Otterchip Model E — independent final-geometry review
2026-11-06 · Claude Opus 5.5 (high) · data: review-final-geometry.json · earlier: initial, provisional visual
Revision r2 (2026-11-06) is a bounded factual correction pass. It changes the canonical scores, the neutral-hand terminology, the black-key bridge datum, the final grille evidence, the UI comparison evidence and the 3D-print gates. My independent judgments are otherwise unchanged.
Revision r3 adds the final optimizer results (optimization/results.json, generatedAt 05:03:35Z), which use the corrected nominal-profile path with no rounded stature override at factor 1. It also adds the music-only subgroup tradeoff and the ±8% screen-top values. My earlier 4-age analysis is approximate. Manufacturing is reviewed separately in review-manufacturing.html.
Revision r4 (about 10:30Z) aligns the final scores with results.json and the refreshed selected-check.json: E is 0.014797756266410417 and M is 0.052349737283120416. The readiness text now points to the build 9 final disposition.
Scope: rendered appearance, app-solver poses and the optimizer objective for the frozen compact E. The mechanical CAD/QA final disposition (build 9) is in review-manufacturing.html.
Verdict. No blocking geometry or visual error for the frozen compact E. Not production-validated. Build 9 export geometry checks pass (QA 2026-11-06T05:24:51Z; see the manufacturing disposition). Stage A coupon and fit files can be prepared for quotation and printing. The full shell is conditional on measured fit, coupons and the acoustic prototype. No physical sit, load, tip, thermal, electrical, acoustic or child-use testing and no FEA have been done. J values are designer heuristics on estimated body proportions. Do not present the M/E J ratio as a percentage improvement in health, comfort or fitness.
Final parameter identity
Frozen: width 355 · rear 92 · height 245 · tilt 18° · setback 180 · deck 62.5 · slope 0 · drop 23 · underlap 39 · pad 7 / crown 1 / gap 10 · round 4 · grille corner 4.
native-browser-final.json (04:53Z, errors [], 16 cases) carries exactly these params, including rear 92. E measures 355 × 338.0 × 415.4 mm, against M at 355 × 316.9 × 382.2.
optimization/native/selected-check.json also shows rear 92 and, in its refreshed form, J 0.014797756266410417, identical to the final results.json (core 1.1.0, nominal profile path). M is 0.052349737283120416. An earlier snapshot showed 0.0147936; the difference was only a tiny stature-rounding path. Zero hard penalty.
- Earlier grille and depth-grid artifacts used rear 100 or the provisional 250/20/159. They are treated as provisional evidence.
- This is a practical compact tradeoff, not a global optimum.
Images read
From /tmp/otterchip-e-final-:
- E-body and M-body;
- E 3/6/9/12 side views, typing and music (all 8);
- E-6 three-quarter views, typing and music;
- M-3 and M-12 side views, typing and music.
Also /tmp/otterchip-e-final-E-orthographic-front.png (frozen 245/18/180, rear 92). In it the grilles visibly share the display's top and bottom lines. It supersedes the provisional /tmp/otterchip-e-review-E-orthographic-front.png.
Raw metrics (app solver, auto furniture with the same algorithm; table heights within 2 mm)
| model | age · activity | view mm | gaze° | neck° | incid.° | top below eye mm | elbow° | wrist° | forearm° | shoulder° | table lift over neutral-hand mm† | horiz. shoulder→wrist mm | wrist disp./(U+F) |
| M | 3 typing | 380 | 14.2 | 0 | 3.7 | 12.7 | 93.9 | −8.5 | 27.7 | 31.6 | 112 | 202 | 0.377 |
| E | 3 typing | 386 | 10.3 | 0 | 7.7 | −8.7 | 95.5 | −2.0 | 19.5 | 25.0 | 83 | 193 | 0.291 |
| M | 3 music | 437 | 12.9 | 0 | 2.4 | 16.7 | 89.2 | 1.0 | 11.9 | 11.0 | 54 | 159 | 0.141 |
| E | 3 music | 439 | 9.5 | 0 | 8.5 | −5.7 | 88.9 | 1.0 | 6.6 | 5.5 | 30 | 145 | 0.073 |
| M | 6 typing | 461 | 20.3 | 0.3 | 9.8 | 78.9 | 98.3 | −3.0 | 20.3 | 28.6 | 111 | 250 | 0.327 |
| E | 6 typing | 441 | 17.7 | 0 | 0.3 | 55.9 | 90.6 | 1.0 | 15.1 | 15.7 | 84 | 211 | 0.192 |
