The Decide console

The real engine, compiled to WASM, running in your tab — zero ML, zero network, zero server. Ask a question, pick candidates, read the decision.

Engine
loading…
WASM
loading…
Network
measuring…
Decide

Ask a question. Pick candidates. Read the decision.

Try an example:

Try a decision

Plain text is sent as-is. JSON pasted here is also sent as text — the engine reads text.

Criteria are composed into the question text — the wire format carries a boolean as text, and the zero-ML rungs read text. The JSON view shows the exact composition.

Same engine, same confidence gate, same refusal codes as the console below — the built request lands there verbatim.
Advanced — raw JSON editor, graph mode, refusal codes

Anything you type stays in this tab — there is no server to send it to. Invalid JSON or invalid requests return the same canonical error codes the HTTP API returns.

Refusal codes — every way this engine says no
ir.empty_candidates
Request carried an empty candidates set — there is nothing to choose between.
schema.*
The request did not match the canonical IR schema — the family covers invalid_json, missing_field, invalid_type, invalid_value, limit_exceeded.
graph.cycle
The declarative DAG contains a cycle; run it after breaking the loop.
engine.missing_backend
Graph needs a rung 5–7 backend (embedding / model / verifier) this tab does not ship.
below the gate (no code)
Not a refusal — when calibrated confidence lands under the gate the answer ships as an HTTP 200 abstain, not an error.
loading

Booting the WASM engine…

Graph mode — declarative DAG

A graph chains steps — normalize, decide, gate, output — in one request. Validation checks shape and acyclicity locally; running feeds it through the same WASM engine.

What this is — and what it is not

  • Real: the WASM module is built from this repo's opencodifier-wasm crate — the identical IR, rule engine, narrowing, lexical scorer, cache, and calibration code the CLI and HTTP server run.
  • Real: the zero-ML stack only. Rungs 5–7 (embedding, decision model, verifier) need backends this tab deliberately does not ship; a graph that needs one returns engine.missing_backend instead of pretending.
  • Not here: network calls, telemetry, or model downloads. The Network chip counts cross-origin requests from the Performance API — if it ever reads anything but zero, the page is broken, and the chip will say so.

Prefer a terminal? The same decisions from cargo run --bin opencodifier-cli -- decide, or serve the HTTP API and hit the documented endpoints.