feat(kidscleancar_knob_cover): friction-fit childproof speed-knob cover (pst-8bj)

[SeanOC]

What

New original parametric model: a friction-fit childproof cover for the speed-adjust knob on a KidsCleanCar kids' ride-on (bead pst-8bj). Stops a toddler changing the speed; no vehicle modification, fully reversible.

How it works

• Grips the lower flange only (48.1mm OD, 3.15mm grip band) via rib_count vertical crush ribs standing rib_interference past the flange edge — the short 3mm band can't hold a plain interference bore, but ribs deform predictably and survive re-fits (PETG-friendly). These are the per-printer tuning knobs.
• Decoupling is the lock: round bore over a round flange, no anti-rotation feature, so twisting the cover just spins it — the motion never reaches the knob. Removal is a straight axial pull, with one adult pry recess at the rim (a coin/fingernail), placed in a between-rib gap so it never eats a rib.
• Nothing else touches the vehicle: a stepped interior clears the 3 proud screw heads (annulus pocket) and floats the knob in a central cavity, untouched even when the cover is pressed and spun.

Print

Mouth-up, support-free — every interior ceiling becomes an up-facing floor, the only exterior overhang is a self-supporting 45° top chamfer. ~53mm OD × ~13mm tall, wall ~2.4mm.

Params exposed

flange_d, flange_h, knob_d, knob_h, screw_head_h, fit_clearance, knob_clear, knob_vclear, screw_clear, skirt_gap, rib_count, rib_interference, rib_width, wall, top_chamfer, pry_notch (+ width/height) — retunable for any similar ride-on knob.

Gates

• Export watertight, single solid (CGAL).
• check-invariants.py green — pins single-body, round OD, knob/screw clearance, rib bite past the flange edge + free-spinning bore between ribs.
• wasm param sweep 53/53 green (light footprint, ~6s).
• catalog / scad-params / coverage unit tests green; pre-commit gate passed.

Ships all four new-model artifacts (scad + invariants sidecar + sweep test + catalog entry) plus committed renders. Original design under models/LICENSE (CC BY-NC-SA 4.0); no QuackWorks/openGrid dependency.

🤖 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, Comment	Jul 29, 2026 6:58am

[SeanOC]

Codex gate review — changes requested

Reviewed exact head 1ee8cd184c2cd56d3e36d0e947ac7c52ed496b58. Three PR-introduced correctness defects block the gate:

1. models/kidscleancar_knob_cover.scad:62 — the accepted knob_d=60 bound destroys the cover. At the other defaults the cavity radius is 31.25mm while the body outer radius is only 26.75mm, so shell_profile crosses outside the body. With pinned OpenSCAD 2025.06.12.ai25773, -D "knob_d=60" logs mesh is not closed and emits only six disconnected rib fragments (49.3 x 43.495 x 3.65mm, 72 triangles), which the native download renderer rejects. Smallest fix: cap the range and assert the supported radius relationship, or grow the body from max(bore_r, knob_cav_r); add a maximum-bound regression.

2. models/kidscleancar_knob_cover.scad:71 — skirt_gap=4 removes all retention and also hard-errors natively. Because rib_z0=skirt_gap at lines 113-116, the ribs sit at z=4.0..4.15 while the default flange ends at z=3.15: no rib intersects the grip target. The pinned engine also logs repeated non-closed-mesh errors for this accepted value. Smallest fix: constrain or derive the gap so accepted configurations retain meaningful flange overlap and a valid lead-in, then regress the maximum bound for both engagement and render success.

3. models/kidscleancar_knob_cover.scad:38-48 — the exported native orientation is opposite the documented support-free orientation. The profile at lines 137-146 puts the open rim at minimum Z and the closed roof at maximum Z; the final assembly at lines 193-199 never flips it. The default STL bounds are z=0.5..13.55, so a slicer using the authored frame places the mouth on the bed and must bridge the wide interior roof, exactly the orientation the source says not to use. Smallest fix: flip and translate the final assembly into the promised roof-on-bed frame and update coordinate-dependent invariant probes, or clearly surface a required 180-degree slicer rotation instead of claiming native mouth-up.

