How a solve flows through Selva

Debounced before it becomes a solve — 150 ms for a slider drag, 400 ms for typing.

One solve runs at a time; a newer value replaces the single waiting slot and the old one is dropped.

Inside the run, before the request: inputs already solved this session are replayed from memory.

Sends the input specs, the values, and a reference to the definition rather than the file.

Three doors before the work starts: too big, who are you, and how often are you asking.

Four quick lookups turn the id into a solvable job. The file itself never moves.

Which Rhino machine solves this — and its connection is usually already open.

The values are translated into the typed tree Grasshopper actually reads.

Identical solves arriving together run once and share the one result.

A hash of definition + input tree; a hit returns instantly without touching Rhino.

As many solves running at once as the machine has workers; the rest wait their turn.

A definition Rhino already holds is sent as a short id instead of the whole file.

Headless Grasshopper runs the definition — with its own definition and result caches.

JSON, gzip above 1 KB, and a Server-Timing header describing every phase.

The library page — not a package — decodes the JSON and matches outputs to the schema.

Meshes straight out of a binary blob into three.js — no other package is involved yet.

The Svelte viewer owns the solve session and the scene. The two packages before it only hand back data.