Try it
Type two messages from different people about the same thing and see what Quorum does with them — extraction, then a plain conflict judgment, live.
Runs the exact same pipeline shown on Architecture — noise gate, extraction, referent resolution, pre-rules, adjudication — not a simplified stand-in for it.
A single message almost never produces a contradiction here — this page only judges claims within the messages you type in this run, with no memory of the ledger's established facts, and only wording the resolver can match will bucket together (see the two prefilled examples below). Type two messages, from different people, about the same thing, in similar wording, for a real disagreement to show up.
How this page works (for reviewers)
Every run goes through the same runExtractionPipeline/runAdjudicationPipeline the ledger and Signals pages use — this is not a simplified demo path. In replay mode, novel input that doesn't match a committed recording's cache key returns a clear “no recording for this input” error rather than a fabricated result (see ReplayMissError in src/core/model/client.ts). Live mode calls the Vercel AI Gateway server-side only — the API key never reaches the browser — and is rate-limited to 10 calls per session per 10 minutes, with an automatic fallback to replay on a 429.
This page has no memory of the ledger — each run only compares claims found within that run's typed messages, never against previously established facts. Referent matching for free-typed text is also heuristic, not alias-aware: real corpus events like indep_event.launch_date resolve via a hardcoded alias list (src/core/aliases.ts), so unrelated free-typed wording (even about the same real-world topic) can extract into separate referents and never get compared at all — one message producing a single-claim bucket that resolves COMPATIBLE by the “fewer than 2 live claims” guard, not because nothing changed.
Full detail in the README: /architecture#sandbox