feat(disney_ear_hanger): two-piece dovetail Multiconnect mount + slim 6mm plate (pst-j3ej)

[SeanOC]

What & why (pst-j3ej, operator-picked A1+B1)

Redesigns the Multiconnect ear-hanger mount, which the operator flagged as (1) far
too thick for a trivially light headband and (2) unprintable without heavy supports
(the one-piece welded the slot plate perpendicular to the saddle).

A1 — slim plate

The Multiconnect slot slab drops from 6.5mm slab + 4mm plate cap (~10mm) to a
slim 6.0mm slab (4.15mm engaging recess + 1.85mm back wall) via patch-0002's
backThickness param, which keeps the engaging profile identical.

B1 — two-piece dovetail

mount_type splits multiconnect into two fanned, separately-printed STLs:

variant	is
multiconnect_plate	slim slot plate + male dovetail rail
multiconnect_saddle	saddle + solid wall-end backing + female socket

earHanger() is untouched, so tab stays byte-identical and opengrid is unchanged.
Joint: the rail slides across the wall so gravity + the cantilever lever-out
bear on the dovetail shoulders, not the slide axis (operator's load rule).
Friction fit (dovetail_clearance ~0.25mm, @PARAM) + an optional glue dot.

Plate orientation — bug fixed, then stood up (the last three commits)

The first cut of the plate had the slot slab's rotation inverted, so the exported
part met a Multiconnect board with a solid face — the slot openings pointed at the
rail instead of the board, and the plate could not engage at all. Codex-review caught
it; no machine invariant did.

• e4cd6eb (codex-gate fix) flipped the slab so the slots face the board — but in a
  flat print that lands the retention undercuts as ~107 mm² of severe
  (<25°-from-horizontal) downward overhang: functional but needs support, which
  breaks the bead's "zero-support both" rule.
• 7d7207d (this branch head) resolves it per the operator's decision this session:
  print the plate STANDING — a wall in the X-Z plane on its z=0 edge, slot openings
  horizontal — the standard zero-support Multiconnect orientation the sibling
  backers (apple_tv) already use. The rail protrudes horizontally off the solid back and
  adds no overhang; its trapezoid is unchanged (24 × 4 × 8→12) so it still mates the
  socket. (A flat print can't both face the slots at the board and stay support-free —
  openings and rail are opposite faces of the same axis.)

New machine guard so this can't regress: mc_plate-slot-direction — AIR just inside
the -Y board face at the central slot, SOLID behind it (verified: fires on a Y-mirrored
mesh). _check_mc_plate was rewritten for the standing frame.

Verification

• All 4 variants watertight / 1 solid / seat min-Z=0 (CGAL export pipeline).
• tab default STL geometry unchanged; opengrid + multiconnect_saddle untouched.
• Standing plate: slots open on the -Y board face across the range; rail flares
  8→12 over 4mm depth × 24mm long, mates the socket; CGAL watertight/1-solid/seats
  z=0 at width/height 1..6, on_ramp, retention, clearance extremes.
• wasm param-sweep 31/31; 218/218 unit tests; invariants ok.
• Plate renders regenerated (standing). Stale hand-made assembled render dropped
  (render-all doesn't maintain it) — the per-part isos + the header assembly note cover
  the two-piece.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
stuff	Ready	Preview	Aug 21, 2026 9:52pm

[SeanOC]

Codex gate review — changes requested

One blocking defect; everything else verified solid.

BLOCKING — the Multiconnect plate is printed slot-side up; it cannot engage a board.
models/disney_ear_hanger.scad:490 applies rotate([-90,0,0]) to multiconnectBack(). In the vendored lib frame the slot tools are recessed from the y=-t face (openings there; blind back wall at y=0) — see scripts/patches/QuackWorks/0002 and the sibling convention models/apple_tv_4th_gen_holder.scad:352-370 (rotate(180,[0,1,1]), openings on the panel face). rotate([-90,0,0]) maps y=-t -> build z=+6, so the openings land on the TOP (rail side) and the blind face on the bed. Measured on exports/disney_ear_hanger-multiconnect_plate.stl: the z=0 cross-section is the full solid footprint (no slot voids), the channels occupy z[1.85,5.35] opening upward, and only a 1.85mm wall sits between the slots and the board face. Mounted flat on a Multiconnect board the plate presents a solid face to the rails — the part fails its sole function. The header comments (scad:69-77, 471-477) assert the opposite orientation.

