Stop edge rounding at 6 %, where the Pixel 10a's coins still settle

[drehtuer]

Measures the edge-rounding range decision 98 left open, and narrows it from 1.5–12 % to 1.5–6 %. The new cap applies to the face designer's slider and to a set file's edge_rounding alike.

Based on #389 (feature/sample-dice-sets), stacked as .claude/CLAUDE.md asks.

Measured on the Pixel 10a, 2026-10-05

Fairness: FairnessTest, every catalogue shape, 20,000 throws each.

Rounding	Coin throws that never settled (bar 2)	d18 worst face (bound 1 %)	Other shapes
1.5 %	0	0.50 %	fair
6 %	0	0.62 %	fair
8 %	2, on the bar	0.68 %	fair, but the d10's χ² is 24.6 against 27.9
12 %	42, fails	0.98 %	fair

Settling: tools/harness.sh -n 200 -c 20 on d4, d6 and d20 at 1.5 %, 3 % and 12 %.

• No throw gave up, none was forced to settle, and no die came to rest on another.
• Median settle times are unchanged.
• Rounder dice tumble a little longer and are re-thrown less.
• The overlap and re-throw bars fail at every value, exactly as they do at the default; those are the known Step 5.4/5.5 items.
• The harness throws no coins, which is why it saw nothing wrong anywhere up to 12 %.

What gives first is the coin: rounded enough, its rim rolls like a wheel until the twelve-second backstop. 6 % is twice the default and the last value with a margin.

Device suites, same session: 131 tests across app, designer, dicesets/install, feature/roll, render/filament and simulation/jolt. 0 failed. 3 declined: the opt-in gallery and harness tests, which skip themselves unless asked. That run includes the designer branch's AtlasTextureTest, with 5 tests passing, its first time on a device.

Change

• DieMaterial.EdgeRoundingRange = 0.015..0.06. It is the single source for the validator's clamp, the designer's slider (now 10 positions), the harness's --rounding and FairnessTest's edgeRounding.
• Tests: those that pinned 12 % as the top now use 6 %. HarnessRoundingTest now also refuses 0.12.
• Docs:
  • docs/physics-and-rendering.md ("How round a die may be") has the tables above, replacing "not yet measured".
  • Decision 100 records the bound; 94 and 98 point to it.
  • Also updated: docs/dice-sets.md, docs/face-designer.md (range, and the visibility table, whose top values halve), docs/build-setup.md, the prototype's slider (max="6"), design/README.md, docs/STATUS.md and docs/TODO.md. The measurement box is gone, and a 100,000-throw follow-up with coins in the harness is left.

Checks (devcontainer)

• Unit tests pass in core/model, designer, feature/designer, render/filament, simulation/api, simulation/jolt, simulation/harness, dicesets/format, dicesets/install and app.
• ktlint and detekt pass on all modules; Android Lint passes on designer and feature/designer.
• simulation/jolt's device-test sources compile.
• verifyDocsIndex, verifyDocsLinks and verifyDesignSystem pass.
• The Documentation job's mermaid and markdownlint checks pass.

Note

A set already installed with edge_rounding above 0.06 now loads clamped, with the validator's usual warning.

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud