Assembly

Current assembly notes for the tracked desk-pi prototype. The task #28 insertion/torque-path audit (2026-07-07) verified a complete install order exists (see "Assembly order (verified)" below). The 2026-07-08 maintenance pass then killed the worst service traps: tilt motor is a rear-access cartridge (tilt_carrier), the worm wheel uses an accessible radial grub on a center-only axle flat, tracks close with a master link (no loop-flexing), the pan race got a BB cage, both joints got stall-homing hard stops, the microSD swaps through a plugged left-wall slot, the positively located side-panel joints, and power is a 12 V PD-trigger + dual-buck belly tray (firmware/WIRING.md). Grub-on-flat backlash remains a bench measurement covered by the coupon below.

Bill of materials

Owned quantities cross-checked against a personal parts inventory (2026-07-07; re-audited 2026-07-13 with bag-label photo reads: TT motors, M8, 608zz, F688 candidates, fuses all settled below). "Need" is per robot.

Electronics

Bearings, race, axles

Fasteners and pins

Printed parts (watertight; tank base + split head)

Materials: print head_back (all four pieces), screen_tray and head_door in PETG, not PLA. The Pi 5 lives in the closed head and sustains 70-80 C bursts under brain load (measured, see CLAUDE.md thermal notes); PLA creeps from ~60 C, so a PLA tray would slowly sag under the screen+Pi module and a PLA back wall would relax its snap tongues and tray pilots. The head_door shares the same hot bay, so it goes PETG too (its snap tongues also live longer in PETG). Everything else on the robot -- chassis, tracks, bezel, neck, pan parts, cosmetics -- can stay PLA. The black CR-PETG spool (Tray 1) covers the PETG set.

• chassis_lower_front / chassis_lower_rear: the open-top tub, split at y=+26 for print speed; join with 2x M3x20 (axis Y, heads in the front pads, captive nuts rear)
• 2x Ø4 dowels in the floor pads at x +-61; the deck and bolted side panels bridge it
• chassis_deck_front / chassis_deck_center / chassis_deck_rear: the pan deck in three plates (seams y 66/-52, half-laps + 2x M3 down through each strip into shelf pilots); the center carries the whole pan seat + its own 4 hold-downs; the corner hold-downs live in the strips
• track_L / track_R: tank track pods: 64 links/side (4 print-in-place strips + master; integral Ø2.0 pins in-strip, Ø1.75 filament pins at the 5 boundaries), 12T pin-pocket sprocket (real circular pin seats since 2026-07-10, r 1.15 on the 19.32 pin circle), 2 end idlers on 2x F688ZZ each (30 wide; front tensions in its panel tower), TWO ground-run drive sprockets (spr_y rear + spr_y2 front; the front motor is OPTIONAL, all fittings modeled), 5 dished road wheels/side (30 wide) on M4x40 bolt-axles off the pod-rail wheel beams
• pan_platform: disc that yaws on the base (central shaft bore + off-axis cable pass)
• pan_race / pan_balls / pan_retainer / pan_cage: captured-BB lazy-Susan race, full-circle uplift retainer, and the BB spacer cage (18x 6 mm BBs; pan_balls is a placeholder for the bought BBs; the cage keeps them spaced so a turret lift doesn't scatter them)
• neck_clevis: rounded column + two cheeks that rise into the head and drive the tilt axle; vertical cable channel
• head_bezel_L / head_bezel_R: front frame split at x=+22 (M3 + Ø4 dowel in pads behind the forehead/chin strips); seam staggered off the back's for interlock
• head_back_frame_L/R + head_back_panel_L/R: rear cover in four (frames = walls + pivot hubs, print front-down, no ceiling support; panels = the flat 4mm back wall, print lying flat). 6x M3 from the back join panel to frame rim tabs; the frame halves share the top-wall flange (2x M3) and the panel halves a tongue/groove
• screen_tray: bench-mounted module carrier (2 rails + spine): the screen+Pi bolts to it on the bench (4x M3x10 into the factory bosses, open access), the loaded tray drops into head_back, 4x M3x10 drive from OUTSIDE the back wall (visible, between door and frame)
• cam_cover: camera board cover and cable trap
• sd_plug: friction plug for the microSD service slot (left wall + trim_rail_L); pull it, reach the card with straight forceps down the eject axis (~61 mm, sight line clear)
• tilt_carrier: removable tilt-motor cartridge plate; the motor's ears drop onto its pin-topped D-posts on the bench (2026-07-16 sandwich retention, see step C), 4x M3x16 clamp it to the neck bracket from the open rear bay
• track_keeper_L/R: master-link keeper bars (2 bars + side tabs per pod, 1x M2 each); the master link body prints as link 0 inside each track_L/R
• worm_wheel / tilt_worm: real generated teeth (docs/WORM.md); the wheel has a round Ø5.2 bore and an accessible radial O2.5 pilot for its M3x4 tip-on-flat drive grub
• Cosmetic / design-ref set (render-only today, print with the head): side rails, forehead led_strip, camera_pod eye shell, rear trim_hatch_frame, chassis trim_fascia + trim_rear, and the placeholder gripper arms (mechanism TBD)

Plastic hardware stand-ins (interim, plate "Hardware stand-ins")

Until the metal order lands, every buy-list hardware row above has a PRINTED stand-in (src/standins/ -> stl/hardware/), so the whole robot dry-assembles in plastic. The BOM is unchanged; swap each stand-in 1:1 for metal on arrival.