Smallest honest fix (verified on a patched copy): at models/disney_ear_hanger.scad:490 replace rotate([-90, 0, 0]) with translate([0, H, mc_thickness]) rotate([90, 0, 0]) — watertight, 1 solid, seats z=0, footprint/height unchanged, slot void z[0,4.15] opening at the board side, 1.85mm blind wall now under the rail (flanks still self-supporting). Correct the inverted comments, add a sidecar probe pinning slot direction (AIR at [W/2, H/2, 1.0], SOLID at [W/2, H/2, 5.0]), regenerate renders/exports.

Why CI missed it: _check_mc_plate (models/disney_ear_hanger.invariants.py:263) probes footprint/height/rail and a corner column outside the slot ladder — it passes for either orientation.

Non-blocking notes:

• renders/disney_ear_hanger/assembled_multiconnect.png is not produced by scripts/render-all.py (views are top/front/side/iso) and will silently go stale; teach render-all an assembled view or drop it.
• No cross-part mating assertion; I verified fit manually (boolean intersection 0 mm³, rail fully engaged with 0.3mm face-seat gap) — a sidecar probe would pin it.

Verified good: dovetail joint mates with zero interference; tab/opengrid byte-identical/unchanged; all exports watertight + seated; invariants and CI green; pins and license split untouched.

[SeanOC]

Codex gate review — pass

codex-review status set to success on e4cd6eb.

What I verified independently (fresh worktree, pinned engine 2025.06.12.ai25773, vendored libs re-cloned at their pins):

• All 4 filename-grid STLs (tab, opengrid, multiconnect_plate, multiconnect_saddle) export watertight, 1 solid, min-Z=0; scripts/check-invariants.py disney_ear_hanger → ok with the rewritten sidecar.
• Tab byte-identical: re-exported from main and from this head with the same engine — identical vertices/bounds/volume (9551.24 mm³).
• Slim-plate claim holds against the vendored source: multiconnectSlotDesign.scad:42 depends on backThickness only through the tool translate, so 6.0mm keeps the ~4.15mm engaging profile identical and adds/removes only the meat behind the blind face.
• Joint mates: rail measured 8.4→11.8mm flare over 4mm depth (design 8→12), socket 8.2→11.9mm (design 8.25→12.25), flank slopes match; an assembly simulation mapping ~99k rail interior points into the mated saddle frame found zero interference. Socket open both ends, backed solid at the blind end.
• pst-cu7n flip verified on the mesh: board-side (z=1.0) air, rail-side (z=5.0) solid — slot mouths face the board.
• No stale references to the old one-piece multiconnect variant anywhere in app/lib/scripts/docs; presets updated. License header + CC BY-NC-SA split intact.

Non-blocking nits (not gate-worthy, suggesting a small follow-up if anyone wants):

1. models/disney_ear_hanger.invariants.py:57 — _DT_CLEAR is defined but never used.
2. models/disney_ear_hanger.invariants.py:117 — _air_span_z reports max−min over all air points between the probe bounds; at deep socket depths a second air interval outside the part inflates the number (14.8 reported vs ~11.9 true span). Measuring the air interval containing dovetail_z would be more precise; harmless for the current monotonic-flare check.
3. dovetail_backing() comment says it refills "the arch channel" — it actually refills the arch's full outer silhouette across the wall slab (including the upstream clip regions, verified on the mesh). Intentional and correct for the flat seat; comment just understates it.

Slot-down/rail-up print orientation judgment call is sound (a saddle-facing rail can't coexist with slot-up in a flat print) — flagging for operator awareness only.

[SeanOC]

Codex gate review — pass

codex-review status set to success on 7d7207d (re-review of the standing-plate commit).

Verified independently (scratch worktree, pinned engine, libs re-vendored at their pins):

• All 4 STLs export watertight / 1 solid / min-Z=0; check-invariants green; CI all green.
• Standing plate: bounds 28×(6+4)×56, slots open on the −Y board face, solid back behind; rail flare 8.6→11.8mm, length 23.6mm; joint re-mated into the saddle with zero interference (~99k rail points).
• Tab byte-identical to main; opengrid/saddle geometry unchanged from the prior revision (vertex-set equal; only CGAL triangle-order noise).
• Zero-support claim tested with the repo's own scan convention (apple_tv header documents it: flag faces >50° from vertical, exclude bed): standing plate = 228 mm², vs 1319 mm² for the shipped apple_tv standing backer under the identical scan — ~6× cleaner. The flat revision it supersedes measured ~1032 mm² of steep retention-dome ceilings + slot-mouth bridges, which this orientation eliminates. Residual is the inherent slot lead-in/on-ramp walls of any standing Multiconnect print.

