Alembic V2 - Native growth cycle

Wet ground now starts its own little garden: fresh water supports plants, leaves and blooms return to ash, ash and water make mud, and exposed wet mud spends itself on a seed or a mite beside food. Crowded native mites pause reproduction; dry patches fade. Existing V2 physics, weather and ice tuning remain intact.

Main's explorer, sharper close-up grains, beetles and newts are preserved. This release fixes hovering after support removal, rejected-tool mutation of the explorer, invalid drop positions and oversized/diagonal pokes. The explorer remains session-only and is not saved in flasks.

• 452 tests pass: 247 V2, 168 original-game and 37 V3, plus lint, typecheck and production build.
• Three fixed-seed ten-minute shores ended with plants and mites and 86–96% of their starting water. Dry scenes decayed to ash. These are finite-window checks, not a permanent-equilibrium guarantee.
• Five alternating desktop A/B rounds pass the 10% median-step regression limit in all four scenes; worst increase 3.6%. Full measurements, tail timings, source hashes and reproduction commands are preserved in the evidence note.
• Independent review found no unresolved concrete defects.

Play V2 · Release log · Evidence and limits · Changes

This remains an alchemical game simulation. Finite water/food can run out, and exact checkpoint continuation is scoped to the same engine/runtime. The native Godot project and separate WebGPU experiment are not part of this release.

Source commit: 3cfde02aedb282c2e119bda2756bd88a06641165.

Alembic V2 - October 14, 2026

Play Alembic V2 · Full release log

Recent additions

• Safer particle collisions: fast grains check thin walls and world boundaries at cell resolution.
• Steadier simulation and tools: fixed 60 Hz ticks with bounded catch-up, elapsed-time fling response, and held tools/decay applied per simulation tick.
• Weather and living creatures: seeded warm spells, cold snaps and gusts; minnows gasp on land, and mites have age, hunger, drowning and freezing limits.
• More control: ¼× speed, pause/single-step and easier phone zoom.
• A visible release log: “what's new” under desktop More and phone Options, plus clearer ice help and six new freezing checks.

Why ice no longer freezes an entire warm pool

The October 13 heat-balance update deliberately removed runaway cooling. Ice still freezes already-cold water; cold snaps make that easier and can skin exposed water. Warm water can slowly melt ice. Use the cool wand to freeze water directly. Brine resists passive freezing. This release clarifies that behavior and tests it; it does not rebalance the thermal rules again.

Validation and limits

Six new fixed-seed checks cover calm and cold-snap contact freezing, warm-water resistance, brine resistance and the cooling wand. Disabling contact freezing in an isolated test copy fails both positive contact tests. All four deterministic fixture state hashes match the previous main revision. Three interleaved benchmark rounds found no scene above the project's 10% regression limit; no performance speedup is claimed.

Local release checks passed: 216 V2, 168 classic, and 37 V3 tests, plus lint, type checks and the production build. An independent review checked the thermal claims, test isolation and layout.

Heat, density and reactions remain intentional cellular game approximations, not a certified scientific solver. Exact checkpoint replay requires the same engine version/runtime. Physical phone performance testing remains pending. V2 still uses WebGL; the separate WebGPU experiment has not shipped.

The Godot V2 Windows playtest is a separate project, currently private, with its own release and save format.

Recent implementation: #27, #29, #30, #31.

Release log, clearer help and freezing regressions: #32. Exact release source: b68128aee6677c8c4da5515a8d378e43fb9d1065.