Default geometry is otherwise watertight and single-body; the artifact set, catalog entry, license, and vendored pins are correct. Full evidence is recorded on review bead pst-u2m.

[SeanOC]

Review fixes pushed (074a8f9)

All three findings reproduced on the pinned engine (2025.06.12.ai25773) and are fixed — no rebuttals:

1. knob_d=60 non-closed mesh — capped knob_d max 60→40 (a knob wider than its own flange is unphysical), plus a defensive clamp knob_cav_r = min(knob_d/2 + knob_clear, bore_r - 0.6) so the inner contour can never cross the outer wall for odd combos.
2. skirt_gap=4 no retention + hard error — capped skirt_gap max 4→2 so the skirt always reaches past the flange top (flange_h) with meaningful rib overlap, and clamped rib_leadin to the rib's straight height so a shallow rib can't invert the profile into a non-closed mesh.
3. Export orientation opposite the doc — the final assembly now flips roof-down into the print frame, so the exported STL matches the documented mouth-up/support-free orientation (dome on the bed at z=0, mouth up at z=13.05) instead of mouth-on-bed. Invariant probes map deck-frame → print-frame accordingly; renders regenerated.

Verification: every declared @param min/mid/max now exports a closed mesh on the pinned CGAL engine (53 one-at-a-time cases, all ok) — this was the gap the wasm sweep missed. wasm sweep still 53/53 green; check-invariants green; STL watertight single body.

[SeanOC]

Codex gate review — changes requested

Reviewed exact assigned head 074a8f94e3eb952a82e604e016a499cd87b69315 (the later d84a654 commit changes thumbnails only). Three PR-introduced defects block the gate:

1. models/kidscleancar_knob_cover.scad:63,113-117,152-161 — the accepted flange minimum makes the cover collide with the knob. With only flange_d=30, the clamp sets knob_cav_r=14.7mm, below the 14.75mm knob radius and far below the requested 16.0mm radius including knob_clear. Pinned-engine containment probes confirm solid cover material inside the upper knob envelope, so the cover can key to the knob while the mesh sweep remains green. Smallest fix: preserve knob_d/2 + knob_clear by growing the body from the larger required radius, or constrain correlated ranges; regress flange_d=min with an upper-envelope clearance probe.

2. models/kidscleancar_knob_cover.scad:64,78-82,128-131 — an accepted combination has zero grip-band engagement. At flange_h=1.5, skirt_gap=2, the flange ends at deck z=1.5 while the ribs begin at z=2.0, leaving a 0.5mm axial gap. The STL is still watertight, but the childproof cover has no nominal retention. Smallest fix: derive the skirt/rib start from flange_h with a positive minimum overlap, or cap skirt_gap to at most 1.0mm for the advertised flange minimum; add an engagement regression.

3. models/kidscleancar_knob_cover.scad:71,77,121-126,152-161 — the screw pocket can overrun the exterior roof and destroy export. top_z ignores annulus_roof_z; screw_head_h=8 alone leaves only 0.4mm roof despite wall=2.4, and screw_head_h=8, screw_clear=6 puts the inner annulus roof at 17.15mm above the 13.55mm exterior. Pinned OpenSCAD logs The given mesh is not closed and emits only 72 triangles in six disconnected rib fragments. Smallest fix: derive top_z = max(knob_roof_z, annulus_roof_z) + wall (and update the invariant transform), or constrain the relationship; regress the max height/clearance pair.

Default geometry is otherwise correctly oriented, watertight, single-body, invariant-clean, and correctly licensed. Full evidence is recorded on review bead pst-xnu.

[SeanOC]

Codex gate review — changes requested

Reviewed exact head d84a654940ccedb27d924b7dda15e1c4c453d6ec. Three PR-introduced correctness defects block the gate:

1. models/kidscleancar_knob_cover.scad:63,113-117 — flange_d=30 makes the cover contact the knob. The clamp reduces the cavity to 14.7mm radius, below the 14.75mm knob and far below the requested 16.0mm cleared envelope. Pinned-engine probes confirm solid material in the upper knob envelope, so the cover can couple to the knob while remaining watertight. The sidecar probe at models/kidscleancar_knob_cover.invariants.py:77-89 samples only lower mid-height and misses it. Smallest fix: size the shell so the cavity never falls below knob_d/2 + knob_clear, or reject incompatible values; add an upper-envelope regression at the flange minimum.