Non-blocking findings:

1. models/disney_ear_hanger.scad:474-481 — the old flat-orientation PART A comment block was left in place with the new standing block appended (482-489); the two contradict each other and 474-481 describes a rotation this revision no longer uses. Smallest fix: delete lines 474-481.
2. Carryovers from the prior review, still present: unused _DT_CLEAR (invariants.py:57), _air_span_z max−min over all air intervals (invariants.py:117), and the dovetail_backing() comment understating its refill scope.

56mm-tall standing wall with a 72mm² bed footprint — brim if tippy is your call, and the header says so honestly.

[SeanOC]

Codex gate review — changes requested

codex-review set to failure on 7d7207d. One blocking finding on the dovetail joint; everything else in this revision verifies clean. Note: this replaces my earlier pass on e4cd6eb — my containment-probe measurements under-sampled the joint's exact flank widths; plane-slice cross-sections (below) expose the defect, which has been present since the feature commit and is therefore in-scope for this PR.

BLOCKING — the dovetail joint is an interference fit, not a clearance fit

What: Exact cross-sections of both exported STLs (plane-slice, no probe ambiguity), plate centre x=14, mated at the wall face:

depth from wall face	rail z-width	socket air gap	clearance
0.02	8.551	8.314	−0.237
1.0	9.422	9.205	−0.218
2.0	10.311	10.114	−0.197
3.0	11.200	11.023	−0.177
3.9	12.000	11.841	−0.159

The rail is 0.16–0.24mm WIDER than the socket at every depth — the parts collide on the flanks and cannot slide together (the joint's entire purpose). At print tolerances of 0.1–0.15mm this is a hard bind, not a snug friction fit.

Why: plate_rail_standing() (models/disney_ear_hanger.scad:528-535) hulls the neck cube from y=−0.3 (the bury embed) to the tip band. The hull spreads every cross-section toward the wider tip: at the plate face (y=0) the rail reads ~8.53mm, not dt_neck (8) — the flank slope becomes 0.89mm/mm instead of the socket's 0.80mm/mm ((dt_tip+c−dt_neck−c)/(dt_depth+0.4)). So even though the socket mouth is dt_neck + c = 8.25 > 8, the actual rail base is already past it before the joint starts. The flat revision's plate_rail() had the identical hull-spread (verified: its base-layer z-width is also 8.551), so this has been present since 5c46507 — but it ships nothing a user can assemble, so it blocks.

Smallest honest fix (pick one):

1. models/disney_ear_hanger.scad:175 — set dovetail_clearance = 0.8;. Then mouth = 8.8 vs rail base 8.533 (+0.27 entry clearance) and clearance ≥0.27 at every depth. One line; the @PARAM already documents "bump it if the parts bind". If you want to keep the 0.25 nominal, compensate for the hull spread instead (see 2).
2. models/disney_ear_hanger.scad:531-532 — anchor the neck cube's top face at the plate face so the hull cannot spread the base: e.g. translate([0, -bury, 0]) cube([dt_len, bury, dt_neck], anchor=BOTTOM) (or an equivalent non-hull prism for the buried root + a pure trapezoid hull above y=0). Rail base then reads 8.0 and the existing 0.25 clearance works as designed.

Either way, please verify post-fix with a plane-slice measurement (rail z-width at x=W/2 vs socket air gap at y=0 across depths 0..3.9) — a containment-probe check will not catch 0.2mm flank errors, which is how this slipped through the sidecar and two gate reviews. Consider adding that slice-based flank-clearance assertion to _check_mc_plate/_check_mc_saddle (the sidecar currently only asserts the flare is monotonic).

Non-blocking

1. models/disney_ear_hanger.scad:474-481 — stale flat-orientation PART A comment block left in place, directly contradicted by the standing block at 482-489; delete 474-481.
2. models/disney_ear_hanger.scad:520-527 — rail comment claims "Same trapezoidal shape ... the saddle socket receives" — false as built (hull spread); update with whichever fix you take.
3. Carryovers: unused _DT_CLEAR (invariants.py:57); _air_span_z max−min over all air intervals (invariants.py:117); dovetail_backing() comment understates its refill scope.

Verified clean in this revision

• Standing plate geometry: watertight/1-solid/min-Z=0 across all 4 variants; footprint 28×56, depth 6+4; slots open on the −Y board face; rail 24mm long, 8→12 flare; bed seat 72mm² wall footprint.
• Zero-support claim holds under the repo's own scan convention: 228mm² flagged vs 1319mm² for the shipped apple_tv standing backer; flat revision's 1032mm² of steep ceilings eliminated.
• Tab byte-identical to main; opengrid/saddle geometry unchanged from e4cd6eb; CI all green; license split intact.