REWORKED 2026-07-16 ("as functional and as close to reality as possible"). The M4 and M8 pairs now carry real ISO threads (src/threads.py) and screw together for real. More importantly, the rework caught three v1 parts that were physically unbuildable -- each had inherited nominal metal dimensions with no print compensation, which looks perfect in CAD:

RULE: a stand-in mates PRINTED-TO-PRINTED or PRINTED-TO-STEEL, never nominal-to-nominal. Budget 0.1-0.2 mm per printed surface, put the compliance on the printed part (crush ribs beat a tight window), and probe the real mating geometry -- never copy the metal part's numbers or the CAD placeholder's.

Limits to respect on plastic (full rationale in src/standins/__init__.py):

• The track tensioner now has a real clamp stack: outer jam nut, tower, washer, inner nut. The idler stays free on the smooth journal while the tower is clamped.
• M4 is a real fastener now: it screws, tightens and re-uses. Snug it; don't lean on it (~400 N strip vs a metal M4's ~2 kN).
• PLA creeps, so preload decays over hours -- but with real threads that decay is bounded and recoverable: re-snug, nothing degrades permanently.
• Tilt homing is usable (~1.3-1.5 deg wind-up, and stall homing calibrates it out). Grub-on-flat backlash remains bench-only. Park at the balance point -- the real cost is 1-3 deg of creep droop over hours.
• The dry pan slews, but slowly and near the limit, and may need a nudge from rest. Grease both grooves. Don't time the dry pan and call it a gear-ratio verdict.
• Idlers are greased plain bearings: ~50x a ball bearing's rolling resistance, but only 0.17-0.70 N to roll, which the track's own weight exceeds.

Verify on arrival (caliper before printing)

The CAD models several bought parts from datasheet/typical dims, not measurements. The rule for every row: caliper the real part -> update the named param(s) -> make build -> make check + make fits -> only then print the dependent parts. Clone boards vary: barrel/jack spacings tend to be stable across clones, mounting-hole patterns are not. Printing a seat before its part has arrived is how reprints happen.

Assembly order (verified)

Task #28 insertion-path + torque-path audit (2026-07-07): a full install order EXISTS end-to-end. Every step below was checked with swept-cylinder driver-line / extraction probes on the neutral-pose mesh; the numbers in parentheses are the worst measured clearance. Read the order constraints and nasty steps at the bottom before you start: several joints are only reachable at ONE point in the sequence.

Tools you need: a 1.5 mm hex key (head-clamp grubs), a standard M3/M2 driver, tweezers, and thread-lock or CA glue (cosmetics). (The old ~95 mm slim driver is retired: the screen tray killed the 88.5 mm blind channels, 2026-07-08.)

Sub-assemblies to build on the bench FIRST (they become unreachable once seated)

• A. Neck + pan platform. With pan_platform upside-down on the bench, drop 3× M3 up through its underside Ø6.5 counterbores (r16.5 circle, clocked 270/30/150) into the neck_clevis base pilots. The counterbores face DOWN into the race once seated, so this bolt-up is impossible after the platform is on the balls. Torque path: screws clamp neck to platform; the platform's integral 16T pinion takes the drive from the pan motor's 32T gear (fast-pan 2:1 gear-up 2026-07-12; the old on-axis D-bore is gone).
• B. Tilt axle prep. File the center-only flat on the Ø5 rod (span and depth in the BOM row above) and verify it with the axle flat + wheel grub coupon below. The wheel and its spacer tubes are NOT mounted on the bench: a mid-span wheel cannot pass a 695 bore, so they thread onto the axle inside the clevis gap at step 10. The tip-on-flat contact is the positive torque path; the threads hold preload.
• C. Tilt-motor cartridge. Drop the tilt 28BYJ onto tilt_carrier: the can passes the plate's Ø29.1 bore and each ear hole lands on a Ø3.8 pin atop the plate's D-posts (2026-07-16: the real motor is 9 mm shorter than the old placeholder modeled, leaving only 0.2 mm in front of the ears -- the old 2x M4 ear bolts are geometrically impossible and are GONE from the BOM). Press the worm onto the D-shaft. Clock the O2.5 radial pilot over either D-flat, fit the M3x3 cup-point grub, and seat it gently with its tip on the flat. This is low-load axial retention, not a structural clamp. The grub stays in the worm and travels with the cartridge during rearward extraction. Hold the motor against the posts while inserting (step 11); driving the carrier's 4x M3x16 captures the ear bar between the post fronts and the neck's pocket-front wall (0.25 mm float, pins lock rotation). It comes OUT the same way for a motor swap, no head teardown.

Chassis + drive (fixed frame)

ACTION BEFORE ORDERING: Measure all 3 Bag-5 TT 1:120 shaft tips now. Record whether each has the Ø2 axial hole. Order M2x25 screws and Ø9 washers only for motors that actually have it. Many TT variants omit this hole.

1. Print + prep the chassis. Confirm the deck pan-seat, the pedestal, and both TT motor pockets are clean.
2. Fit the four structural side panels. The old separate pod_rail_L/R parts and their dowel joint are retired. Each panel carries its wheel beam and end tower as one print. Seat the front/rear splice tongue and locating pads, install its two M3 screws into captive nuts, then bolt the panel feet to the lower tub. The tongue and separated feet hold alignment before tightening. Before fitting the track, slide two M3 nuts into each shoe trap and press the matching track_shoe_{L,R}_{rear,front} upward onto its two locator pins at y=-68 or +90. Drive two M3×10 countersunk screws from below. The shoe is 10.6 mm wide in X, 14.76 mm along Y and 4.6 mm thick, with its running face at z=10.4. Fit all four shoes for a twin drive; the +90 front pair may be omitted with the optional motors.

