See the decoder make a real decision.
Each run samples a fresh syndrome from the locked local circuit, executes the decoder backend and returns measured BP iterations, beam expansions, correction and runtime. It is live software, not a prewritten animation.
One case. Every decision exposed.
The dashboard changes because each click creates a different syndrome. “BP work” is the measured work on that case; “Beam work” appears only if the engine escalates.
ORBIT-D1 · Connecting to local decoder…
Reading the evidence-backed product status from the local API.
Adaptive Decoder Studio
Real decoder API connected. Every run uses a fresh syndrome generated from the locked benchmark circuit.
- Code
- [[144,12,12]] bivariate bicycle
- Circuit
- memory X · 12 syndrome rounds
- Noise model
- circuit-level p = 0.001
- Controller policy
- safe fallback v1
- Actions available
- BP · Beam8/32/64 · memory
If BP converges, the controller stops. If BP does not converge, it sends the real syndrome to Beam8. The learned multi-action policy is disabled until held-out validation.
Choose a normal case to see fast BP, or a hard Beam case to make the service find a syndrome where BP genuinely needs escalation.
Compare all real actions on one hard syndrome
Runs BP, Beam8/32/64 and memory actions against the same syndrome. This is an engineering comparison, not a published benchmark.
The last eight real local controller requests in this browser session.