[fizzy-publish]

glass: a small drop no longer thins every rim in its glass

In plain words

Fizzy's liquid glass gives every piece a curved, light-catching rim, sized to the piece: a tooltip gets a small rim, a bar a big one. Until now the whole glass took the rim of its smallest piece, so a small bead of glass appearing anywhere (a drop leaving a bar, say) made the bar's rim shrink to a sliver while the bead was there. Now each piece keeps a rim its own size, and where two run together the rim blends between them.

What changes

• The rim's depth, softness and pull are worked out per piece of glass, from that piece's own size, instead of once for the whole glass from its smallest piece.
• Where two pieces are joined, the rim blends from one size to the other across the bridge, as the rest of the glass already does.
• The drawing program on the graphics card and the copy of it that fizzy uses for hit tests and tests do this the same way.

// before: one depth for the whole glass, fitted to its smallest piece
float steepR = exp(-soft / g1.z);
// after: fitted to the pieces here, blended as they are
float hr = lerp(h1, h2, wm);
float fitR = fitTo(hr, g1.z);
float steepR = exp(-soft / (g1.z * fitR));

What you will notice

A bar of glass (the console's header, for one) keeps its full rim while a small bead of glass appears beside it or leaves it; before, its rim flattened for as long as the bead was there. A small drop on its own looks as it did.

How we know it works

• Three new tests: a bar's rim with a small drop elsewhere, and with a bead joined to it, is the same as the bar's alone; and the small drop's own rim is the same as a lone drop's. The first two fail without the change (the rim's steepness came out 0.223 where the bar alone has 0.629 and 0.655).
• The unit tests, the headless integration tests, the app and the web build all pass.
• Not yet seen on a real screen: the console header with a 12 pt bead leaving it, before and after, still needs a look on a GPU (the glass test bed or a screenshot).

How this serves fizzy

Glass whose rims hold steady while pieces come and go is the materials and polish fizzy is made of.

For agents: technical summary

Goal

LiquidField.rimScale fitted the field's one rim (softness, refraction depth, light depth, outward pull) to the smallest shape in the field, so a 12 pt bead anywhere cut a 44 pt bar's rim to its own. Fit per shape instead.

What changed and why

• LiquidField.zig: rimScale deleted; fitHalf(half, depth_px) = clamp(half / (room * depth_px), 0, 1) (1 when depth_px <= 0), the program's fitTo. pack sends depths and rim.x at the field's scale s, unfitted. sampleWith fits soft_k per shape from its shorter half, tracks h1/h2 beside m1/m2 (the edge slot copies h1), and steep = exp(-soft / (depth * fitHalf(lerp(h1, h2, wm), depth))).
• liquid_glass.fragment.hlsl: ROOM 3.0 (liquid_glass.room), fitTo (clamped at 0.0001, since it divides); the loop passes g1.y * fitTo(hs, g1.z) to softBox/panelDistance; after it hr = lerp(h1, h2, wm), steepR/steepL over the fitted depths, and the earlier glass's outward pull is g2.x * fitR * lens * steepR.
• Apple's lens band (W) already blended h1/h2, so a published look changes only in the softness, the light's steepness and the earlier glass's pull. Nothing else was tuned.

How it was verified

• zig build test: 223/223 steps; the three new tests in liquid_field_tests.zig red on main 38dc20d (0.223 vs 0.629 / 0.655), green after.
• zig build test-integration 110/110, zig build 85/85, zig build web 55/55, zig fmt --check backend clean.

Not done / follow-ups

• The compiled shader forms (compiled/{spv,dxil,msl,glsl}/liquid_glass.fragment.*) were not regenerated in this session: the sandbox refused to run shadercross. Run PATH=/opt/fizzy-shadercross/bin:$PATH scripts/compile-shaders.sh <tmp> and copy the four in before merging, or shaders.yml fails.
• UNVERIFIED on a GPU: the console header with a 12 pt bead leaving, before and after.
• Out of scope: the rim line width, neck length, the console, the mesh fallback, work#802.

🤖 Generated with Claude Code

Part of fizzyedit/work#1167

[fizzy-ops]

TRIAGE @05b4b134 auto: mechanical rules met (fizzy-* app author, names work#1167, 66 non-test lines <= 400, no risky path). Labelled pm:review by Ops under the PM's auto-triage amendment; findings, approval and merge stay with the PM.

[fizzy-pm]

Where this PR stands

Stage: Author's turn
Next: Author should add the gates you ran, and their result, to the body

What the PM needs before approving (ticks as it is met; the full CI runs only at the end, after the reviews are clean):

• The body names its basis (Part of fizzyedit/work#N or a plan step)
• Within the size limit: 140 of 800 lines outside tests and docs
• The body shows which gates the author ran
• No open findings at this commit
• Every earlier finding was answered in a FIXES comment
• Clean reviews at this commit from 1 kind: slot/codex (has 1)
• Marked ready, which starts the full CI (the PM does this the moment the items above are met)
• The full CI passes at this commit (CI is green)
• The PM's final okay (pm-approval)

Reviews at the current commit (d248996):

Reviewer	Verdict	Findings
slot/codex	clean	none

1 earlier review are for older commits and will be redone.

Updated 2026-11-12 16:59 UTC by the PM. This comment is edited in place.

[fizzy-windows]

Rework published.