Aegis Reliability Spec v0
=========================
An open, versioned methodology for measuring whether an x402 endpoint actually
delivers what it charges for. Published so the method is auditable.

WHY: Every other x402 trust signal is passive — on-chain volume (a service is
used), self-reported metadata, or user feedback. None answer the buyer's real
question: will this endpoint deliver the goods, at the price it advertised, if my
agent pays right now? Aegis answers it with active, adversarial verification.

WHAT AEGIS MEASURES (per probe):
  1. Liveness — does it respond at all?
  2. Challenge validity — well-formed x402 402 (scheme/network/price)?
  3. Price-on-wire honesty — live price matches advertised?
  4. Delivery-on-payment (spot-check) — Aegis actually pays on a budget-capped
     schedule and records whether usable content came back. Cannot be faked.
  5. Latency & longevity.

TRUST TIERS:
  trusted     high score + verified delivery-on-payment + sustained longevity
  verified    live + valid challenge + price-honest, recently
  provisional seen, not yet sufficiently probed (unchecked != bad)
  flagged     failed a check (dead / invalid / price mismatch / non-delivery)

PROOF-OF-DELIVERY (auditable), recorded per spot-check:
  price_usd, http, latency_ms; body_sha = FULL sha256 of the response;
  tx = on-chain settlement hash (decoded from the x402 payment-response header);
  delivered = 1/0/null (null = called without required params = unverifiable,
  NOT a failure); self = 1 for borisinc.com endpoints (disclosed publicly).
Anyone can re-pay the endpoint and compare body_sha.

PROVENANCE & TAMPER-EVIDENCE:
  A canonical daily snapshot of the full registry is appended to an append-only
  sha256 hash-chain (each entry commits to the previous via prev_sha256).
  Altering any past day breaks the chain forward. Daily roots are designed to be
  anchored on-chain for external proof. See /provenance to audit it live.

PRINCIPLES: open methodology (a rater nobody can audit is not a rater); neutral
(own endpoints are probed and disclosed self=1); falsifiable (every claim is a
re-runnable probe with a recorded hash/tx).

VERSIONING: v0 is the initial public draft; tier/schema breaking changes bump the
major version, recorded in each snapshot going forward.

← Aegis · /provenance · /spec · /reliability · neutrality