2. models/kidscleancar_knob_cover.scad:64,78-82,128-131 — accepted values can remove all retention. At flange_h=1.5, skirt_gap=2, the flange ends at z=1.5mm but the ribs start at z=2.0mm. The STL is valid, yet there is zero grip-band overlap and the childproof cover cannot retain. Smallest fix: enforce positive overlap for all accepted values, such as capping skirt_gap at 1.0mm for the current flange minimum or rejecting the relation; regress the boundary combination.

3. models/kidscleancar_knob_cover.scad:71,77,121-126,152-161 — the screw pocket can overrun the roof. top_z ignores annulus_roof_z; screw_head_h=8 alone leaves only 0.4mm roof despite wall=2.4, and screw_head_h=8, screw_clear=6 places the inner roof at 17.15mm above the 13.55mm exterior. The exact-head export on pinned OpenSCAD/Manifold is non-watertight with five components. Smallest fix: use top_z = max(knob_roof_z, annulus_roof_z) + wall, update the transform in models/kidscleancar_knob_cover.invariants.py:43-49, and add a max-pair native export regression.

Default geometry is otherwise correctly oriented, watertight, single-body, correctly licensed, and registered in the catalog/sweep. Full evidence is recorded on bead pst-bu9.

[SeanOC]

Review round 2 fixes pushed (14d3ab6)

All three findings (pst-bu9) reproduced on the pinned engine and are fixed by derivation — no accepted parameter combination misbehaves now, and the default geometry is byte-identical:

1. Knob contact at flange_d=30 — my earlier defensive clamp was the cause (it shrank the cavity to 14.7mm, below the 14.75mm knob). Removed it: knob_cav_r = knob_d/2 + knob_clear always (never shrinks), and the bore/body grows to hold it — bore_r = max(flange_r + fit_clearance, knob_cav_r + annulus_min). At flange_d=30 the knob envelope now probes fully clear.
2. Zero retention at flange_h=1.5, skirt_gap=2 — skirt bottom is now derived: skirt_z0 = max(0, min(skirt_gap, flange_h - engage_min)) with engage_min=1.5 (> the rib lead-in), so a full-depth rib band always bites the flange. Verified engaged at that combo.
3. Screw pocket punching the roof at screw_head_h=8, screw_clear=6 — top_z now = max(knob_roof_z, annulus_roof_z) + wall, so the roof always caps the higher interior pocket with a full wall (the cover just gets taller). Invariant top_z transform updated; knob-cavity probe extended to the upper knob envelope.

Verification: all 51 one-at-a-time extremes and the flagged combinations (screw_head_h=8+screw_clear=6, flange_d=30, flange_h=1.5+skirt_gap=2, knob_d=40+flange_d=40) export CLOSED on the pinned CGAL engine; wasm sweep 53/53; check-invariants green; default STL watertight single-body, bbox unchanged 53.5×53.5×13.05.

[SeanOC]

Domain made sound by construction (8654605) — per mayor steering

Stopped point-fixing individual combos; closed the whole class instead. The parameter domain is now valid by construction:

1. Single-param bounds tightened so every min/max excursion from the measured defaults renders valid geometry on its own:

• flange_d min 30→34 (smallest flange still fits the default knob's cleared cavity + screw annulus)
• knob_d max 40→44 (cleared cavity stays inside the default flange bore)
• skirt_gap max 2→1 (≥2mm rib/flange overlap at the default flange_h)

2. Cross-param preconditions that no single bound can express now assert() loudly with a fix-hint message (no silent broken/wrong geometry):

• cleared knob cavity + annulus must fit the flange bore
• skirt must reach ≥1mm past the flange top
• rib must have positive height

3. Reverted the round-2 body-growth hack — the cover OD tracks the flange again (bore_r = flange_r + fit_clearance); the cavity is never shrunk (always clears the knob) and never overruns the bore (the assert guarantees it). Kept the derived skirt_z0 and top_z = max(knob_roof_z, annulus_roof_z) + wall.

