ERNA Weather API

Weighted multi-model ensemble + METAR pacing correction + LLM reasoning. This is the primary ERNA daily Tmax forecast.

Raw weighted ensemble used by the edge/arbitrage module (no pacing, no LLM). Values are stored in °C; _c/_f aliases are added.

We send POST with Content-Type: application/json and an 8-second timeout. Both http and https are supported (use https in production — http leaks your HMAC secret in transit). Failed deliveries retry with backoff (1m, 5m, 30m, 2h, 12h). After 10 consecutive failures the webhook is disabled automatically.

All strategy pick events share one envelope: event_id, event_type, event_version, created_at, attempt, market, trade and result. result is null on *.pick.created and filled on *.pick.resolved. Temperatures in *_c fields are always Celsius; bucket_low/bucket_high/bucket_label use the unit Polymarket uses for that city (market.bucket_unit).

{
  "event_id": "3f1c9e6e-2b8a-4a1f-9d55-5a2d0f8c1a11",
  "event_type": "no24.pick.created",
  "event_version": 1,
  "created_at": "2026-09-01T22:10:04.512Z",
  "attempt": 1,
  "market": {
    "city": "warsaw",
    "market_date": "2026-09-02",
    "market_tz": "Europe/Warsaw",
    "bucket_unit": "C",
    "event_slug": "highest-temperature-in-warsaw-on-september-2-2026",
    "polymarket_url": "https://polymarket.com/event/highest-temperature-in-warsaw-on-september-2-2026"
  },
  "trade": {
    "pick_id": 812,
    "strategy": "no24",
    "staking": "flat",
    "side": "NO",
    "model": "erna_pro",
    "model_exact_pct": 12.5,
    "predicted_max_c": 24.3,
    "bucket_slug": "24c",
    "bucket_label": "24°C",
    "bucket_low": 24,
    "bucket_high": 24,
    "no_ask": 0.72,
    "stake_usd": 10,
    "shares": 13.888889,
    "max_slippage_bps": 200
  },
  "result": null
}

The same NO pick can be delivered under three stake-sizing rules. Subscribe to whichever you want your bot to trade: no24.pick.created (flat $10), no24.pick.created.tiered ($5/$10/$15/$20 by the city's prior win rate) or no24.pick.created.kelly (half-Kelly, $5–$25). Win rates only use days strictly before the traded day. Envelope and fields are identical; only trade.staking, trade.stake_usd and trade.shares differ. When half-Kelly finds no edge (or the optimal stake falls under the $5 minimum), no Kelly event is sent at all — the flat event still fires.

{
  "event_type": "no24.pick.created.kelly",
  "trade": {
    "pick_id": 812,
    "strategy": "no24",
    "staking": "kelly",
    "side": "NO",
    "no_ask": 0.72,
    "stake_usd": 18,
    "shares": 25
  },
  "result": null
}

In webhook settings you can set a minimum historical ROI per variant (flat / tiered / kelly, default −50%). If a city's full-history ROI for that variant is below your threshold, the no24.pick.created* event for that city and variant is not delivered. The filter never applies tono24.pick.resolved*: settlement events are always delivered, so a bot that received an entry signal always learns how the position closed. Every NO24 payload also carries meta.no24_city_roi with the city's historical ROI per variant (null when there is no settled history yet), so your bot can apply its own filtering too.

{
  "event_type": "no24.pick.created.kelly",
  "trade": { "pick_id": 812, "strategy": "no24", "staking": "kelly", "side": "NO", "stake_usd": 18, "shares": 25 },
  "meta": {
    "no24_city_roi": { "flat": 0.0574, "tiered": 0.0697, "kelly": -0.32 }
  },
  "result": null
}

Same envelope; result carries the settlement. no24.pick.resolved.tiered and no24.pick.resolved.kelly repeat it with that variant's stake, payout and ROI. yescore.pick.resolved, ernamax.pick.resolved and taf.pick.resolved use the identical result block.

