# PayBench calibration provenance — R11 MPP-on-Lightning settlement-finality
# Created 2026-06-05 with first-party n=30 (REAL-RAIL — supersedes the 2026-06-04 doc-proxy 0.8s estimate).
# NOTE: the prior doc-proxy version of this file mis-labelled MPP as "Lightning Multi-Path Payments" —
# corrected here: MPP = Machine Payments Protocol (mpp.dev), the SAME protocol as R10, Lightning settlement.

rail: "R11 MPP-on-Lightning (Machine Payments Protocol, via Lightspark/Spark)"
dimension: settlement-finality
unit: seconds

distribution:
  family: lognormal
  parameters: {median_s: 16.52, sigma_log: 0.156}   # EMPIRICAL n=30, real-rail Spark hosted regtest
  approach: first-party-empirical

sources:
  - type: first-party
    ref: "poc/rail-lightning-mpp measure-finality.ts, n=30 Spark hosted regtest; samples/R11-mpp-lightning.samples.jsonl"
    date: "2026-06-05"
    sample_size: 30
    note: >
      0 failures. Real-rail MPP (Machine Payments Protocol) on Lightning via the
      @buildonspark/lightning-mpp-sdk (extends mppx). request->200 wall-clock for the full L402-family
      flow: 402 + fresh BOLT11 invoice (seller Spark wallet) + Lightning payment + preimage proof + 200.
      median 16.52s, sigma_log 0.156, min 8.45, p50 16.75, p90 18.33, p99 23.66.
  - type: documentation
    ref: "mpp.dev/payment-methods/lightning; Lightning HTLC resolution semantics"
    date: "2026-06-05"
    sample_size: 0
    note: "Raw Lightning HTLC resolution is sub-second; the measured ~16.5s is dominated by the Spark layer, not HTLC."

calibration_date: "2026-06-05"
confidence: "high (central, Spark-path); spread is quiet-regtest and likely understates mainnet"
fixture_content_hash: "sha256:b31f79451b67c68ffc97dbbb4846a497cd7f530385a99ed8decb421da9e18910"
notes: >
  REAL-RAIL — supersedes the doc-proxy 0.8s / sigma 0.80 MEDIUM estimate. CRITICAL ATTRIBUTION: this
  ~16.5s is the MPP-on-Lightning-VIA-SPARK number, NOT raw Lightning HTLC speed. Each payment is a
  multi-round-trip to Spark's hosted operators — FROST threshold-signing ceremony + SSP preimage swap
  (/spark.SparkService/initiate_preimage_swap_v3) — and that ceremony, not Lightning HTLC resolution
  (which is sub-second), dominates wall-clock. Puts R11 in the same finality band as R9 Solana-finalized
  (14.63s), an entirely different regime from what "Lightning = instant" assumes. A raw-LND L402 path
  (cf. R3 Polar regtest) would settle the Lightning leg far faster; R11 benchmarks the Spark production
  path specifically. Measured on Spark's HOSTED regtest (not a local node); mainnet operator latency may
  differ. sigma_log 0.156 is quiet-regtest and likely understates mainnet congestion — widen toward a
  doc-informed spread until a mainnet run is captured (same caution as R9). MPP = Machine Payments
  Protocol (same protocol envelope as R10, different settlement rail); the SDK README confirms this is
  NOT Lightning Multi-Path Payments. One transient preimage-swap failure was seen only under concurrent
  duplicate loops (operator error, since fixed); the clean single-process n=30 had 0 failures.