| M | 6 music | 501 | 18.9 | 0 | 8.4 | 81.3 | 88.9 | 1.0 | 9.7 | 8.5 | 53 | 189 | 0.111 |
| E | 6 music | 500 | 15.9 | 0 | 2.1 | 58.9 | 88.8 | 0.5 | 5.5 | 4.3 | 29 | 176 | 0.059 |
| M | 9 typing | 486 | 24.8 | 4.8 | 14.3 | 123.6 | 89.0 | 1.0 | 18.9 | 17.9 | 120 | 254 | 0.226 |
| E | 9 typing | 491 | 22.1 | 2.1 | 4.1 | 106.8 | 89.0 | 1.0 | 14.7 | 13.7 | 91 | 241 | 0.174 |
| M | 9 music | 556 | 22.5 | 2.5 | 12.0 | 131.6 | 89.2 | 1.0 | 9.2 | 8.3 | 59 | 223 | 0.107 |
| E | 9 music | 560 | 19.9 | 0 | 1.9 | 113.0 | 89.2 | 0.5 | 5.6 | 4.8 | 35 | 209 | 0.063 |
| M | 12 typing | 533 | 28.8 | 8.8 | 18.3 | 176.4 | 89.0 | 1.0 | 16.7 | 15.7 | 118 | 282 | 0.198 |
| E | 12 typing | 536 | 26.2 | 6.2 | 8.2 | 159.0 | 89.1 | 1.0 | 12.9 | 12.0 | 90 | 268 | 0.153 |
| M | 12 music | 601 | 26.2 | 6.2 | 15.7 | 184.4 | 89.0 | 0.5 | 8.1 | 7.2 | 57 | 249 | 0.093 |
| E | 12 music | 602 | 23.9 | 3.9 | 5.9 | 166.1 | 89.0 | 0.5 | 4.6 | 3.6 | 32 | 233 | 0.050 |
- Horizontal shoulder→wrist is the raw quantity the root graph plots, in mm/cm, from the shoulder joint to the wrist centre.
- Wrist displacement ÷ (upper arm + forearm) is the optimizer's loss input. Its good threshold is 0.243.
- My earlier straight-line chain ratio is a separate descriptive metric. No graph change is requested.
- View distance, gaze, incidence and screen-top values use
device.screenCenter, device.screenNormal and solution.points.eyes. They match the native equivalence raw values (E age 3 typing: incidence 7.66°, view 386.1 mm). Horizontal reach is my recomputation from exported points.
- † Table lift over the neutral-hand construction (solver field
keyAboveNeutralMm) = auto table height − neutral-hand table height for this seat. The neutral-hand construction has the upper arm vertical, the forearm level, the wrist at 0°, the palm level, and the activity's curled fingertip drop down to the key target. The value is how much higher the table, and so the key target, sits than that construction. Auto furniture accepts this lift to keep 20 mm of leg clearance. It is not key height above the elbow joint. The raw forearm slope and wrist angle show the arm consequence more directly.
Real benefit and limitations (raw)
Benefit:
- Typing forearm rise reduced by 3.8–8.2°. Music forearm slope is +4.6…+6.6°.
- Table lift over the neutral-hand construction: typing 111–120 → 83–91 mm, music 53–59 → 29–35 mm.
- Shoulder flexion at age 6 typing: 28.6 → 15.7°.
- Age 3 typing wrist: −8.5 → −2.0°.
- Incidence at ages 6–12: 8–18° → 0.3–8°.
- Gaze at age 12 typing: 28.8 → 26.2°.
Limitations:
- Music-only subgroup tradeoff (final optimizer, nominal stature). E's music loss is higher than M's at ages 3–7, for example age 3 at 0.0242 vs 0.0041 and age 4 at 0.0086 vs 0.0010. The cause is mainly viewing: shallow gaze and the screen top above the eye. Raw music arm values still improve. E's aggregate music J is lower (0.0117 vs 0.0226) because ages 8–12 improve. The 70/30 aggregate (E 0.014797756 vs M 0.052349737) and typing improve at every age, but not every activity improves at every age.
- ±8% stature sensitivity (not percentiles): at age 3 and 0.92 stature, the screen top is about 37 mm (typing) and 35 mm (music) above the eye, with gaze about 6°.
- Typing still needs 83–91 mm of table lift over the neutral-hand construction to keep leg clearance, and typing forearms still rise 12.9–19.5°.
- The screen top is above the age-3 reference eye: 8.7 mm in typing and 5.7 mm in music (M is 12.7 and 16.7 mm below). Age-3 gaze is shallow at about 10°, and incidence there is worse than M. This is reported against OSHA's adult top-at-or-below-eye reference. It is not a child hard limit.
- Age-12 gaze of 26° is still above the adult 15–20° reference. A single fixed screen cannot centre both ends of the age range.
- All poses are static.
Age 3 vs 12 in E:
- Typing: wrist −2.0 / +1.0°, elbow 95.5 / 89.1°, view 386 / 536 mm.
- Music: wrist +1.0 / +0.5°, elbow 88.9 / 89.0°, view 439 / 602 mm.
Independent checks
- 70/30 arithmetic: verified. Activity weights are 0.7/0.3, renormalised so D/W get typing-only scores. Categories (view, elbow, wrist, reach) are 0.25 each. J = 0.5·mean + 0.25·±8% mean + 0.25·worst-age + hard. The canonical values come from the native cross-check,
native-final-equivalence.json: 113 checks with 0 failures, and browser geometry equals the optimizer adapter. Direct core.score on the captured E age-3 solution gives typing 0.02582332267241637 and music 0.024163303025379834. 0.7·typing + 0.3·music = 0.025325316778305407, exactly the selected-check age-3 value. My earlier hand recomputation (≈0.0261 / ≈0.0240) is approximate only and does not indicate any model discrepancy.
- What drives J. The wrist category is 0 in all 16 cases and elbow is ≤ 0.006, so half the category weight barely separates designs. J differences come from (a) age-3 typing wrist displacement (M 0.30 → E 0.038) and (b) viewing at the age extremes: better for E at age 12, worse at age 3. Forearm slope and table lift over the neutral-hand construction are not directly in J, so I report them raw.
- Weight sensitivity: superseded by the final optimizer. My earlier 4-age recomputation was approximate. The final per-age results (
optimization/results.json, 05:03:35Z) show E below M at every age 3–12 for the 70/30 aggregate and for typing. Music-only, E is worse than M at nominal ages 3–7 and better at 8–12. My earlier claim that E ranks ahead under every activity split is withdrawn for that subgroup.
- Hard and physical gates, reported separately.
- White-key target: bridge clearance is 40.0 mm (my check, consistent with the native 60 cases). This covers the representative white-key target only. It does not prove the whole playing area.
- Secondary black-key contact (mechanical corrected datum):
- The viewer reports 26.1 mm black-key exposure.
- In CAD, the flat black-key top begins at local z 15.6, the keybed centre is at 97.8 and the solid bridge/front lip is at 88.35, which gives 25.05 mm of actual exposure (gate 23). The earlier 28.55 mm figure was the keycap ramp envelope, not the flat playing top.
- The CAD lip extends 1.05 mm forward of the viewer bridge. The optimizer's black-target capsule clearance of 9.7 mm is therefore conservatively 8.65 mm, still above the 3 mm allowance.
- The bridge underside is 6.4 mm above the black-key tops (gate 3.4), and penetration is 0.
- These are static geometric checks, not a claim that every key fits every hand. Donor measurement and finger trials are pending.
The E-oriented physical check flags M's bridge gap at 3.36 mm against 3.4 mm; this is informational only, and M is unchanged.
- Comparison graphs (native evidence from
native-comparison-ui.json; I did not run the UI browser myself):
- Viewports: desktop 1440×1080 and mobile 390×844.
- Points: typing 40 (M, D, E, W), music 20 (M, E), weighted 20 (M, E).
- Default model M; the camera stays fixed through compare, person, age and activity changes; 0 errors.
- Data: all 32 weighted rows pass the arithmetic check, M and D raw typing metrics are identical, and D/W music is absent.
- Raw reach is labelled as horizontal. A duplicate 0.1 loss-axis tick is minor polish, not a blocker.
- Neighbourhood. All 26 one-variable ±1/±2 variants have zero hard flags, with J 0.0139–0.0187. Some neighbours score lower: pad rise −2 gives 0.0139, and the pad 14.5/6 candidate gives 0.0139. The frozen point is a practical rounded choice, not a local minimum, and the difference is small next to M's 0.0523.
- Display depth.
- Best grid result: 222/18/250, J 0.0107. Stage-2 best at setback 250: J 0.0092.
- Best with setback ≤180: 0.01486. Frozen 245/18/180: 0.01479 (screen-grid snapshot). The final full-solver J is 0.014797756266410417.
- Setback 250 would make the case about 485 mm deep: roughly 70 mm deeper than E and 103 mm deeper than M. E itself is +33 mm deep and +21 mm tall vs M.
- In J arithmetic, the compact choice gives up about 0.004 in exchange for about 70 mm less depth.
- Collisions vs projection. Side views show 2D overlap of fingers with the keybed front and of hands with the pad. The three-quarter views show no real intersection, and exported penetration is 0. At E age 3 typing, the wrist-to-pad gap is 0 mm: contact, not penetration, which fits resting during pauses but leaves no hover margin.
Resolved earlier concerns
- Incidence and correlated viewing: resolved. They form one view category with declared subweights.
- Discontinuities: mitigated. The chosen point was re-solved and its neighbourhood checked; continuous hip refinement was not done.
- Reach: the primary gate is resolved. Far-key and control reach with repositioning is still a secondary, pending check.
- Music bridge: resolved at geometry level. The white-key target clears by 40 mm. The secondary black-key contact has 25.05 mm CAD exposure and capsule clearance of at least 8.65 mm. This is not a whole-keyboard fit claim, and donor and finger trials are pending.
- Furniture: resolved (auto, same algorithm for every model).
- Component floors: partially resolved. The 62.5 mm deck is conservative, and the LPK25 reserve still needs measuring.
- Grille alignment: resolved by complete evidence on the frozen params, and no fix is requested.
grille-alignment-final.json was independently re-run on 245/18/180, rear 92. Screen and grille share Y min 167.19759085844805 mm and Y max 323.14232783536437 mm, with exactYAlignment: true. The final orthographic front image visibly confirms it. My provisional observation was a perspective artefact, and the earlier provisional-params grille-alignment.json is superseded.
- Side features: withdrawn. The upper side rectangle is the working volume wheel. Jacks and the power rocker are recessed low behind the keyboard region, and the generic USB rectangles are gone. The top strip is stereo microphones, with no camera.
- Labels: corrected. The graph reach is horizontal shoulder→wrist. The optimizer's robustness is ±8% stature sensitivity, not percentiles.
Beauty at final proportions
- E reads as a clean, fluid fixed body. The compact 245/18/180 is visibly less monumental than the provisional 250/20/159, and the upper mass now balances over the lower deck.
- The front of the music bay is uncluttered and cleaner than M's front ports.
- The broad cream sweep is the dominant surface. With a clean shell preferred, it reads as intentional negative space.
- The volume wheel, the recessed side plates, the mic strip and the flush 7 mm pad are all discreet.
Required appearance changes: none. Optional, with no functional loss:
- carry the round-4 corner family to the side-plate and pull-tab pockets in CAD;
- use a low-gloss or matte finish on the large sweep. Big printed surfaces show layer lines, seams and fingerprints, so confirm the achievable finish with the chosen print process and vendor.
No grooves, logos or light bars are needed. Any change to the key or screen geometry needs objective re-evaluation.
Chart notes
- Native UI evidence shows the graphs meet the earlier requirements: typing for all 4 models, 70/30 only for M and E, D/W music absent, and horizontal reach labelled.
- Wherever the ambiguous UI metric
keyAboveNeutralMm is shown, rename or explain it as “table lift over neutral-hand construction (for leg clearance)”. Prefer forearm slope and wrist angle for readers.
- Label J as a heuristic and show no percent improvement. Label ±8% as sensitivity, not percentiles.
Blockers and gates
Blockers (geometry/visual): none.
Done, mathematical only:
- App-solver poses with zero hard flags (60 optimizer cases; 16 rechecked by me).
- Native equivalence: 113 checks, 0 failures.
- White-key-target bridge check (40 mm). Secondary black-key contact: 25.05 mm CAD exposure, capsule clearance ≥ 8.65 mm. Penetration 0.
- One-variable neighbourhood check.
- Exact final grille alignment.
- Native comparison UI and data audit.
Pending, CAD and manufacturing (separate final CAD review):
- Done in build 9: CAD and export checks pass (QA 05:24:51Z; native audit 746 checks, 0 failures). All known manufacturing items are resolved; see review-manufacturing.html. The stale supervisor metadata wrappers are root packaging metadata only.
- 3D-print production checks:
- material suitability for a child product, and heat deflection near the electronics;
- wall, rib and boss thickness;
- print orientation and support marks;
- layer-line strength at load paths and clips;
- dimensional tolerance of the print process and vendor;
- finishing plan.
- Donor measurements: LPK25 reserve and frame, flat black-key top and keybed datums, controller and ribbon, GK64X stack, AMOLED mount.
- Fit of the side jacks, rocker and volume wheel, including plug body and cable bend above the table.
Pending, physical validation (not done):
- Child sitting and leaning loads on the deck, pad and cover.
- Tip stability: E is taller and deeper than M, with a tall screen mass.
- Drop and impact, small parts, pinch points and edges, and materials under the applicable product standards (to be determined).
- Thermal, electrical safety and EMC.
- Acoustics: speakers, grille open area and microphones.
- Real child fit trials with actual movement.
Known limitations
- Body proportions are estimated and stature is the CDC midpoint, so no clinical optimum is claimed.
- My earlier per-activity score recomputation is approximate. The canonical scores come from the native
core.score cross-check.
- The comparison UI checks are native evidence; I did not run them myself.
- I read 17 images, and only age 6 has three-quarter views.
- This is not a CAD review.
Final export metadata reconciliation
Native release check, 2026-11-06T10:28:35.496803+00:00: the supervisor assembly manifest and export audit have now been refreshed against final build 9. All 83 published assembly mesh hashes match; both wrappers reference the final QA hash. This closes the metadata-only pending item described above. Detailed manufacturer reference shapes are omitted from the public assembly; fabricated parts are unchanged. No physical validation is implied.
Exact hashes and timestamps