Benchmarks
The numbers behind the paper
Section 12 of the yellow paper cites these pages. Everything here comes from scripts in one package with fixed seeds, so you can run it and get the same results.
Timings for every primitive the protocol uses: BLAKE3, AES-256-GCM, Ed25519, ML-DSA-65, X25519, ML-KEM-768, the hybrid handshake, Merkle roots and ledger ID derivation.
tasqnetwork.io/benchmark/cryptoWhether Eq. 6 predicts quorum capture, what the distinct-machine rule buys, and the attack economics that led to hidden audits.
tasqnetwork.io/benchmark/quorumHow fast the reputation rule removes faulty nodes, and how much noise an honest node can have without being excluded.
tasqnetwork.io/benchmark/reputationA discrete-event simulation of mode R with hidden audits. Latency, utilisation and wrong-acceptance rate, random versus κ-weighted selection.
tasqnetwork.io/benchmark/schedulerA cost model for mode A against mode R, and a study of floating-point reordering.
tasqnetwork.io/benchmark/gpu-modelReproduce
Microbenchmarks ran on a 2 vCPU Intel Xeon(R) Processor @ 2.10GHz with one thread, Python 3.11.15, median of 7 runs. Timings depend on hardware. Simulations reproduce exactly.
Raw data and scripts
| File | Page | Download |
|---|---|---|
| micro_crypto.csv | crypto | CSV |
| S1_quorum_capture.csv | quorum | CSV |
| S2_distinct_machines.csv | quorum | CSV |
| S3_attack_economics.csv | quorum | CSV |
| S3b_breakeven.csv | quorum | CSV |
| S4_reputation.csv | reputation | CSV |
| sim_scheduler.csv | scheduler | CSV |
| M1_modeA_overhead.csv | gpu-model | CSV |
| M2_mode_costs.csv | gpu-model | CSV |
| M3_breakeven_premium.csv | gpu-model | CSV |
| D1_nondeterminism_cpu.csv | gpu-model | CSV |
| scripts/micro_crypto.py | source | View |
| scripts/sim_protocol.py | source | View |
| scripts/sim_scheduler.py | source | View |
| scripts/model_gpu.py | source | View |