One HTTP API, keyed and versioned

• GET/v1/strategiesEvery strategy, with its immutable version+

  Returns the strategies the platform can run, each with the version that produced its record, its lifecycle (production or research-only), the timeframes it supports and its execution mode. Every strategy is in paper mode: live order routing is locked.

  examplecopy

  {
    "data": [
      {
        "id": "brain",
        "name": "Decision Brain",
        "version": "1.0.0",
        "lifecycle": "production",
        "mode": "paper",
        "timeframes": ["15m", "1h", "4h"],
        "markets": ["binance_usdm:perpetual"]
      }
    ],
    "live_routing": "locked"
  }

• POST/v1/strategies/:id/promoteAsk for paper or live — live is refused+

  Needs the control scope. Paper is the only mode available, so a request for paper succeeds without changing anything, and a request for live is refused while live order routing is locked.

  examplecopy

  { "mode": "live" }
  
  → 409 { "error": { "code": "live_routing_locked",
                     "message": "Live order routing is locked on this platform; every strategy runs in paper mode." } }
  → 200 { "id": "brain", "mode": "paper", "changed": false }   // for { "mode": "paper" }

• GET/v1/decisionsEvery evaluation, including the rejections+

  Newest first. Filter by verdict (accepted or rejected), symbol and since (an ISO date). The rejections are the point: they are the only place you can see what the system nearly did, and why it did not. Pages of up to 200; pass next_cursor back as cursor for the next page.

  examplecopy

  GET /v1/decisions?verdict=rejected&since=2026-09-01&limit=50
  
  {
    "data": [
      {
        "id": "dec_41",
        "ts": "2026-09-24T09:15:00+00:00",
        "symbol": "XRPUSDT",
        "strategy": "Decision Brain",
        "verdict": "rejected",
        "quality_score": 48,
        "blocked_by": "quality",
        "reason": "Quality score 48 is below the minimum of 60.",
        "rules_failed": ["min_quality_score"],
        "executed": false
      }
    ],
    "next_cursor": "dec_41"
  }

• GET/v1/decisions/:id/replayNot available yet+

  Decisions are stored with their scores, rules and reason, but not the full market inputs they were made from, so they cannot be re-run. The endpoint answers 501 until those inputs are kept.

  examplecopy

  → 501 { "error": { "code": "not_available", "message": "Decisions are stored with their scores and rules, not the full market inputs they were made from, so they cannot be re-run yet." } }

• GET/v1/positionsOpen paper positions with mark and R multiple+

  Every open paper position, with the latest observed mark, unrealised P&L and R multiple. Stops and targets are managed by the engine on every candle — they are not orders held at an exchange, and the response says so.

  examplecopy

  {
    "data": [
      {
        "id": "7c1e…", "instance_id": "eb1a2ce8", "symbol": "XRPUSDT",
        "side": "long", "size": 1250.0, "entry": 0.5412, "mark": 0.5498,
        "r_multiple": 0.86, "unrealized_pnl": 10.75, "mode": "paper",
        "protective": { "stop": 0.5312, "target": 0.5712, "managed_by": "engine" }
      }
    ]
  }

• POST/v1/positions/:id/closeClose a paper position at the current mark+

  Needs the control scope. Closes at a fresh observed price and never at a guessed one — if there is no recent mark the request is refused with no_mark. The close is written to the audit log under the key's name.

  examplecopy

  { "reason": "manual flatten before travel" }
  
  → 200 { "id": "7c1e…", "status": "closed", "close": { "exit": 0.5498, "pnl": 10.75 } }
  → 409 { "error": { "code": "no_mark", ... } }

• POST/v1/backtestsQueue a backtest of one strategy+

  Needs the control scope. Runs a built-in strategy over the Binance candles cached on the server (300 to 1,500 bars). If the server has no data for that symbol and timeframe, the backtest fails with that reason rather than reporting an empty result. Parameter sweeps are not available through the API yet.

  examplecopy

  { "strategy": "brain", "symbol": "BTCUSDT", "timeframe": "15m", "bars": 1000 }
  
  → 202 { "id": "bt_2f77a1c9e0b4", "status": "queued" }

• GET/v1/backtests/:idResults, gross and net of costs+

  Gross performance and performance after modelled fees, spread and slippage are reported side by side, with the difference in R. A strategy whose edge disappears once costs are applied shows as exactly that. Results are kept while the server runs; only the key that queued a backtest can read it.

  examplecopy

  {
    "id": "bt_2f77a1c9e0b4",
    "status": "complete",
    "result": {
      "gross": { "trades": 38, "win_rate": 42.1, "expectancy_r": 0.12, "net_r": 4.56 },
      "net":   { "trades": 38, "win_rate": 39.5, "expectancy_r": -0.05, "net_r": -1.9 },
      "costs": { "cost_pct_per_side": 0.0006, "net_r_drag": 6.46 }
    }
  }

{
  "id": "evt_dec_41",
  "type": "decision.rejected",
  "occurred_at": "2026-09-24T09:18:00Z",
  "sequence": 41,
  "idempotency_key": "dec_41:rejected",
  "data": { "...": "the same object GET /v1/decisions/dec_41 returns" }
}

Delivery is at-least-once, so handlers must be idempotent — the idempotency_key is stable across retries. Any answer other than 2xx is retried after 30 seconds, doubling up to an hour, for 24 hours. Every attempt, with the status code your endpoint returned, is listed under the endpoint in the dashboard.