[dfattal]

What

David's panel feedback: with the 0.03 default the depth cursor floats too far in front of the content. Runtime ADR-046 Amendment 2 A.2 is halved to 0.015 eye baselines, about 1 mm of crossed on-screen disparity at a 63–65 mm IPD instead of about 2 mm. The runtime makes the same change in DisplayXR/displayxr-runtime#1965, so web and native cursors still agree.

• js/inline3d-cursor-depth.js: CURSOR_DEFAULT_TUNING.margin 0.03 → 0.015. This is the single source. DepthCursor (./three), cursor: 'depth' on ./model / ./splat (three and PlayCanvas), and the stereo-tile pair cursor js/camera/pair-cursor.js all take the margin from the core. A grep found no local copy left: every margin option is margin > 0 ? {...CURSOR_DEFAULT_TUNING, margin} : CURSOR_DEFAULT_TUNING. The only other 0.03 hits were in comments and JSDoc, and those are updated too.
• Types and docs: cursor-depth.d.ts, model.d.ts, splat.d.ts, the DepthCursor JSDoc, and the cursor section of docs/authoring-inline-3d.md.
• CHANGELOG.md: a new ## Unreleased entry marks this as a deliberate default change in both the core and preview tiers. To get the 1.41.0 look back, pass margin: 0.03.
• Test wording in the e2e harness is updated. The harness already reads the margin from CURSOR_DEFAULT_TUNING.

Evidence

• npm test: tsc passes, and 918 of 918 tests pass.
• Mutant: with the core margin restored to 0.03, 4 tests fail: the default-margin pin in cursor-depth.test.mjs, and the target, units and PairCursor tests in pair-cursor.test.mjs.
• The default-margin test now also checks the behaviour: cursorTarget(T, true, -0.2) === -0.215.
• npm run -s api:check: the snapshots match.
• npm run -s check:snippets: 22 files typecheck.
• npm run -s test:e2e:cursor on the real viewers in headless Chrome: PASS.
  • The model runs playcanvas/model, three/model, playcanvas/model/screen and three/model/world all give cursor −0.1688 with the cube front at −0.1538, which is front − 0.015 exactly.
  • Splat: playcanvas −0.0400, three −0.0410. In every run the off-subject cursor is null (nothing is drawn).

Not tested

• On a real panel: whether 0.015 reads right through the DisplayXR Browser weave needs an eyeball on the panel. Stock Chrome has no inline 3D.
• No release or tag is cut. The change ships with the next sdk-v*.

🤖 Generated with Claude Code