Sampled AI-compute validation
Price the proof before you trust the output.
Estimate the workload created when selected AI tasks are independently replayed by three validators and cross-checked. Results are assumption-bound, not a protocol audit.
Consensus pathThree independent replays
Replica 1replay
Replica 2replay
Replica 3replay
LOAD A WORKLOAD TO CALCULATE
GPU overhead--
Incremental cost--
Consensus miss--
Detection if validators are independent--
Run the planner to see workload, cost, and the assumption that governs the three-way agreement result.
Formula: samples = tasks x rate; replays = samples x 3; unanimous-invalid = q^3.
What this plans
Selected tasks, three validator replays, and the incremental GPU burden of cross-validation.
Selected tasks, three validator replays, and the incremental GPU burden of cross-validation.
What it cannot prove
No live Crynux data, protocol implementation, validator independence, price feed, or security guarantee is claimed here.
No live Crynux data, protocol implementation, validator independence, price feed, or security guarantee is claimed here.