3. TT motors. Set each TT gearmotor: shaft +X into the sprocket hub's double-D socket, front tab into the rear-wall pocket, nub into the wall pocket, 2× M3 through the gearbox + wall. Push each upper nut past the slot-mouth crush nibs until it clicks into the captive seat; it stays retained while the motor is positioned. The M2x25 plus Ø9 washer is the preferred axial retainer only when the measured shaft has the Ø2 tip hole. Otherwise use the positive cross-pin fallback:

4. Slide the sprocket fully onto the shaft and mark the shaft through the vertical Ø2.1 hub bore.

5. Pull the sprocket off and file an approximately 1 mm deep notch across the round shaft arc at the mark. Do not cut through the plastic shaft.
6. Reassemble and drop a straight Ø2 filament pin through the upper hub wall, the filed shaft notch, and the lower hub wall.
7. Trim both ends flush with the Ø12 hub. The pin must never stand proud of the hub OD because the hub slides through the side-panel journal.
8. Secure the upper pin end against vibration walk-out with a small drop of CA in the bore mouth, or briefly heat-mushroom it with the iron tip, still sub-flush. The bore is open-top in service, so an unsecured pin can migrate up and out.

The pin positively locks axial motion and torque. The socket crush ribs are only a handling aid. Service the pin from above with the track open. 4. Track running gear. Press an F688ZZ into EACH face of all four end idlers (Ø15.95 seat + Ø18.5 flange recess both sides). Each end wheel: M8 bolt from outboard through the bearings, add the OUTBOARD jam nut against the tower, then the flatted washer + NYLOC nut inboard, started ON THE BENCH with the deck upside down. Orient each nut HEX FLATS FORE-AFT: as the deck drops onto the tub the nut descends into its prow-cheek NUT CHANNEL (y-walls 13.8 apart), which grips the flats -- all torquing happens from the outboard head, no inboard tool ever needed. Before fitting each front NYLOC, deburr and press one 12x28x1.0 steel strip fully flush into the tower's x62 bearing-face recess. Its square ends must seat against both printed end shoulders. Rear pair: snug. Front pair: leave loose, set tension AFTER threading the tracks, then tighten from the head -- the channel holds the nut, the nut clamps the slot. Fit the sprockets on the TT shafts (they mesh the ground run under the hull; the robot's weight seats them). Push an M4 nut up each wheel-beam slot until it clicks past the two crush nibs. The nuts remain retained while the panel is handled; inserting them on the bench is convenient but not required. Then bolt each road wheel with its M4x40 from outboard -- snug, then back off 1/8 turn so the wheel spins free.

Each side-panel beam now closes around the Ø12 sprocket hub over the absolute |X|=70.2..74.5 band with a Ø12.5 full-circle printed journal. Apply a thin film of plastic-safe grease to the hub before insertion and renew it when the track is open. The remaining |X|=74.5..81.2 crossing is open on top for cross-pin access. The owned MR105 bearing cannot replace this land: its 5x10x4 bore is too small for the Ø12 hub, while the hub itself contains a Ø6 free bore, so there is no compatible race seat. 5. Join the strips + close with the master (print-in-place chain, 2026-07-12). Each pod is 4 PRINT-IN-PLACE strips (16+16+16+15 links, hinges already free off the printer -- flex every joint once to crack any sag bonds) + 1 separate master. On the bench: pin the strips end-to-end with 3 Ø1.75 filament pins (each boundary = one strip's Ø2.2 open A-bores interleaving the previous strip's Ø2.2 far bores), then pin the MASTER's far end to strip 1's first link (4th filament pin). Wrap the open chain around the pod with the front idler retracted. Seat the last filament pin (5th) in strip 4's final-link far bores, swing the master's open jaws down onto it, slide the two track_keeper bars into the jaw slot from the side faces, and lock each with its M2 into the side-face pilot. Tension the idler -- run the chain TENSIONED: the PIP joints carry 0.35 radial slop (~8 mm of loop slack vs the old filament chain) that otherwise lands as top-run sag; if the front tension slot runs out of travel on the physical chain, that is the expected first fix (more slot travel), not a link redesign. Track removal forever after: 2 M2s out, slide the keepers, lift the master off its pin.

Pan joint

6. Pan motor + drivers. Fit the printed 32T pan_gears onto the motor's D-flats, then drop the pan 28BYJ into the pedestal can pocket (fast-pan 2026-07-12: the shaft sits OFF-axis at (-19.2, 0), gear up; ears run along X), clamp the 2 ears with M3 into the pedestal pilots from ABOVE (deck open). Mount ULN #1 and the 2nd ULN/MX1588 board on their standoffs; wiring box leads exit the pedestal +Y relief.
7. Power tray, ballast, then belly plate. Screw the PD trigger to the rear-wall M2 pilots (jack aligned with the USB slot), the 5A buck to the belly plate's 40×20 post grid, zip the mini buck beside it, and wire per firmware/WIRING.md (leave 60 mm slack on every tray run; zip the incoming wall cable to the floor anchors as strain relief). Fit chassis_pedestal to the loose belly plate with its two printed pins near mid-slot and start the four M3x12 csk screws into the captive pedestal nuts. Leave them loose enough for X adjustment. The swept countersinks keep every screw flush anywhere in the slot instead of pulling the pedestal back to nominal. Backlash setting: mesh the 32T motor gear against the platform's 16T pinion with a 0.10 to 0.15 mm paper feeler between the flanks. Snug all four pedestal screws, remove the feeler, then rotate the platform one full turn by hand and check for tight spots. Re-set if it binds. The target is perceptible but small lash at the teeth. Load the low ballast into the rear bay + the belly-plate pockets from BELOW, then bolt on belly_plate (6× M3 csk, flush at z=7). Ballast must go in before the plate closes the floor.
8. Pan race. Grease the pan_race lower groove, seat it on the deck floor, lay the pan_cage ring over it, and drop the 18× 6 mm BBs through the cage pockets into the groove with tweezers. The cage keeps them spaced; any later turret lift leaves all 18 sitting evenly in the groove instead of bunching and rolling out. Load all six retainer nuts through the inward-opening seat-wall slots before seating the race, since the platform and retainer close that access.
9. Lower sub-assembly A (neck+platform) onto the BBs so the platform's upper groove captures them and its integral 16T pinion drops into mesh with the motor's 32T gear (fast-pan 2026-07-12: fit the 32T on the motor D-flats BEFORE this step). Screw drop pan_retainer straight down over the seated platform. Its 90.8 mm lip ID clears the platform's 90.0 mm top-band OD throughout the insertion path. Seat all six lobes in their deck pockets and drive 6x M3x10 into the captive nuts. The continuous lip reaches over the platform rim rebate to hold the top-heavy head down. Check the platform spins free.

