Skip to content

SLO.md

Latest commit

 

History

History
95 lines (73 loc) · 5.86 KB

SLO.md

File metadata and controls

95 lines (73 loc) · 5.86 KB

PEAC Protocol SLO

Status: Operator-facing service-level objectives for the open-source reference verifier and the published @peac/protocol package. Numbers are baseline measurements against a documented machine profile; see Benchmark methodology. These are SLOs, not SLAs. The open-source reference verifier carries no contractual SLA. The managed Hosted Verify instance is operated separately under its own contract and is not part of this repository.

Baseline metadata

Field Value
baseline_commit 0c6d2047
baseline_date 2026-05-21
baseline_machine_profile See Benchmark methodology
Capture method PEAC_BENCH_JSON=<path> pnpm exec vitest run tests/perf/wire02-slo.test.ts

The local protocol operations table below carries captured values for issue() and verifyLocal() measured against the baseline_commit on the documented machine profile. The reference verifier and MCP round-trip tables remain unmeasured this cycle; those rows are outside the current measured baseline and require dedicated benchmarks. Real-world performance varies with hardware, network, workload, and record size.

Local protocol operations (@peac/protocol v0.14.4)

Operation Median (p50) p95 p99 Notes
issue() (Ed25519, ~1 KB record) 0.4 ms 0.5 ms 0.6 ms Measured on the Wire 0.2 issuance path used by issue(); Ed25519 only.
verifyLocal() (Ed25519, ~1 KB record) 1.5 ms 1.7 ms 1.9 ms Local validation; no network.

The existing SLO gate in tests/perf/wire02-slo.test.ts asserts verifyLocal p95 MUST be ≤ 10 ms. That gate runs in CI on every PR.

Reference verifier (apps/api /v1/verify)

Operation Median (p50) p95 p99 Notes
/v1/verify (Ed25519, ~1 KB record) Not measured this cycle Not measured this cycle Not measured this cycle Loopback; no JWKS resolution
/v1/verify with JWKS resolution (single issuer, warm cache) Not measured this cycle Not measured this cycle Not measured this cycle Measured after first request; JWKS cache hot

Reference verifier rows are outside the current measured baseline and require dedicated benchmarks against the same machine profile.

MCP tool-call round-trip with record issuance

Operation Median (p50) p95 p99 Notes
MCP tools/call round-trip emitting _meta.org.peacprotocol/receipt_jws Not measured this cycle Not measured this cycle Not measured this cycle Local stdio transport

MCP round-trip row is outside the current measured baseline and requires a dedicated benchmark against the same machine profile.

Record size classes used in the table

  • Class A: ~1 KB record (typical agent observation; used in the rows above).
  • Class B: ~8 KB record (commerce evidence with attached upstream artifact).
  • Class C: ~64 KB record (per-extension-group kernel constraint ceiling; see Kernel constraints spec).

The published numbers reflect Class A unless a row explicitly states otherwise.

What these SLOs do not promise

  • The open-source reference verifier carries no uptime SLA. Self-hosters operate the verifier under their own availability commitments.
  • Hosted Verify (managed, multi-tenant, SLA) is a separate offering operated outside this repository. See Hosted Verify contract.
  • Baseline numbers describe loopback measurement on the documented machine profile. Production performance varies with hardware, network, workload, and record size.
  • Carrier-specific capture overhead is not measured this cycle. The local signing and verification cost for records emitted by those carriers is represented by the issue() and verifyLocal() rows above.

Regression policy

  • verifyLocal() p95 is gated by the existing CI benchmark regression check in tests/perf/wire02-slo.test.ts.
  • issue() baseline is published here. It is not a blocking regression gate until a dedicated gate is added and documented.
  • Reference verifier /v1/verify baseline is not published in this cycle. It enters regression policy only after a measured baseline is captured against the documented machine profile.

Related documents