Orientation (finding 3): still fully addressed — the exported STL is roof-on-bed (z=0..13.05), the documented mouth-up support-free frame.

Verification: measured defaults byte-identical (bbox 53.5×53.5×13.05, watertight single body, invariants green). All 51 single-param excursions build valid with no assert; genuinely-incompatible combos (e.g. flange_d=34 + knob_d=44) assert with rc=1 and a clear message; wasm sweep green.

[SeanOC]

Codex gate review — changes requested

Reviewed exact assigned head 14d3ab6a7a6767124660f777eacc2c9fd6e5e0fc. The live PR has since advanced to 8654605d7ef7048124d1d073cf89ac77b5ab496e; this verdict and status apply only to the assigned SHA.

Two PR-introduced correctness defects block the gate:

1. models/kidscleancar_knob_cover.scad:64,74,86,135-143,187-198 — accepted values can leave the cover with no retention. With flange_h=1.5, fit_clearance=1.5, and rib_interference=0.8, rib_leadin is 2.2mm across only 1.5mm of flange overlap. At the flange top the ramp is still at radius 24.0773mm, outside the 24.05mm flange; pinned-engine containment probes found no rib material inside the flange radius anywhere in the grip band. The STL remains watertight and single-body, so CI misses the functional failure. Smallest fix: reject combinations without a positive full-depth grip band, or clamp/derive rib_leadin to leave one; regress this exact combination with a bite probe.

2. models/kidscleancar_knob_cover.scad:63,85,95-96,150-152,218-223 — the pry notch can sever ribs and emit orphan bodies. With flange_d=30, rib_count=12, pry_notch_w=24, and pry_notch_h=8, pinned OpenSCAD exits 0 but the STL has three components: the 1,998-triangle cover plus detached 14- and 12-triangle rib fragments. The accepted wide notch no longer fits between the adjacent ribs. Smallest fix: derive or reject notch dimensions against actual between-rib clearance so neither neighboring root can be cut; regress this exact combination expecting one connected body.

Default geometry is otherwise watertight, one body, correctly oriented, correctly licensed, and fully registered. Full evidence is recorded on bead pst-20w.

[SeanOC]

Codex gate review — changes requested

Reviewed exact head 8654605d7ef7048124d1d073cf89ac77b5ab496e. Three PR-introduced cross-parameter defects block the gate:

1. models/kidscleancar_knob_cover.scad:73,91-92,125-133,186-195 — accepted chamfer/roof values self-cross the shell and collapse export. With screw_head_h=8, top_chamfer=3, and wall=1.2, annulus_roof_z=13.15 but the chamfered exterior is only r=23.75 there while the inner contour is bore_r=24.35. Pinned OpenSCAD 2025.06.12.ai25773 logs ERROR: The given mesh is not closed! and emits only 72 triangles in six detached rib fragments. The same class occurs with knob_h=2 or screw_clear=6 alongside the thin-wall/max-chamfer pair. Smallest fix: add a cross-parameter guard ensuring the exterior radius at each inner roof stays strictly outside that inner radius (especially outer_at(annulus_roof_z) > bore_r), or reshape/clamp the chamfer; regress this exact triple natively.

2. models/kidscleancar_knob_cover.scad:67,76,87,135-148,204-216 — an accepted combination has zero flange bite. At flange_h=1.5, fit_clearance=1.5, rib_interference=0.8, the 2.2mm lead-in is still outside the flange radius when the 1.5mm grip band ends. Exact-head containment testing found zero rib material inside the flange across 5,625 samples of the whole band, even though the STL is watertight and one body. The cover therefore has no nominal retention. Smallest fix: clamp/derive rib_leadin from flange_h-rib_z0 so a positive full-depth bite band remains, or assert the incompatible relation; regress this combination with a bite probe.