{
  "event_type": "no24.pick.resolved",
  "trade": { "pick_id": 812, "strategy": "no24", "staking": "flat", "side": "NO", "stake_usd": 10, "no_ask": 0.72 },
  "result": {
    "outcome": "win",
    "actual_tmax_c": 26.1,
    "payout_usd": 13.888889,
    "roi": 0.3889,
    "resolved_at": "2026-09-03T01:12:44.180Z"
  }
}

NO1 adds the safety fields it used to accept the trade: consensus_c, consensus_sources, distance_c, bucket_gap, top_yes_label, top_yes_price, taf_risk and metar_tmax_so_far_c. no1.pick.resolved uses the same result block as no24.pick.resolved.

{
  "event_id": "9a3d0e11-77bc-4c02-9f4a-6de2b1c7a005",
  "event_type": "no1.pick.created",
  "event_version": 1,
  "created_at": "2026-09-04T11:38:02.104Z",
  "attempt": 1,
  "market": {
    "city": "warsaw",
    "market_date": "2026-09-04",
    "market_tz": "Europe/Warsaw",
    "bucket_unit": "C",
    "event_slug": "highest-temperature-in-warsaw-on-september-4-2026",
    "polymarket_url": "https://polymarket.com/event/highest-temperature-in-warsaw-on-september-4-2026"
  },
  "trade": {
    "pick_id": 41,
    "strategy": "no1",
    "side": "NO",
    "bucket_slug": "12c",
    "bucket_label": "12°C",
    "bucket_low": 12,
    "bucket_high": 13,
    "no_ask": 0.99,
    "yes_at_entry": 0.01,
    "stake_usd": 5,
    "shares": 5.050505,
    "consensus_c": 21.4,
    "consensus_sources": 4,
    "distance_c": 8.4,
    "bucket_gap": 8,
    "top_yes_label": "21°C",
    "top_yes_price": 0.54,
    "taf_risk": "calm",
    "metar_tmax_so_far_c": 20.6,
    "max_slippage_bps": 100
  },
  "result": null
}

YES Strategy adds the per-city bias correction fields corrected_max_c, bias_c and edge.

{
  "event_type": "yescore.pick.created",
  "market": { "city": "madrid", "market_date": "2026-09-02", "market_tz": "Europe/Madrid", "bucket_unit": "C" },
  "trade": {
    "pick_id": 511,
    "strategy": "yescore",
    "side": "YES",
    "model": "standard",
    "model_exact_pct": 31.2,
    "predicted_max_c": 32.4,
    "corrected_max_c": 33.1,
    "bias_c": 0.7,
    "bucket_slug": "33c",
    "bucket_label": "33°C",
    "bucket_low": 33,
    "bucket_high": 33,
    "yes_ask": 0.34,
    "edge": 0.08,
    "stake_usd": 10,
    "shares": 29.411765,
    "max_slippage_bps": 200
  },
  "result": null
}

MAX Strategy is paper-traded (mode: "paper") and reports the local entry_hour (11 or 15) plus the Erna Max distribution used.

{
  "event_type": "ernamax.pick.created",
  "market": { "city": "toronto", "market_date": "2026-09-02", "market_tz": "America/Toronto", "bucket_unit": "C" },
  "trade": {
    "pick_id": 77,
    "strategy": "erna_max_11",
    "side": "YES",
    "mode": "paper",
    "entry_hour": 11,
    "mu_c": 25.8,
    "sigma_c": 1.1,
    "n_models": 6,
    "model_prob": 0.41,
    "bucket_slug": "26c",
    "bucket_label": "26°C",
    "bucket_low": 26,
    "bucket_high": 26,
    "yes_ask": 0.62,
    "edge": 0.09,
    "stake_usd": 10,
    "shares": 16.129032,
    "max_slippage_bps": 200
  },
  "result": null
}

TAF Strategy fires once per distinct TX value in the day's TAF bulletins. leg is the escalation step (1 = $5, 2 = $10, 3 = $20, …) and taf_issue_time/taf_tx_at identify the bulletin used.