Printed race acceptance (do this after the head is fitted, step 12+, so the race sees the real top-heavy load). Bench physics, not CAD: ~8 N/ball Hertz on PLA cold-flows and abrades; the 0.2 mm groove clearance sits inside FDM scatter, so a given print may rattle or pinch and will worsen with wear. The post-gear-up torque budget is ~15 mNm at the platform vs ~7 mNm assumed race friction (little headroom).

• Measure breakaway torque: push sideways at a known lever (e.g. a kitchen gram scale against the head side at a measured radius from the pan axis). Torque = force × radius. Example: 10 gf at a 100 mm lever ≈ 10 mNm.
• ACCEPT at <= 10 mNm breakaway.
• REJECT above 10 mNm, or if the platform rattles axially more than perceptibly, or if breakaway grows noticeably after a few hundred sweeps. Then fit the commercial ring (BOM "Bearings, race, axles"; print-3 option below).
• Derate (between accept and upgrade): at 7-10 mNm of friction the pan still slews but acceleration roughly halves (net accel torque 15 - 10 = 5 mNm vs the budgeted ~8); a 180 deg sweep stretches from about 2 s toward 3 s. Below ~5 mNm friction, nothing changes vs budget. Above ~12 mNm, stall-homing detection gets unreliable and the commercial-ring upgrade is mandatory (do not ship on the printed race).
• Upgrade path (print-3, NOT modeled now): fitting the commercial ring means re-cutting the deck seat and the platform underside for the ring's races (a variant of chassis_deck_center and pan_platform). Frozen shells stay clean (separate-by-stability rule); model against the bearing actually bought, dims VERIFY_ON_ARRIVAL. Until then the printed race + BBs is the default path.

Tilt joint + head (all on the pan group)

10. Bearings + axle cartridge. Press a 695-2RS into each neck cheek from the clevis gap (seats open flush to the inner face; a light lead-in helps start the press). Insert the prepped axle (B) through one cheek bearing, thread the worm_wheel and its two spacer tubes onto the leading end inside the gap (wheel meshes the worm), then drive the axle on into the far bearing. The round wheel bore and tubes pass freely over the round journals. With the head not yet hung, use the open rear bay to center the wheel, clock the filed flat to the -Y hub pilot, and seat the M3x4 grub flush. Loosen this grub before pulling the axle.
11. Insert the tilt cartridge. Slide sub-assembly C (carrier + motor + worm) in from the open rear bay: the worm passes the bracket plate's Ø12.2 bore, its tail lands in the open-top cradle groove, the can registers in the Ø29 pocket, the motor's ear bar rides the neck's ear-bar channel to 0.2 behind the pocket-front wall, and the carrier's 4 bosses land on the plate/column rear faces. Drive 4× M3×16 from the rear into the captive nuts in the neck blocks -- this also captures the ear bar between the carrier's D-posts and the neck wall (hold the motor against its pins until the first screws bite). (Extraction reverses this and is UNCONDITIONAL since the 2026-07-12 3-start worm: the mesh back-drives, so with a dead motor just hand-nod the head while pulling and the worm screws itself out. The old rule -- drive the head fully UP first because the single-start pull needed ~46° of nod against ~34° of stop travel -- is retired, 2026-07-13.) Route the tilt ULN wiring on the column back standoffs, board centered at z 93, below the carrier (motor + driver both ride the pan group, so no leads cross a joint).
12. Hang the head on the axle. Before lowering head_back, push each M3 clamp nut through its downward-facing slot mouth until it clicks past the crush nibs and seats on the cross-bolt axis. The front-down bench pose is still convenient, but retention no longer depends on it. Lower the head so its side hubs take the axle ends, then drive the two M3 pinch-clamp bolts from the rear bay. The axle now turns with the head; the worm holds tilt with the driver off ONLY marginally (fast-tilt 2026-07-12: the 3-start worm back-drives; hold is 28BYJ detent+gear friction through 4:1 -- firmware energize-hold or park at the balance point; see CLAUDE.md fast pan/tilt pass).
13. Screen + Pi module (tray). On the BENCH: bolt screen_tray to the combined touchscreen+Pi with 4× M3×10 pan heads straight into the display's factory 126.2×65.65 bosses (the pillars are z-offset from the bores, so the driver line is open). Power-test the module on the desk if you like -- it is now a self-contained unit. Then drop the loaded tray into head_back from the open front (the pillars pass z-clear of the clamp tubes) and drive 4× M3×10 from OUTSIDE the back wall into the pillar-end pilots: short, visible screws in the fixed strip between the door outline and the hatch-frame opening. The bezel pocket locates the glass when the head closes (step 15).
14. Camera + forehead LED. Mount the CM3 front-face-in on the 4× M2 pier bosses, trap it with cam_cover (2× M2 + ribbon pinch), drop the CSI ribbon into the pocket. Seat the WS2812 forehead segment in its recess and route the 3 wires through the wire pass to the Pi. Route the Pi power pair per the cable step below.