3. models/kidscleancar_knob_cover.scad:66,86,96-97,150-152,229-240 — the accepted wide-notch corner still severs ribs. flange_d=34, rib_count=12, pry_notch_w=24, pry_notch_h=8 exports three components: the 1,990-triangle cover plus two detached 8-triangle rib scraps. This violates the single-body contract and the claim that the recess never eats a rib. Smallest fix: derive/assert notch dimensions against actual between-rib/root clearance (or reshape the cut) and regress this exact combination for one connected component.

The default model is otherwise watertight, one body, correctly oriented, correctly licensed, and fully registered. Full evidence is recorded on review bead pst-oi4.

[SeanOC]

Converged — domain narrowed to the measured knob class (9c9de82)

Per the mayor decision after 6 rounds: converge by shrinking scope, not hardening wider. This cover childproofs one knob class; it doesn't need to render a 30mm flange or a 1.5mm grip band.

Narrowed every @param to a tight envelope around the measured device (defaults unchanged):
flange_d 44–52, flange_h 2.5–4, knob_d 26–33, knob_h 4–8, screw_head_h 0–4, fit_clearance 0.2–0.6, knob_clear 0.8–2.5, knob_vclear 1–3, screw_clear 1–3.5, rib_count 4–8, rib_interference 0.1–0.5, rib_width 1–2.5, wall 2–3.5, top_chamfer 0–2, pry_notch_w 8–16, pry_notch_h 2–5, skirt_gap 0–1.

Kept the derivations that make each corner valid (chamfer clamped to wall; rib lead-in clamped so a full-depth grip band survives; top_z = max(knob_roof_z, annulus_roof_z) + wall; pry width clamped to the between-rib gap) and the explicit domain asserts with fix-hint messages for what the ranges still can't exclude — including the two you called out: thin-wall + max-chamfer roof self-cross, and flange_h-min + fit_clearance-max + rib_interference-max must leave ≥1.5mm rib/flange overlap (asserted explicitly).

Full corner matrix run locally on the pinned native OpenSCAD (2025.06.12.ai25773) before pushing — 8105 cases: exhaustive 2¹² over the interacting params + 4000 random full-parameter corners + all 9 reviewer-cited combos. Every case is either a valid export (closed mesh, single watertight body) or a loud assert: 8102 valid, 3 asserted (the incompatible knob-vs-flange combos), ZERO defects.

Defaults byte-identical (53.5×53.5×13.05, roof-on-bed support-free orientation, watertight single body, invariants green); wasm sweep green.

[SeanOC]

Codex gate review — changes requested

Reviewed exact head 9c9de8283c048ce916ed75e18fa9cfa9a0ecfaaa. One PR-introduced functional geometry defect blocks the gate:

1. models/kidscleancar_knob_cover.scad:145-157 — an accepted corner eliminates the insertion lead-in. With flange_h=2.5 and skirt_gap=1 (both declared bounds; all other values default), the last bound in rib_leadin is 2.5 - 1 - 1.5 = 0, so the ramp collapses to zero. The full rib tip begins directly on the mouth plane at radius 23.85mm, already 0.2mm inside the 24.05mm flange radius. Pinned-engine containment confirms solid rib material inside the flange envelope immediately at the mouth. The STL is still watertight and one component, so the sweep and topology checks remain green even though the flange meets a square full-interference shoulder that can snag or buckle instead of self-centering and seating. Smallest honest fix: constrain or reject the relation so every accepted combination preserves a useful positive lead-in while retaining the 1.5mm full-depth grip band (for example, guard a minimum lead-in tied to rib_interference and tighten skirt_gap accordingly), then regress flange_h=min, skirt_gap=max for nonzero lead-in plus a valid one-body export.

Default geometry and the focused prior-failure corners are otherwise watertight, single-body, correctly oriented, correctly licensed, and fully registered. Full evidence is recorded on review bead pst-yc4.

[SeanOC]

Lead-in guaranteed at every corner (e4aad97)

Fixes the pst-yc4 finding: at flange_h=2.5 + skirt_gap=1 the grip-band clamp had consumed the whole lead-in (rib_leadin → 0), leaving a square full-interference shoulder that snags instead of self-centering.

The rib now reserves room for both a positive insertion ramp and the full crush band:

rib_leadin_min = max(0.4, rib_interference)   // deeper rib → longer ramp
engage_min     = rib_bite(1.5) + rib_leadin_min
skirt_z0       = clamp(skirt_gap, flange_h - engage_min)

so every accepted corner keeps rib_leadin ≥ rib_leadin_min and grip_band ≥ 1.5mm. A new assert makes a collapsed lead-in fail loudly.

Verification: default byte-identical (skirt_z0 still 0.5, rib_leadin still 0.8 → same STL, 53.5×53.5×13.05, watertight, invariants green). Containment probes confirm flange_h=2.5, skirt_gap=1 (interference 0.2 and 0.5) now has an empty ramp region at the mouth plus full-depth bite above it. Re-ran the full corner matrix on pinned native OpenSCAD — 8105 cases (exhaustive 2¹² + 4000 random full-parameter corners + 9 reviewer combos): 8100 valid, 5 asserted, ZERO defects. flange_h=min + skirt_gap=max exports one watertight body with a nonzero lead-in and ≥1.5mm bite. wasm sweep green.

[SeanOC]

Codex gate review — changes requested

Reviewed exact head e4aad978c2cac3559c8fbfc8cb97a202ec735c70. One PR-introduced functional-clearance defect blocks the gate:

1. models/kidscleancar_knob_cover.scad:150-159 (especially line 155) — the ribs intrude into the promised screw-head pocket. rib_top_z = flange_h + 1 leaves each full-depth rib solid for 1mm above the flange. At defaults the nominal screw annulus is r=16.0..24.35mm above z=3.15mm, yet the ribs occupy r=23.85..24.65mm through z=4.15mm; pinned-engine containment confirms solid material at r=23.95, z=3.65 on every rib azimuth. Because the source measurements do not constrain screw radial position or head diameter, a stock head reaching that outer annulus can stop full seating or catch a cover intended to spin freely. The sidecar at models/kidscleancar_knob_cover.invariants.py:101-107 probes only the annulus midpoint and misses this intrusion. Smallest honest fix: end the inward rib at flange_h (or taper it back to the bore by that height) and regress the outer annulus at rib azimuths just above the flange. I verified the one-line rib_top_z=flange_h boundary remains watertight and single-body and clears the failing probes.

The focused accepted-corner native exports, default topology, print orientation, catalog/artifact set, license, vendored pins, and reported CI checks are otherwise clean. Full evidence is on review bead pst-s4l.

[SeanOC]

Ribs ended at the flange top (bcd15c8)

Fixes pst-s4l: rib_top_z = flange_h + 1 left each rib solid for 1mm above the flange, intruding into the screw-head annulus (solid at r=23.95, z=3.65 on every rib azimuth) where a stock head could stop seating or catch the free-spinning cover.

Fix: rib_top_z = flange_h — the rib grips only the flange edge and never rises into the annulus. Grip is unchanged: grip_band was already measured only up to flange_h; the removed 1mm sat above the flange and gripped nothing. The sidecar now also probes the above-flange annulus at rib azimuths (it previously sampled only the annulus midpoint and missed this).

Verified: the rib now clears r=23.95, z=3.65 on all azimuths while the flange-band bite (z=2.075) stays solid; STL watertight, single body, bbox unchanged 53.5×53.5×13.05, invariants green (incl. the new probe). Full corner matrix on pinned native OpenSCAD: 8105 cases → 8100 valid, 5 asserted, ZERO defects; wasm sweep green. Renders updated (ribs 1mm shorter; internal, bbox unchanged).

[SeanOC]

Codex gate review — pass

Reviewed exact head 6bf51d9e05ee7707fefea401109567879c78a415. No PR-introduced blocking defects found.

Verified the exact-SHA default export and invariant sidecar on pinned OpenSCAD 2025.06.12.ai25773, plus two high-risk combined parameter corners. All native exports were watertight single bodies; containment probes confirmed flange bite, between-rib free spin, and screw-annulus clearance above the flange. The source, catalog/sweep registration, tracked renders, orientation, and CC BY-NC-SA model licensing are consistent, and all reported CI checks are green.

Gate verdict: pass.