S0 · Shadow benchmark of a partner's own execution (no keys, no spend)

{
  "protocol_id": "S0-shadow-execution",
  "version": 1,
  "experiment": "S0",
  "status": "draft",
  "title": "S0 · Shadow benchmark of a partner's own execution (no keys, no spend)",
  "hypothesis": "A partner's current guard, exit and launch transactions have a measurable landing rate, Δslot tail, fail rate and cost per landed transaction. Nothing is compared and nothing is sent: this is the baseline every later head-to-head (A2, B, C) is judged against.",
  "customer_value": "The partner learns how its own execution performs today, from chain data it can re-check, without giving Rout keys or traffic. Rout learns the workload it would have to beat.",
  "unit": "One partner intent: a transaction the partner decided to send, identified by the partner with the on-chain transaction that triggered it (or, failing that, its decision time) and every signature it sent for it.",
  "arms": [
    {
      "arm_id": "partner_path",
      "description": "The partner's existing sending path, exactly as it runs today. Route labels are the partner's own and stay partner-confidential."
    }
  ],
  "primary_metric": {
    "id": "landed_within_window",
    "definition": "Share of all partner intents with a landed_ok copy at slot - anchor.slot <= landing_window_slots. anchor = the slot of the trigger transaction the partner names (anchor.kind = trigger_tx); intents with only a decision time are reported separately, never pooled."
  },
  "secondary_metrics": [
    {
      "id": "delta_slot_quantiles",
      "definition": "Δslot quantiles over landed intents with order-statistic intervals."
    },
    {
      "id": "fail_classes",
      "definition": "Every intent by outcome class: landed_ok, landed_err (with the on-chain error), not_landed."
    },
    {
      "id": "double_landed_intents",
      "definition": "Intents with more than one landed copy: executions the partner did twice."
    },
    {
      "id": "spend_per_landed",
      "definition": "Chain-observed fees, plus tips where the partner names its tip accounts, divided by landed intents."
    },
    {
      "id": "route_win_share",
      "definition": "For intents sent through several routes, which route's copy landed (partner-confidential)."
    }
  ],
  "parameters": {
    "confidence": 0.95,
    "quantiles": [
      0.5,
      0.9,
      0.95,
      0.99
    ],
    "bootstrap_resamples": null,
    "landing_window_slots": null,
    "commitment": "finalized",
    "window_start": null,
    "window_end": null,
    "intent_kinds": null,
    "min_intents": null,
    "strata": [
      "intent_kind",
      "congestion_tercile",
      "leader_window_position",
      "route"
    ]
  },
  "randomization": {
    "unit": "none (observational)",
    "method_id": "none",
    "method": "No assignment. The partner fixes the time window and the intent kinds before sharing, and shares every intent in it, failures included.",
    "strata": []
  },
  "denominator": "Every intent the partner made in the agreed window and kinds. A gap in the partner's own sequence numbers is reported as missing data, never filled.",
  "exclusions": [
    "None after the fact. Intents the partner cannot tie to signatures are counted as 'unattributed', not dropped."
  ],
  "stopping_rule": "Single analysis of the agreed window.",
  "acceptance": {},
  "redact_by_tier": {
    "partner_confidential": [],
    "public": [
      "mint",
      "size_raw",
      "quote",
      "endpoint_region",
      "refusal_code",
      "send_error",
      "route",
      "host_id",
      "nonce_account",
      "stratum"
    ]
  },
  "requires": [
    "PARTNER-SIGNATURES",
    "SHADOW-IMPORTER",
    "LICENCE"
  ]
}