15. Close the head. Bezel to back: 8× M3×35 through the perimeter posts into the captive nuts in the back bosses (screws from the front). Fit the rear service head_door (the stepped rear pod): engage the two top hook tabs first, swing the door in until both leg snap tongues CLICK behind the wall band beside the void (tool-free; replaced the 2× M3 csk 2026-07-10). To open: firm pull on the pod's bottom edge (35 mm proud, that IS the grip); the barbs' back ramps cam the tongues inboard and release.

    SERVICE (door / dead head): the tilt drivetrain occupies the pod cavity at the ±33.8° stalls by design (hollow clearance, not a closed-door clash). tools/probe_door_stall.py (2026-07-16) swept the tool-free open path (hook pivot 0..24°, then 15 mm lift along swung -Y) at tilt -33.8 / 0 / +33.8 against tilt_carrier, motor_tilt, tilt_worm, worm_wheel, and neck_clevis: all three poses path-clear on the baseline cavity (no enlargement). Stall-open is therefore geometrically possible, but the margin is millimeter-scale (closed-pose drivetrain-to-door clearance is ~2-3 mm at the stalls, and the swept path was checked at 2 deg / 1 mm resolution); neutral tilt remains the comfortable service pose (more free volume, easier hands, print tolerance and an un-held nodding head both eat into that margin). For a dead-motor or dead-Pi head: the 3-start worm back-drives, so hand-nod the head to roughly neutral first, then either hold it there with a second hand or wedge a soft prop (folded cloth / foam block) between the head chin and the chassis deck before pulling the door. Why prop: without energize-hold the head can slowly nod under its own imbalance while you work, and a hard-down park is awkward even though the open path does not collide.

15b. Antenna drives. Push one metric O-ring 6.0 x 1.0 into each modeled top-wall gland, then drop each mast in from above (rack facing the pinion slot), then hang ant_bracket on the back wall (spine + locating shoulders and 4x M3). Insert each printed ant_output_* half-shaft/pinion through its Ø4.2 bushings, fit ant_idler_axle_* and the compound ant_idler_gear_*, press ant_motor_gear_* onto the motor double-D, and bolt each 28BYJ nose-through its face plate (2x M3 into the vertical-ear pilots, shaft inboard). Each mast has its OWN motor + ULN2003 (independent control); wire both to the Pi in the head. Homing: leave park at the lowest step rate for maximum torque, then ramp. BUY: 2x metric O-ring 6.0 x 1.0, NBR or silicone; silicone is preferred for low breakaway. All 30T/12T/27T m0.8 gears, axles, shafts, and rack teeth are exported printable parts.

**BENCH BREAKAWAY PROTOCOL:** assemble one mast and ring with NO gears fitted. Measure
axial breakaway with a 0-1 N spring gauge, or hang calibrated 10/20/30 g weights from
the mast tip. Sliding breakaway target: <=0.15 N (~15 g). Park-groove breakaway target:
0.2-0.3 N (20-30 g). If outside the window, silicone grease lowers sliding friction;
a 6.0 x 0.8 ring gives a lighter section/squeeze variant; or reprint the cheap vertical
mast with `ant_park_groove_depth` adjusted by +/-0.1 mm. Groove depth is the park-force
tuning knob. Final friction and breakaway are bench validation. CAD fixes only the
geometry and the acceptable window.

Cables (per docs/CABLE-CHECK.md + firmware/WIRING.md)

16. Wall PD brick → rear PD trigger (12 V) → belly tray bucks. The Pi-rail pair (18 AWG, from the 5A buck) coils a 2-turn service loop (~600 mm) at r≈48, z≈38–45 in the cavity → 16×8 deck pass → platform slot → neck channel → the column top-left exit window → through the bottom-rear head slot to the Pi's GPIO 5V/GND pins with ~60 mm of free head lead for the tilt drape. Set usb_max_current_enable=1 + EEPROM PSU_MAX_CURRENT=5000 (GPIO power skips PD negotiation). Software-limit pan to ±90° (hard stops at ±93.3) so the loop never over-winds.

Cosmetics LAST (glue + locating pins)

17. Press-and-glue the pin-located cosmetic parts: trim_rail_L/R, trim_hatch_frame (after the screen, its band overlaps the 4 driver-channel mouths, up to 2.3 mm), camera_pod, and the chassis trim_fascia / trim_rear / sensor_rear grille cap. Fit fascia electronics if used: HC-SR04P barrels through the Ø16 passes, amber corner lamps, front LED strip, rear buzzer/speaker.
18. Final check + homing. Power on; firmware stall-homes pan against its ±93.3° deck stops and tilt against its ±33.8° fin stops, backs off, and zeroes. Sweep pan ±90 and tilt ±30 and confirm the screen/Pi stack, worm, cheeks, axle, cables and bottom head edge stay clear (matches the make check-sweep gate).

Order constraints (do NOT reorder)

• Body↔pod join + TT screws (step 3) before ULN #1 (step 6): the board blocks cavity access.
• Neck↔platform bolt-up (sub-assembly A) before seating on the balls, counterbores face down.
• Six retainer nuts before the race, cage, and BBs (step 8); race, cage, and BBs before the platform; platform before the drop-on retainer ring (step 9). Power tray + ballast (step 7) stay before the belly plate closes the floor.

Nasty-but-possible steps (measured)

• Head-clamp grubs (step 12): 1.5 mm hex key driven blind UP through the motor bay, 4.0 mm clearance. Kept deliberately: the grubs give continuous tilt-zero trim.

(2026-07-08 passes retired the old tilt-motor-ear reach, the inaccessible wheel grub, the last-track-pin loop flex, and the 4× 88.5 mm blind screen-standoff screws -- the worst step in the build is now four short bench screws plus four visible wall screws.)

Squaring up: datum + shim procedure (first build)

The tilt axis and the screen optical plane sit at the end of a long printed stack (chassis lower tub → z46 deck seam → pan race → pan platform → neck clevis → cheek 695 bearings → axle → head). CAD gates (make check / fits / joints) check part PAIRS at nominal, not accumulated tolerance. A first physical build needs this procedure to square itself up without reprinting anything.

Params cited below: clevis_half=22 (cheeks at x ±22), tilt_axis_y/tilt_axis_z = -18 / 153, scr_mount_pts factory pattern 126.2 × 65.65, base_h=66 (deck pan-mount plane), neck→platform 3× M3 on r16.5 clocked 270/30/150 about (0, neck_y=-17), head-clamp grubs at x ±30, pan race 18 BBs on Ø80 (pan_race_n / pan_race_circle_d).

Master datum

Master datum plane = the pan platform's top face as seated on the BB race. The deck's pan seat defines it (base_h=66). Nothing below the race is adjustable: the stack under it is edge-bearing (side panels prop the deck), so its errors tilt the whole turret together and the pan sweep averages them out. Errors that matter to the eye are platform-to-screen. All squaring happens between the platform and the head.

What to measure

Bench method: calipers + a machinist square. No dial indicator assumed. Do (a) after step 12 (head hung), (b)/(c) after step 13 (screen tray in), (d) anytime the race is seated.

a. Tilt-axis parallelism to the platform. Hold the platform still. Measure from the platform top face up to each exposed axle end (or the head clamp bosses at x ±30). Difference over the ~60 mm span (clamp-to-clamp, or cheek-to-cheek at clevis_half ±22) is the axis tilt. Target ≤ 0.3 mm (~0.3°).

b. Screen upright / lean. Stall-home tilt, command zero. Stand a square on the deck (or platform) and measure the gap to the glass at the top and bottom bezel edges. Difference over the ~66 mm vertical mount span (scr_mount_pts 65.65) is lean.

c. Screen twist (rotation about Y). Same square, gap to glass at the left and right bezel edges at one height. Difference over the 126.2 mm horizontal mount span is twist.

d. Pan wobble sanity. Pan slowly ±90 and watch the head top edge against a fixed reference. Cyclic rise/fall means race/platform debris or an unseated BB, not a shim case. Re-seat the 18 balls on the Ø80 groove (step 8) and recheck.

Where shims go

Clevis-foot clocking (which foot raises which side of the axis):

• 270° foot (rear, -Y from pan axis through neck_y): raises the rear of the clevis → lifts the rear side of the axle plane (axis pitch relative to platform).
• 30° foot (+X, slightly +Y of the column): raises the right cheek → lifts the +X axle end.
• 150° foot (-X, slightly +Y): raises the left cheek → lifts the -X axle end.

Shim the low side. Equal shims under 30°+150° raise the front of the clevis without rolling the axis; rear alone does the opposite pitch.

Shim stock + order

• Stock: printed 0.1 / 0.2 / 0.4 mm shim washers (Ø7 × Ø3.2, print a strip of each), or aluminum foil layers (~0.02 mm each) for fine trim. M3 steel washers are 0.5 mm and too coarse for anything but gross correction.
• Order is load-bearing: 1. Square the tilt axis at the clevis feet before hanging the head (or re-hang after: step 12 grubs must come off to drop the head). 2. Hang the head; set tilt zero at the x ±30 grubs (step 12). 3. Screen-tray shims last (step 13). 4. Re-run the stall-homing sweep after any shim change -- stop angles move with the stack (step 18).

The CAD gates verify pair fits at nominal; this procedure is where the accumulated real-print tolerance gets taken out, and it is expected to be needed on a first build.

Track coupon protocol (Track coupon plate -- print BEFORE any strip plate)

The Track coupon plate carries two 5-link print-in-place strips + 1 loose master link + both keeper bars (~1 h sliced in PETG):

Each strip is open-A first link, 3 integral-pin mids, open-far last, keels on. Production loop geometry, tools/probe_track_pip.py, and the long strip plates stay on Ø2.7 only; the Ø2.56 strip exists solely on this coupon plate (stl/base/track_coupon_tight.stl, written under EXPORT=1).

Static checks (both strips)

1. Every PIP hinge frees after break-in flexing -- no fused knuckles.
2. Hinge radial slop: pull the 5-link strip taut and measure total stretch vs 40.0 nominal; that gives real per-joint slack. Scale ×59 and check it still fits the 6.5 mm front idler outboard tension travel via delta_L / 1.84 (dL/d(idler_y) on the raised loop). Accept if scaled stretch ≤ 6.5 × 1.84 ≈ 12.0 mm of full-loop elongation budget; reject if a single 5-link strip already exceeds ~1.0 mm total stretch after static break-in (that scales past the tension budget).
3. Ø2.0 integral pins unbroken after 20 full ±35 deg articulations.
4. Keel faces clean (no sag scars), grousers flat.
5. Ø2.2 boundary bores accept Ø1.75 filament.
6. Master jaw drops onto the end pin; keepers slide and seat; M2 insert pilots hold.

Any static fail: adjust track_bore_pip_d / track_pin_print_d and reprint the COUPON, not a strip. Prefer the Ø2.56 strip only if it frees cleanly and beats the Ø2.7 strip on the powered stage below; otherwise keep production at 2.7.

Powered coupon (deciding bench test for mid-drive reliability)

After both strips pass the static checks, run a powered coupon on the production (Ø2.7) strip first; optionally repeat on the Ø2.56 strip for A/B wear data.

Setup (concrete):

1. Mount a real mid-drive sprocket (printed drivewheels_* sprocket) on a TT gearmotor shaft, fixed to a bench block so the sprocket axis is horizontal.
2. Pin the 5-link coupon into a short loop segment against the sprocket ground-run mesh (use Ø1.75 filament at the open ends, or clamp the strip as a weighted flat run over the sprocket with ≥200 g hanging on the free end so pins stay seated in the conjugate pockets). A hold-down shoe or a wooden block with a 0.9 mm gap over the link crowns is enough to keep the mesh from lifting.
3. Drive the TT through the chassis MX1588 (or a bench H-bridge) at the firmware 80% PWM cap (never 100%: 27 W budget rule). Forward 60 s, reverse 60 s, then continuous forward for the break-in block.

Break-in: 30 minutes continuous at 80% PWM, or 5000 sprocket revolutions, whichever comes first. (At ~120:1 TT gearing and ~200 RPM motor free speed under light load, 30 min is on the order of several thousand pin engagements; count revs with a mark on the sprocket if you can.)

Re-measure after break-in (same gauges as static):

This powered coupon is the deciding bench test for mid-drive reliability under representative load and reverse mesh. Do not commit the 6.8 h strip plates until the Ø2.7 strip passes. If Ø2.7 fails ovalization/elongation but Ø2.56 passes, consider adopting 2.56 into production (update track_bore_pip_d, re-run probe + coupon) only after a second powered coupon confirms it.

Mid-drive engagement is quantified, not assumed: the conjugate 14T mesh has contact ratio CR 1.48 and the rigid-chain pitch-ratchet barrier from tools/probe_track_pip.py is 2.14 mm. Fixed road-wheel bolt-axles cap lift to about 0.1 mm at their stations; each hold-down shoe caps the sprocket mesh window to 0.9 mm.

Links / strips / masters are CONSUMABLES

PIP hinges are fatigue/wear items: integral Ø2.0 pins hammer Ø2.7 bores oval, designed slack grows past the ~8 mm tension budget, the 2.14 mm skip barrier erodes, and a broken integral pin scraps a whole 16-link strip. Expected wear life is unknown until the powered coupon above has real numbers (log hours and km of desk travel when you have them).

Replacement procedure:

1. Open the master link: remove the 2 keeper M2 screws, slide the keeper bars out, swing the C-jaw off the closure pin.
2. Count links from the master along the loop to the damaged strip segment (production layout per side: strips of 16+16+16+15 + 1 master).
3. Drop out the worn strip at its two filament boundary pins (or cut a fused pin if a mid-strip pin failed), seat the spare strip, re-pin the boundaries with Ø1.75 filament (press tip into the Ø1.6 step).
4. Re-close the master, refit keepers + M2s, re-tension the front M8 until the mesh bites without top-run flogging.

Print one spare production strip per side with the first batch (same PETG spool as the loop; the 16-link size covers the three long segments; also keep one 15-link if you can spare the plate). Masters and keepers are also consumables on the same service path (2 M2s + jaw wear).

Bearing-seat coupon protocol (print BEFORE neck_clevis and any wheel plate)

Print hw_coupon_695 and hw_coupon_f688 in the supplied orientations. Their five steps are marked by 1..5 tick notches and ascend through 0.075 / 0.100 / 0.125 / 0.150 / 0.175 mm rib crowns. Press the REAL bearing into the steps in ascending order. The correct step is the LOOSEST one where all of these are true:

1. Seating needs a firm thumb or arbor push, not a tap-in drop.
2. The seated outer race cannot be turned by finger torque.
3. The bearing cannot be pushed back out by thumb from behind.
4. There is no whitening or cracking at the rib roots.
5. For the 695 only, the bearing still spins freely after seating, proving its shields were not crushed.

One step looser will creep. If two adjacent steps pass, choose the looser one. Re-key brg695_rib_proud or idler_rib_proud to the winning crown, run make build and make check, and only then print neck_clevis or the wheel plates. Repeat this protocol for every filament brand or material change. Results do not transfer between spools.

Axle flat + wheel grub coupon

Print the worm_wheel, or a short hub coupon with the same Ø11 outside diameter, round Ø5.2 bore and radial Ø2.5 pilot. Also print a simple fork gauge with two open slots: 4.1 mm GO and 3.9 mm NO-GO. This gauge is documented rather than exported as a stand-in because adding a new export, plate entry and invariant is more than the roughly 40-line trivial-part threshold.

1. File a 9.5 mm long flat to 1.0 ±0.1 mm depth. The 4.1 GO slot passes over the across-flats dimension; the 3.9 NO-GO slot does not.
2. Slide the round hub over an untouched round journal and center its pilot over the flat.
3. Seat an M3x4 cup-point grub flush with the hub OD. Confirm its cup lands on the flat.
4. Hold the axle and reverse wheel torque by hand. Confirm no hub rock and record backlash feel.
5. Loosen the grub and confirm the hub again slides over both round journal sections.

Grub-on-flat backlash versus the retired chordal D-key is bench-only. If the coupon rocks or slips, adjust wheel_grub_pilot_d or axle_flat_depth, then reprint the coupon.

Tilt torque + holding bench coupon

Run this protocol on the complete physical head before choosing the production worm pair:

1. Measure residual imbalance about the axle with a spring scale at a measured lever arm at tilt 0, +15, -15, +30 and -30 deg. Record the worst value in mNm.
2. De-energize at +30, -30 and 0 deg. Watch each position for 10 minutes and record any nod or drift.
3. Slew under the real head load at full speed in both directions through the full -30 to +30 deg working range. Record stalls, missed steps or hesitation.
4. Energize-hold off balance for 10 minutes. Record coil current and motor temperature.

Decision rule: if worst imbalance exceeds about 50 mNm, any de-energized nod is not acceptable to firmware parking, or the loaded slew stalls, swap to the committed single-start fallback pair. It is a drop-in, see docs/WORM.md. A spring-ball neutral detent on the wheel hub remains the print-3 option and is deliberately not modeled, per the 2026-07-13 decision. Prove the need on the physical head first.

Recommendations (bigger than this pass)

• Print the axle-flat + wheel-grub coupon above and bench-check tip seating, hub rock and reversal backlash before committing the filed axle.
• Print supports: paint a support enforcer under the neck's two tilt-stop posts (x 20..32, z 150..156 -- their outboard halves start mid-air with the neck on its back) and extend the clamp-tube supports over the head_back stop fins (55° overhangs feeding a ~0.9 mm homing margin; treat the first stall-home angle as calibration). Print the track keepers lying on their 13.3×1.9 bar face so the slot-critical 1.9 width is XY-accurate.
• Tilt stop faces are sacrificial homing surfaces (fin/post PLA-on-PLA at +-33.8°; fins crush-hardened 2026-07-16 to x-span |26.5..32|, 1.375x contact width; inboard 0.5 running-clear of the cheek face). Home at reduced drive once per boot only (firmware/WIRING.md "Tilt homing + hold"). If home zero drifts after months of boots, inspect the fin faces and post tops. Escalation if bench wear is real: a replaceable stop-cap (printed slip-on pad) on each cheek post -- deliberately not modeled now; keep the post z-faces and the +-33.8° first-contact angle if you add one.
• Track gauge is still ~184 vs a 205-wide head (~10 mm overhang/side): widen the gauge or accept it (tracked elsewhere).
• If field-servicing the screen without un-gluing the hatch frame matters, notch the frame band clear of the 4 channel mouths.

Wiring

See firmware/WIRING.md (2026-07-08; power review 2026-07-16) for the full architecture: 15 V PD-trigger input from a 45-65 W brick (12 V = 27 W fallback), dual-buck belly tray (5.1 V Pi rail + 5 V motor rail), hard current limiting (CC pot + TT-branch polyfuse), the brownout test protocol, what crosses each joint, Pi 5 config flags, connector/labeling rules, and the buy-list delta. Short version: only the Pi-rail pair crosses tilt; the pan loop carries that pair plus the thin motor-rail/signal bundle; DSI and CSI ribbons never leave the head.

Order now (by lead-time importance)

1. 8x F688ZZ flanged bearings (8x16x5, flange Ø18): most specific part, slowest to source; the end-idler seats are modeled around them (2 per wheel x 4 since the mid-drive). 1b. 10x M4x40 road-wheel bolt-axles to the hard shank spec: under-head 40 mm, plain journal 34.0..35.5 mm measured with calipers before accepting the bag. REJECT full-thread kits and stock DIN 931 / ISO 4014 M4x40 (~20 mm plain). REJECT Ø5 shoulder screws (bores are Ø4.2 / Ø4.4). M4 nuts are owned. Until metal arrives, print hw_m4_bolt (JOURNAL_L=34.7).
   • 4x M8x70 + 4 jam nuts + 4 NYLOC nuts + 4 washers (end bolt-axles; Bag 13 has no M8).
2. 4x HC-SR04 (forward + rear obstacle + 2 cliff; zero owned; plain 5V is fine on the Arduino I/O plane). TT gearmotors are COVERED (own 3); buy 1 more only for the optional twin-drive 4th station.
3. Power electronics (firmware/WIRING.md, re-specced 2026-07-16): a 45-65W USB-C PD brick that advertises 15V (the official 27W is a 12V fallback with firmware co-scheduling load-bearing), 15V PD trigger, XL4015-class 5A buck (CC/CV variant), MP1584 mini buck, 2A-hold polyfuse (MF-R200 class) for the TT branch, JST-XH kit + crimper, 1 m 18 AWG silicone pair plus 24-26 AWG high-flex silicone for the pan-loop runs, inline blade-fuse holder (the 5A blade fuse itself is owned).
4. 1 m narrow addressable LED strip (4–5 mm wide, SK6805-2427 / WS2812-2020, ≥160 LED/m), one purchase covers the forehead 8-LED segment and the front 7-dot strip. (Alternative: widen led_slot to ~54x11 and buy two common 8x5050 sticks.)
5. Ø5 SOLID rod ~100 mm (tilt axle; NOT tube) + Ø4x12 dowels for the remaining registered shell/seam joints. The retired body-to-pod rail joint no longer consumes four.
6. 6 mm airsoft BBs (bag of 100+; need 18): cheap, everywhere.
7. Optional: MAX98357A I2S amp x1 (pairs with the owned 8Ω speaker), amber LEDs x2–5.

Every board/hardware item above lands in the "Verify on arrival" table -- caliper before printing its seat.