{
  "schemaVersion": "dxrg-agentic-trading-benchmark-card/v1",
  "version": "1.1.0",
  "title": "DXRG Agentic Trading Benchmark Card",
  "publishedDate": "2026-07-18",
  "modifiedDate": "2026-07-20",
  "reuseTerms": "No public license is granted by this file. Contact poof@dxrg.ai for reuse terms.",
  "purpose": "A disclosure and evaluation template for autonomous and semi-autonomous market systems. Completing the card does not establish profitability, safety, or suitability.",
  "completionInstructions": "Replace each null value with the disclosure for one evaluated configuration. Keep the field definitions unchanged or publish a new schema version.",
  "evidenceClassEnum": [
    "conceptual",
    "backtest",
    "event-replay",
    "simulation",
    "paper",
    "shadow",
    "bounded-historical-live",
    "current-live-observation"
  ],
  "fields": [
    {
      "id": "evaluation_subject",
      "label": "Evaluation subject",
      "required": true,
      "description": "Versioned model, harness, prompts, tool schemas, memory and retrieval policy, risk code, execution code, deployment configuration, asset universe, venue, dates, and starting portfolio."
    },
    {
      "id": "decision_rights",
      "label": "Decision rights",
      "required": true,
      "description": "Decisions the system makes autonomously, proposes for approval, cannot make, or delegates to deterministic controls, including fund-transfer authority."
    },
    {
      "id": "information_set",
      "label": "Information set",
      "required": true,
      "description": "Sources and timestamps available at each decision, point-in-time controls, survivorship handling, revision handling, and tests for model-memory or historical contamination."
    },
    {
      "id": "action_space",
      "label": "Action space",
      "required": true,
      "description": "Permitted actions and parameters; abstention; tool errors; validation, repair, retry, and human-override behavior."
    },
    {
      "id": "execution",
      "label": "Execution",
      "required": true,
      "description": "Information, inference, validation, submission, and fill timestamps; order handling; commissions, spread, slippage, impact, gas, borrow, funding, failures, and partial fills."
    },
    {
      "id": "baselines",
      "label": "Baselines",
      "required": true,
      "description": "Same-information, same-universe, same-cost comparators and ablations, including hold, simple rules, traditional models, and agent-component removals where relevant."
    },
    {
      "id": "attribution",
      "label": "Attribution",
      "required": true,
      "description": "Instruction-to-outcome trace plus decomposition of return, exposure, turnover, concentration, drawdown, tail loss, and execution effects."
    },
    {
      "id": "repeated_runs",
      "label": "Repeated runs",
      "required": true,
      "description": "Independent run count, seeds, distribution and worst case, action stability, prompt and version sensitivity, regime splits, and untouched holdout."
    },
    {
      "id": "safety",
      "label": "Safety and controls",
      "required": true,
      "description": "Identity, permissions, allowlists, mandate-compilation cases, state-integrity checks, exposure and loss limits, audit logs, pause, kill, recovery, and credential rotation."
    },
    {
      "id": "evidence_class",
      "label": "Evidence class",
      "required": true,
      "description": "The exact support class: conceptual, backtest, event replay, simulation, paper, shadow, bounded historical live, or current live observation."
    },
    {
      "id": "artifacts",
      "label": "Artifacts",
      "required": true,
      "description": "Immutable commit, benchmark version, prompt and configuration hashes, data lineage, fill model, run outputs, trace schema, representative traces, and audit references."
    },
    {
      "id": "limitations",
      "label": "Limitations and corrections",
      "required": true,
      "description": "Unresolved confounds, excluded data, failed runs, scope limits, conflicts, privacy constraints, correction route, and revision history."
    }
  ],
  "values": {
    "evaluation_subject": null,
    "decision_rights": null,
    "information_set": null,
    "action_space": null,
    "execution": null,
    "baselines": null,
    "attribution": null,
    "repeated_runs": null,
    "safety": null,
    "evidence_class": null,
    "artifacts": null,
    "limitations": null
  }
}