure)

• D1 fixed - standoff bearing face moved from the hole plane to the display boss REAR plane: boss lip measured 2.500 (y 22.534..25.034, identical all 4 mounts, annulus census); new face y=22.480 (lip 2.5 + 0.05 seat clearance, scr_boss_lip/scr_seat_clear). Probes: 0/400k screen samples inside head_back (0 per site at all 4 mounts), 0/30k inside head_bezel; glass front 30.994 (flush, target 30.994 +-0.05); screw path 0/400 hits per mount; 6 mm seat solid 400/400 per mount.
• D2 fixed - worm_len 16 -> 14 + face_y offset 8 -> 9.5: thread span now y -32.00..-16.29 = 15.71 mm = 4.00 axial pitches (>=4 teeth-equivalent), bare Ø5 stub forward of y=-16 (max r fwd of -15.5 = 2.500); cradle groove band moved to y -15.5..-13 on the stub, riser/ pad rear split into two side prongs by an r5.9 envelope-relief bore about the worm axis. Probes: worm vs neck static 0.000; ROTATION swept envelope (64-slice surface of revolution, max thread r 5.344) vs neck 0.000; worm_cd()=11.500 vs mesh axis distance 11.500, delta 0.0000; wheel-worm placeholder contact preserved (32.9 mm3).
• D3 fixed - removed the extra R(TAU/4, y) clocking on the worm's double-D bore (dbore_neg(axis="y") already cuts flats +-X = shaft flats); comment corrected. Probe: worm vs motor_tilt 17.4 -> 0.000 mm3.
• D4 fixed - tilt-ULN standoffs lengthened 8 -> 8.5 and buried 0.5 into the column back face. Probe: neck_clevis = 1 watertight body (was 5 disjoint).
• Neck-on-platform done - neck_y -38 -> -17.0 (derived: footprint max r = sqrt(14^2+(|ny|+13)^2)+10 <= 44 gives |ny| <= 17.98; measured 43.10 at z 52..55). Cascade: bolt circle re-clocked (90,210,330)r12 -> (270,30,150)r16 in BOTH neck pilots and platform holes (old layout put a hole 2 mm off the pan axis, inside the D-bore hub); cable channel decoupled to neck_chan_y=-26 (behind the notch; platform slot start + deck pass aim follow it); cheeks re-anchored at (+-22, -33, 155) on a new 52x14x10 cheek-root block (anchoring at ny=-17 grazed the resting stack and buried 11 mm into the -30 sweep); column front notch re-derived: the -30 sweep (shell chin to y=-16.7 @ z~102, display back band to y=-14 right up to z 150) forces the whole column front above z=94 back to y=-19.5. Probes: footprint max r 43.10 <= 44.0; pan sweep -90..+90 step 15 vs all fixed parts 0.000 (only the known 0.031 ball tangency, intended contact, = stage-4's 0.030); tilt sweep -30..+30 step 5: head solids vs neck group all 0.000, screen samples in neck group 0 at all 13 angles; cable route: Ø7 rod through the deck pass (12,-24, z 30.5..55) vs platform/race/ chassis 0.000, Ø7 rod down the channel (0,-26, z 45..150) vs neck/platform 0.000; all 3 bolt holes on solid platform (r 16.5/16.5/33.0, annulus 300/300 solid, bore open, clear of hub/rebate/groove/slot), neck pilot rings 300/300 solid.
• Build health: plain + EXPORT=1 exit 0; all 10 printed STLs watertight; single-body everywhere except the by-design multi-body track_L/track_R (52) and pan_clips (3). Renders (st5_iso/side via port 8770) inspected: assembly coherent, no floaters.

Stage 6, independent re-verification of stage 5 [DONE: READY TO COMMIT]

Same methodology as stage 4 (own probes, scratchpad/stage4_*.py + stage6_d123.py; screen mesh checked by surface-sample containment, booleans raise on failure, nothing reused from the stage-5 agent's scripts).

Docs: CLAUDE.md drift from stage 4 is resolved (10 parts, 12T, Pins_Out_v8; the one remaining "24T" is a historical "was 24T" note, not a stale claim).

Verdict: READY TO COMMIT. No defects found. Standing model-only artifacts (not defects): static motor_pan shaft vs the swept platform D-bore (shaft co-rotates in reality), placeholder worm/wheel tooth interpenetration (regen in BOSL2 before printing), 0.03 mm3 ball-seat tangency.

Wallcheck gate findings (2026-07-13, skills fix campaign)

The new make wallcheck min-wall gate (src/wallcheck.py, in make all) surfaced and closed these on its first full runs: