Servo shafts, horns and the remaining hardware values (2026-10-31, #62)
Question: where every servo shaft sits (which tab hole it is near, which face of the printed lug the flange rests on, which way it points), the horn geometry, and the other assumed hardware values, from the reference files. Ground-truth order (CLAUDE.md): Bambu STEP > PurpleShark’s STLs > guide PDF images > remote configs > Cansin’s photos > datasheets. Every conflict below names both sources and the winner.
Sources (local copies under docs/references, gitignored):
- STEP-SERVO / STEP-HALF / STEP-LINE / STEP-CROSS:
cyberbrick-cad/servo/CAD_9g Servo Motor with Clutch Protection-PG002-PG001/*.stp, read exactly with OCP (uv run --with cadquery-ocp, the reader of model/engine/step_mesh.py): cylindrical faces (radius, axis, extent) and solid boxes. - STL-*:
01-eyes/models,02-pan-tilt/models,03-mouth/models(assembled frames, model/geometry.py), read with trimesh: ray casts along the hole axes, voxel slices,model/fits.pycircle fits. - EYES-pN / PT-pN / MOUTH-pN: the three guide PDFs, page N (“Page N of …”); pictures extracted with
pdfimages. PT-p7 and MOUTH-p14 are the designer’s CAD renders (servos and horns drawn). - CONFIG:
03-mouth/remote-config-eyes-pandt-jaw.json(every servoinitial_value90). - PHOTO: docs/references/build-photos/eyes-lid-linkage-side.jpg (Cansin’s build).
- C3: Espressif ESP32-C3 datasheet, Table 5-7 “Wi-Fi Current Consumption Depending on RF Modes” (https://documentation.espressif.com/esp32-c3_datasheet_en.pdf).
Method (deterministic, script in the session scratchpad, not committed): for each servo, 2 near holes x 2 shaft directions x 2 lug faces = 8 placements of STEP-SERVO on the measured hole axes; each is voxelised (0.5 mm, eroded one voxel so face contacts do not count) against the printed parts around it; survivors are then decided by the guide pictures and by the linkage closing at the designer’s neutral (CONFIG: 90 for all), and the result is checked against the measured faces the bars must ride on.
Servo STEP (STEP-SERVO), local mm
Section titled “Servo STEP (STEP-SERVO), local mm”- Tab plate z -2.3..0 (2.3 thick); tab holes r 1.1 at x -2.55 and 25.25, y 5.9, counterbore r 2.1 z -1.3..0. Case x -0.1..22.8, y -0.1..11.9; body z 0..16.3 on the +z side of the tabs; gear housing to z -10.0 (boss r 5.3), collar r 2.4..3.1 to z -11.2, output spline r 2.35 from z -11.2 to the tip at z -14.52 (3.32 long), screw hole r 0.95.
- Shaft at x 16.69: 8.56 from the near hole (25.25), 19.24 from the far one.
- Lead: three wire solids x 16.69..41.99 at z 10.55..11.47 leave the +x end face (x 22.8), i.e. the cable leaves the end nearest the shaft, 5 mm above the body bottom.
Horn STEPs
Section titled “Horn STEPs”- All three horns: hub r 3.63, 4.15 tall (local z 0..4.15); spline bore r 2.15 from the z 4.15 face down to z 1.8 (2.35 deep); screw hole r 1.25 (z 1.1..1.9); screw-head recess r 2.45, 1.1 deep in the z 0 face; arm 1.6 thick on z 0..1.6. So z 4.15 is the servo side and the arm lies on the outer face (z 0).
- Arm holes r 0.5 every 2 mm: half arm (STEP-HALF) at 4.75, 6.75, 8.75, 10.75 (r 0.55), 12.75, 14.75, tip r 1.9 at 14.15; line arm (STEP-LINE) the same both sides; cross arm (STEP-CROSS) long side 4.75..16.75, short sides 4.75 and 6.75. “Second from last” (EYES-p15, PT-p11) on the half arm = 12.75 (geometry.HORN_STEP pin, unchanged).
- Seat: the horn goes on until the spline tip (z -14.52) meets the bore bottom: inner face at STEP z -12.17, outer face at -16.32, i.e. 11.52 beyond geometry’s tab plane (z -0.65), outer face 15.67. Conflict: geometry.SERVO_STEP horn_seat 10.35 (Fusion measure of the collar top, z -11.0) puts the hub 1.17 mm down the spline into the bore bottom; STEP wins: horn_seat 11.52. Three independent checks agree with 11.52: the eyeball bar (STL y 2.5) lands on the eye horn’s outer face at y 2.52; the tilt bar between the horn’s outer face (x 10.98) and the tilt arm lever face (x 8.0) is 2.98 for a 3.0 bar; the lid horn’s outer face (x -23.74) is 0.26 from the lid ears’ outer face (x -24.0). Conflict: the model placed the horn with local +z away from the servo (arm against the servo, hub outward); the bore/recess says the opposite. STEP wins: the horn is turned over (same extents along the shaft, arm outermost).
Per servo (frames as model/geometry.py; “lug” = the printed tab plate with the two screw holes)
Section titled “Per servo (frames as model/geometry.py; “lug” = the printed tab plate with the two screw holes)”Lid servos (L; R mirrors about x 29.5)
Section titled “Lid servos (L; R mirrors about x 29.5)”- Lug x -7.42..-3.40 (STL-L servo bracket; the face is -7.34 at the upper hole, 0.08 proud at the lower one).
- Voxel test: only placements with the tab on the lug’s outer face (x < -7.42) clear the bracket; both shaft directions and both near holes survive it.
- Shaft points -X (outboard): EYES-p32 photos and PHOTO show the horn and crank bars outside the bracket.
- Near hole = the upper one (y -4.65): in PHOTO and EYES-p32 the horn hub is about 10 mm below the upper tab screw and 18 below the lower (STEP: 8.56 / 19.24). The four-bar agrees: with the horn along +Y at neutral the pin is at (y -0.46, z 40), 41.9 and 41.6 from the two lid pins for 41 mm bars, symmetric about the eye centre; the lower hole would give 46.2 and 40.0.
- Flange: tab’s body side on the lug’s outer face, body through the bracket window, gear housing outboard.
- Result: tab plane x -8.07; horn seat x -19.59, outer face -23.74; shaft (y -13.21, z 40.0). Was (-16.0, -13.5, 40.0) inferred. Settled.
- The straight eyelid bar rides on the lid ears’ outer face (x -24..-27, [M]) 0.26 off the horn face; LID_LINKS unchanged (the old “4.1 mm off the horn face” note is gone).
Eye servos (L; R mirrors)
Section titled “Eye servos (L; R mirrors)”- Lug y -11.50..-6.50.
- Shaft +Y (the horn drives the eyeball bar above it, STL eyeball bar y 2.5..5.5). Voxel test: with +Y only the tab under the lug (shaft side on the lug’s bottom face y -11.5, gear housing up through the lug window) clears it; EYES-p16 step 6: “first servo … screws from underneath”, photo shaft up.
- Near hole = the outboard one (x -6.0): the eyeball bar’s horn pin is at (x -10, z 27) [M], 12.56 from a shaft at x 2.56 (pin 12.75); the inboard hole puts the shaft at x 13.24, 23.2 away, which no horn hole reaches.
- Result: tab plane y -13.15; seat y -1.63, outer face 2.52 (the bar’s 2.5); shaft (x 2.56, z 26.9; z 27.0 since the mating check, below). Was (2.9, 0.0, 26.9) inferred. Settled.
Tilt servo (PT frame)
Section titled “Tilt servo (PT frame)”- Lug x 20.0..25.0 on the pan base tower (two blocks).
- Shaft -X toward the tilt arm, near hole y 22.54 (toward the lever): PT-p7 CAD top view (horn at the +Y end of the body, body outside the tower, tab ears on the blocks’ outer side).
- Flange: tab’s shaft side on the lug’s outer face x 25.0, gear housing between the blocks. The other face (inner, x 20) would put the horn at x 4.68..8.83, through the arm lever (x 3..8); the chosen one leaves 2.98 between the horn’s outer face and the lever face for the 3 mm tilt bar.
- Result: tab plane x 26.65; seat x 15.13, outer face 10.98; shaft (y 13.98, z 18.38; y 14.09 since the mating check, below). Was (12.15, 14.1, 18.4) inferred. Settled. Tilt bar plane x 9.5 (on the lever face, x 8.0..11.0). Was 11.6.
Pan servo (PT frame)
Section titled “Pan servo (PT frame)”- Mount tops z 15.0 (STL-PandT servo mount plain/cutout); the tabs rest on them, body down between the mounts (PT-p16 photo and text “screws from above”), shaft +Z.
- Near hole y 29.55 (the end toward the pan base): PT-p7 CAD (horn at the pan-mechanism end of the body, arm and bar toward the upright lever). Linkage: hub to lever hole (-5.5, 25.5) 19.6 for horn 12.75 + bar 15; the other end (y 48.79) is 27.7 away, the bar at full stretch with no room for +-60 deg.
- Conflict: STL places the cutout mount at y 54.9..59.9, the end away from the shaft; PT-p16 (text “the mount with the cutout is at the cable end”, photo: the cable runs through the cutout at the shaft end) and STEP-SERVO (cable at the shaft end) put it at the shaft end (y 27..32). STEP + guide picture + linkage win for the shaft; the mount placement stayed as STL here (swapped since by the mating check, below); the lead Port stays at the shaft end, y 32 face.
- Result: tab plane z 15.65; seat z 27.17, outer face 31.32; shaft (x -20.5, y 38.11). Was (.., 38.1, 25.35) inferred. Settled.
- Pan bar: PT-p7 draws it under the horn arm, on the upright lever top (z 26.8): bar z 26.8..29.8, plane 28.3 (the arm’s underside is z 29.72: 0.08 overlap). Was 31.0 (on the old hub).
Jaw servo (eyes frame)
Section titled “Jaw servo (eyes frame)”- Lug x 73.42..78.42 (STL-jaws servo mount).
- Conflict: the model had the shaft -X (inboard, seat 65.55, near the lower hole). MOUTH-p9 step 3 photo (the shaft goes through the mount plate, the body and the rail foot on the other side) and MOUTH-p14 CAD (horn facing the R side, arm down at neutral) put it +X, outboard. Guide wins.
- Near hole = the upper one (z 42.23): MOUTH-p14 CAD (shaft toward the back end of the mount), and the loop: with the arm down at neutral, the pin (y -61.35, z 33.67) is 44.23 from the jaw lug hole for a 44.1 bar; the lower hole (shaft z 22.99) gives 33.6.
- Result: tab plane x 71.77; seat x 83.29, outer face 87.44; shaft (y -48.6, z 33.67). Was (65.55, -48.6, 23.1) inferred. Settled.
- Connector bar (STL-jaws connector bar), conflict with the old note (“holes along Y, 31.4 apart”): the holes run along local Z at (-22.05, -2.5, 6.5) r 1.1 and (22.05, -2.5, -6.5) r 1.35 (fits.py), 44.1 apart in the plane, and the 3 mm strip is cranked 13 mm along the pin axis. STL wins. Smaller hole on the horn, larger on the jaw (MOUTH-p12 step 8). Jaw end on the bottom jaw lug’s outer face (x 70.42..73.42, MOUTH-p12 photo: bar outside the jaw), horn end 13 mm outboard (83.42..86.42), under the horn arm (85.84..87.44): 0.58 mm overlap, unresolved by any source.
- MOUTH-p10 says “drill … the servo horn at the second position”; the MOUTH-p12 photo shows the bar at the second-from-last hole like the other horns: pin 12.75 kept.
Other values
Section titled “Other values”- Core radio: C3 Table 5-7 (chip, 3.3 V, peak): TX 335 mA at 802.11b 1 Mbps 21 dBm, RX 84 mA (802.11b/g/n HT20). The model’s WIFI_TX_A / WIFI_RX_A already use these; now cited instead of “(assumed)”. The Core’s own regulator path is not published.
- XH2.54 lead 100 mm: PT-p8 hardware list “100mm Wire with 2Pin XH2.54 Connectors” (EYES-p11 prints “00mm”, a typo).
- Servo lead length: not in STEP-SERVO (the wires are cut 25 mm past the shaft), not on the store page (https://eu.store.bambulab.com/en-si/products/9g-servo-motor-with-clutch-protection) or in the electrical note. Stays 250 mm, settled as a best guess (docs/research/2026-11-01-best-guesses.md).
- Mouth channel: CONFIG has the mouth on receiver 2 PWM2; MOUTH-p18 text says “the third position”. Config is the higher-ranked machine source here only by being the file actually loaded; geometry.CHANNELS (PWM2) unchanged.
Switch board 1 (#79, PT frame)
Section titled “Switch board 1 (#79, PT frame)”- Conflict: the model stood switch board 1 on the standoff at (6, 68.5, 13) between the receiver shields (inferred); Cansin’s build (docs/references/build-photos/eyes-top.jpg, confirmed on the real build), EYES-p13 and PT-p23 put it in the base’s two-clip holder at the back edge. Build wins. STL-PandT base: the holder’s two clip shelves (x 18..26 and 43..51, y 111.5..126) and the post (34.5, 108.5) all top z 9 (the post was recorded 7.7: STL wins); clip walls’ inner faces x 18 / 51, back wall y 126. STEP-switch: PCB 20.5 x 32, hole 3 mm from one short edge, the slide switch’s pins 4.6 mm under the PCB. Placed flat, parts up, hole on the post: PCB x 18.5..50.5, y 105.5..126; the pins drop into the gap between the shelves (floor z 3); the lever (tip z 24.2) at the back edge, reachable from behind.
- Its lead to shield 1’s XH socket now runs along the plate’s +x edge past shield 2 (inferred), 69.0 mm of 80; pack 1’s lead 71.6 mm (68.9 straight). The standoff pair at x 6 carries no switch in the build; set 2 keeps the back one, inferred, until a follow-up ticket removes set 2 (stock: one pack, one switch, a Y cable). Done in #81, below.
The base electronics as built (#81, PT frame)
Section titled “The base electronics as built (#81, PT frame)”Sources added: BUILD-TOP / BUILD-SIDE (docs/references/build-photos/eyes-top.jpg, docs/references/build-photos/eyes-top.jpg and
eyes-side.jpg), REST-TOP / REST-FRONT (docs/references/build-photos/eyes-rest-top.jpg, eyes-rest-front.jpg: Cansin, 2026-10-31,
the Mech at the program’s rest), STEP-RECEIVER, STEP-SWITCH, STEP-BATTERY, STL-PandT base, PT-p10 / p15 / p16 / p23,
EYES-p13 / p14 / p15 / p26 / p29.
- Receiver standoffs. Conflict: STANDOFFS had their tops at z 13 (from the switch standoffs); STL-PandT base has the receiver standoffs r 3 to z 16 with r 1.1 locating pins to z 18 on one diagonal of each set (also the #56 parts measurement). STL wins: both shields 3 mm up. The PandT wire holder’s posts top at z 9 (was 7.7, unused).
- The Core. Conflict: the model flipped the Core onto the shield top “per photo”; STEP-RECEIVER has it plugged in under the shield, and BUILD-SIDE shows it there (USB-C between the PCB and the plate), EYES-p14 and PT-p15 show only the shield’s top. Build and STEP win: the Core under the shield, unflipped; its USB-C bottom is 2.3 above the plate top.
- Shield orientation. Cansin’s words (#81): shield 1 rotated so the Core’s USB-C faces outward, away from switch board 1, the opposite of stock. BUILD-TOP and BUILD-SIDE: the shield’s XH socket and the Core’s USB-C both face the plate’s +x edge, which is what the model already had (the RECEIVER_AXES comment said “XH socket to the front”; the code put it at +x). The 23 x 24 standoff pattern allows only that or the 180-degree turn. EYES-p14 (the eyes guide’s own base) has the socket toward the switch board, which is the stock orientation Cansin’s build turns away from; PT-p23 has both shields’ sockets at +x too. Build wins; no rotation was needed. Shield 2 keeps PT-p23’s orientation.
- One pack, one switch board, a Y lead. Conflict: the model had a second pack stacked on the first and a second switch board on the x 6 standoff (both inferred); PT-p10 (the Y adapter “to power both receivers from a single battery”) and PT-p23 (one pack, one switch board, the Y lead’s heat-shrunk joint, both shields’ sockets) have one of each. Stock wins (the build has one shield so far). Battery 2 and switch board 2 are gone; the leads are the pack to the switch board and the Y’s two branches from its output to each shield (each taken as 100 mm: half the 100 mm XH lead plus the battery box’s cut lead, whose length is not published, inferred).
- The x 6 standoff pair. BUILD-TOP shows the eyes build’s pink wire keeper on it, the eye leads under it; PT-p23 step 18 moves “the original cable hold from part 1 … onto the new base” and passes the servo wires under it. The keeper is now in the model on those posts (STL-wire keeper, holes 24 apart, 3 mm thick), and the eye leads route under it.
- The pack. Conflicts: BATTERY_POCKET_PT (x -48..-24, y 71..127) against STL-PandT base’s pocket x -50..-21.5, y 73..125, floor z 3 (STL wins: the pack is 51.2 x 28.4); and the pack’s lead at the back end (the STEP’s frame as it came) against BUILD-TOP and PT-p23, where the lead leaves the pack’s front end. Build and guide win: the pack is turned end for end, its back end on the pocket’s back wall, the plug at (-39.25, 72.5, 14.65).
- Switch board sockets (STEP-SWITCH): two XH housings at the PCB ends (x -5.75..4.25, y 1.4..7.6, z +-5..15.5); the pack’s lead goes into the -x one and the Y lead leaves the +x one (BUILD-TOP). The corridors now start and end there instead of at the mounting hole.
- Corridors, routed by BUILD-TOP: the pack’s lead from its front end across the plate and back up the middle to the switch board (91.2 mm of 100); the Y lead over the plate’s +x edge and along it into each shield’s socket from +x (94.9 mm to shield 1, as the photo’s loop; 61.9 to shield 2). A lead no pose stretches (min = max over every pose) needs no slack: the 20 mm is for movement (PT-p23 step 18), so its limit is the lead’s length. The eye leads pass under the keeper on the desk (stage 1) and on the neck (stages 2 and 3).
- Electrical budget: one pack carries both shields: every servo at stall plus both shields and a Wi-Fi burst each is 5.34 A against the pack’s 3 A continuous (each shield stays under its own 3 A: 2.995 / 2.345). The stock wiring’s worst case; one pack per shield is a planned edit (docs/model-notes.md). Runtime is now the one pack’s.
Horns at servo 90 (#81, REST-TOP, REST-FRONT)
Section titled “Horns at servo 90 (#81, REST-TOP, REST-FRONT)”- Build facts (Cansin, 2026-10-31): all four eye-mech horns take their bar at the second hole from the end; the two lid horns have that hole drilled out, the eye horns not. Same as stock: EYES-p15 step 5 (“in 2 of the servo horns, drill 2mm holes at the second from last position”), p26 step 17 (the undrilled ones on the eyeball bars), p29 step 22 (an M2 screw and nut through the drilled hole and both eyelid bars). The model already pins every bar at 12.75 (STEP-HALF, the second from the end); geometry.HORN_STEP pin_r records r 1.0 for the lid horns, 0.5 for the eye horns.
- Commanded at rest (the repo’s own mapping: eyes/eyes.py
restwith eyes/calibration.json, model/sim.joint_deg): l_lid 117.96 (joint -27.96, flipped), r_lid 67.8 (joint -22.2), both eyes 88.0 (joint -2.0). The model reads lid fraction 0.28 as horn -14.49 (model/traces.py, between the model’s closed -30.7 and open 27.2). - Read off REST-TOP (1125 x 2000; scale and the Mech’s 1.5-degree turn in the picture from the two eye-horn hub screws, 53.88 mm apart = 699 px): eye horn arms (hub screw to bar screw) L 7.1 deg toward the back (joint +7.1), R 4.7 toward the back (joint -4.7); lid horns (seen end on, the bar screw’s offset along Z from the shaft at z 40) L -15.2, R -16.3 joint. Taken at face value the spline tooth at servo 90 would be L lid +12.8, R lid +5.9, L eye +9.1, R eye -2.7.
- Not settled. Those eye readings put the eyeballs 12 degrees apart, which REST-FRONT does not show (both pupils centred); the common part is a camera tilt: the bar screws sit 3 mm above the hub screws, so a camera leaning about 23 degrees moves both 1.3 mm the same way, which makes both eyes +1.2 (tooth +3.2 each). The same tilt moves the lid screws (6 mm below the reference) by 2.6 mm and the lid readings to about -3 / -4 (teeth +25 / +18). The picture cannot separate tooth from tilt by more than about 12 degrees on the lids. Neither reading clears KNOWN_CALIBRATED: the lid ends land at -41 / -40 or -29 / -28 (the crank step meets the bracket from -38) and the eye ends stay past +-24 either way. They stay listed; a side picture of each horn at servo 90 (or the horn-arm angle on a protractor) would settle it.
- Settled for the lids (#134) by the calibration itself: the real Mech winks shut at lid_closed (0.1: L servo 134.7, R 52.5). In the model the edges meet at horn -30.7 and the lid crank meets the eye bracket from -32 (penetrating from -33), so the shut, collision-free pose needs a tooth of L 12.7..14.0 and R 5.5..6.8. The face-value reading (+12.8, +5.9) lands in both windows and the camera-tilt one (+25 / +18) leaves the lids 26 deg open: face value it is. geometry.HORN_AT_90 takes L +13, R +6; model/sim.joint_deg and model/traces.py add it. The calibrated ends then land at L -41 / +52, R -40 / +45: past the model’s free travel (-32..+39) on both sides, the crank 1.0 mm into the bracket and the lower lid 1.3..2.0 mm into the servo bracket, while the real Mech runs them inside its stops. The model’s lid travel (71 deg) is short of the build’s (93 / 85 servo deg): model/test_report.py KNOWN_CALIBRATED.
Limits re-checked (make sim, engine sweeps)
Section titled “Limits re-checked (make sim, engine sweeps)”- Eyes: model +-20 -> +-22 (68..112): the eyeball bar rides on main bracket v2 (0 mm contact) and enters it 0.10 mm at +-24, 0.26 at +-28. Designer 60..120 (L) / 75..120 (R) not reached.
- Tilt: lower end stays 30 (horn 0.06 mm clear at -62, 0.38 deep at -66, 1.16 at -70; 30 keeps the pan x tilt grid on neutral).
- Mouth: with the shaft outboard the opening side is negative horn as before; model 50..90 -> 15..90 (horn -75..0 = jaw -58..0 deg; the connector bar enters the jaw servo 0.28 mm at -80). Positive horn still drives the jaw further shut (not swept: the hinge pair) and has no loop solution past +22: one-sided.
- Lids and pan: unchanged (lids clean -30..39; pan -50/-45 still unreachable, the gate’s KNOWN entry).
- The calibrated ends the real Mech runs (eyes/calibration.json) still collide in the model and are now listed with their reasons (model/test_report.py KNOWN_CALIBRATED) instead of warning as “pivot inferred”: lids at -54 / -46 (the eyelid crank’s step 1.0 mm into main bracket v2’s outer edge from -38) and eyes at -32 / +28 (bar 0.26..0.5 mm into the bracket, the eye horn 0.2..0.5 mm into the brackets at -32).
Left open here, settled since
Section titled “Left open here, settled since”Servo lead length; whether the horns seat 11.52 past the tab plane on the real spline (the STEP spline is simplified: bore r 2.15 vs spline r 2.35); the jaw bar/horn-arm overlap; the eyelid crank’s plane against main bracket v2’s edge and the eye horns’ tooth (the calibrated-end collisions). Each is settled by a documented best guess in docs/research/2026-11-01-best-guesses.md (#170).
Mating check (#67, model/engine/mating.py, make sim)
Section titled “Mating check (#67, model/engine/mating.py, make sim)”Every screw, pin and shaft joint is checked on the meshes: each part’s round feature (hole, pin, spline, bore) fitted near its declared centre (fits.py), carried by the part’s placement; axes within 0.2 mm. Each servo must touch the mount its near tab screw goes into and enter no part (the horn on the simplified spline excepted), and its lead must leave the case end through no printed part. 64 joints over the three sub-assemblies, plus the seven seats and leads (tab screws, splines, horn pins, bar pins, lid ears on the bracket pins, eyeball screws and pivots, bracket feet, jaw hinges, mount feet, the tilt and pan pivots, the upright screw); the electronics are not in it (their STEP holes are not clean circles; their standoff placement is stages.placements). First run on the #62 model, and what settled each:
- Lid bars on the lid pins: 0.63 (outside) and 0.90 (inside) mm. Conflict: the lid STLs’ pose with the 41 mm bars (STL-eyelid bar/crank, fits 41.0) vs the horn along +Y at neutral (an inference of the model, no source). STLs win: the horn’s neutral arm turns 3.59 deg from +Y toward the lids (geometry.LID_HORN_ARM), the best fit of both loops: the pin is 41.14 from the inside lid’s pin and 40.86 from the outside one’s, 0.14 mm left at each (no one horn angle closes both: the STL lids are 0.28 apart in that sense). The lid curves move: +20 deg horn now gives inside 21.02 / outside -19.81 deg (was 21.24 / -19.33). The bars’ placement tolerance drops from inferred (3 mm) to measured.
- Eye servos 0.10 mm into the L/R servo bracket. Conflict: the eye lug holes (STL, fitted z 26.88) put the case side (STEP-SERVO case y -0.1..11.9: 6.0 from the hole axis) at z 20.9, under the bracket floor face z 21.0 (STL). Both STLs stand: the case rests on the floor, shaft z 27.0, and the tab screws take 0.12 (inside the 0.2). The eyeball bar’s horn hole is at z 27.0 too (STL), so the eye horn pin is now purely radial: 12.56 from the shaft against the STEP horn’s 12.75 hole, 0.19 mm (passes; the bar self-taps its r 1.05 hole).
- Tilt servo: far tab screw 0.22 mm, gear housing 0.11 into the far tower block. Conflict: the tower’s holes are 27.58 apart (STL-PandT pan base, fitted y -5.04 / 22.54), the servo’s tabs 27.8 (STEP-SERVO). Each part keeps its own; the servo sits centred on the pair (each screw 0.11 off) and its housing end then clears the block face (y -2.58). Shaft y 13.98 -> 14.09.
- Pan servo lead into the plain mount. Conflict (above): the STL has the cutout mount at the far end, the lead leaves at the shaft end (STEP-SERVO) through the cutout (PT-p16 text and photo). STEP + guide win over the STL’s placement: the mounts trade places, 27.9 apart (the base’s foot holes, geometry.MOUNT_SWAP_Y); their tab holes are then 28.0 apart, 0.1 off each of the servo’s; the pan shaft is unchanged.
- Jaw horn pin: the connector bar enters the horn arm 0.28 mm (the 0.58 mm overlap along the shaft above). The #68 guide sheet read MOUTH-p13 (steps 8-9) as the bar on the horn’s second-from-end hole with the model “four holes nearer the shaft”. Measured: the bar’s horn hole is 12.75 from the bore axis, coaxial (0.000) with the STEP horn’s 12.75 hole, the second from the end (the arm ends at 16.05, past the 14.75 hole), as the photo shows (the horn just past the bar screw); the sheet’s “four holes nearer” is not borne out (STEP holes > STL bar > guide image agree on 12.75). The photo also has the horn outside the bar, as modelled; how far the bar sits under the arm it does not show. Settled since as flush seating (docs/research/2026-11-01-best-guesses.md).
- Everything else lines up first time (worst: tab screws on the lid lugs 0.10, the pitch 27.7; the upright screw 0.035).
- Ground-truth order since (Cansin, 2026-10-31): his photos for how things are assembled (docs/references/build-photos, local only), then measurements of the real Mech (eyes/calibration.json), then the stock sources. No build evidence touches the four fixes above: the eyes-only build has no neck or mouth; docs/references/build-photos/eyes-top.jpg agrees on the eye servos (shaft at the outboard end, arm outboard to the eyeball bar) and holds no 0.1 mm detail; PHOTO shows the lid horn at an unrecorded servo angle, so it does not fix the horn’s neutral). The calibrated ends the build runs (model/test_report.py KNOWN_CALIBRATED: lids at -54 / -46, eyes at -32 / +28) still collide in the model, unchanged by these fixes: the build wins, so they are model errors, but nothing in the repo says which geometry (most likely each horn’s spline tooth at servo 90); they stay listed until a build measurement settles them.
noggin v0.2.26-6-g09bf9b7 · 09bf9b7