Reproduce a result

GeoSpec results are reproducible: the same model bytes, claims and engine version give the same result bytes. That makes a result something you can store, diff and check again later.

Keep what the result depends on

• The model bytes. The subject's content hash is the SHA-256 of these bytes.
• The request bytes. The plan, its claims and their payloads, encoded as JSON.
• The engine version. Exact byte equality holds within one engine profile, not across different numerical backends or versions.

Compare two runs

Hash the raw response bytes rather than a re-serialized object, so nothing in your own code can reorder them:

import { createHash } from 'node:crypto';

const raw = engine.processRequest(request);
console.log(createHash('sha256').update(raw).digest('hex'));

Two runs of the tutorial on the same engine print the same digest.

Canonical JSON

Plans and results use canonical JSON: UTF-16 key ordering and finite binary64 number rendering. The canonicalize export applies the same encoding to any JSON bytes:

import { canonicalize, initialize } from 'geospec-engine';

await initialize();
const bytes = canonicalize(new TextEncoder().encode('{"b":1,"a":2}'));
console.log(new TextDecoder().decode(bytes)); // {"a":2,"b":1}

canonicalize encodes JSON. It does not normalize a claim's meaning: two payloads that say the same thing differently stay different.