Fintech benchmark — measured, not modeled
All numbers below are parsed live from embedded bench files in the substrate. Source paths shown per table.
Card-authorization budget
Real credit-card authorization cycles must complete in < 200 ms
end-to-end (network + issuer + scheme + acquirer). Anything attesting inside that flow
must be a fraction of that budget. We report 3 tiers below — be honest about which one applies.
Tier 1 · Signing-only micro-bench
loading…
These are primitive timings (struct hash, keccak digest, secp256k1 sign, verify) on a single core. They do not represent end-to-end card auth — they are the irreducible signature work inside a capsule.
Tier 2 · Fastpath sweep (Linux, in-process)
loading…
Concurrent fastpath calls against the in-process attester. c=1 is the meaningful number for single card-auth latency. c=128 stresses contention and shows where saturation pushes p99 past the 200 ms budget — flagged in red.
Tier 2 · Fastpath sweep (Windows, in-process)
loading…
Provenance
Source files embedded in the API binary at build time:
- data/bench/signing_only.txt
- data/bench/sweep_summary_linux.txt
- data/bench/sweep_summary_windows.txt
Reproduction: build the kernel from anima_matrix/kerneld, run
go test -bench=. ./fintech/.... Numbers here are not aspirational — they are
whatever the substrate just measured.