{
  "event_type": "taf.pick.created",
  "market": { "city": "beijing", "market_date": "2026-09-02", "market_tz": "Asia/Shanghai", "bucket_unit": "C" },
  "trade": {
    "pick_id": 34,
    "strategy": "taf",
    "side": "YES",
    "leg": 2,
    "taf_issue_time": "2026-09-02T05:00:00+00:00",
    "taf_tx_c": 31,
    "taf_tx_at": "2026-09-02T07:00:00+00:00",
    "bucket_slug": "31c",
    "bucket_label": "31°C",
    "bucket_low": 31,
    "bucket_high": 31,
    "yes_ask": 0.28,
    "stake_usd": 10,
    "shares": 35.714286,
    "max_slippage_bps": 200
  },
  "result": null
}

Not a trade: this is the daily calibrated Erna Max forecast for a city, with the full bucket probability distribution. Use it to build your own sizing logic.

{
  "event_id": "9a4f77c1-0f2e-4c33-8f10-9d1a1c77e001",
  "event_type": "ernamax.forecast.created",
  "event_version": 1,
  "created_at": "2026-09-02T09:05:11.004Z",
  "attempt": 1,
  "market": {
    "city": "warsaw",
    "market_date": "2026-09-02",
    "market_tz": "Europe/Warsaw",
    "bucket_unit": "C"
  },
  "forecast": {
    "model": "erna_max",
    "version": "1.2",
    "mu_c": 24.6,
    "sigma_c": 1.05,
    "mu_models_c": 24.1,
    "pacing_weight": 0.45,
    "pacing_source": "metar",
    "n_models": 7,
    "top_bucket_slug": "25c",
    "top_bucket_label": "25°C",
    "top_bucket_prob": 0.38,
    "top_bucket_market_yes": 0.29,
    "top_bucket_edge": 0.09,
    "bucket_probs": [
      { "slug": "23c", "label": "23°C", "prob": 0.11 },
      { "slug": "24c", "label": "24°C", "prob": 0.29 },
      { "slug": "25c", "label": "25°C", "prob": 0.38 },
      { "slug": "26c", "label": "26°C", "prob": 0.17 }
    ],
    "reasoning": "METAR pacing above climatology; 30d bias +0.4C applied."
  }
}

Verify the HMAC over the raw request body (not the re-serialised JSON) before trusting the payload. Use a timing-safe comparison.

// Node.js / Express
import { createHmac, timingSafeEqual } from "crypto";

app.post("/erna-webhook", express.raw({ type: "application/json" }), (req, res) => {
  const sig = (req.headers["x-erna-signature"] || "").toString();
  const expected = "sha256=" + createHmac("sha256", process.env.ERNA_WEBHOOK_SECRET)
    .update(req.body).digest("hex");
  const a = Buffer.from(sig), b = Buffer.from(expected);
  if (a.length !== b.length || !timingSafeEqual(a, b)) return res.status(401).end();

  const event = JSON.parse(req.body.toString("utf8"));
  // idempotency: skip if you've already processed event.event_id
  handle(event);
  res.status(200).end();
});

# Python / Flask
import hmac, hashlib, os
from flask import request, abort

@app.post("/erna-webhook")
def erna_webhook():
    raw = request.get_data()  # bytes, NOT request.json
    sent = request.headers.get("X-Erna-Signature", "")
    expected = "sha256=" + hmac.new(
        os.environ["ERNA_WEBHOOK_SECRET"].encode(), raw, hashlib.sha256
    ).hexdigest()
    if not hmac.compare_digest(sent, expected):
        abort(401)
    event = request.get_json()
    # idempotency: dedupe on event["event_id"]
    handle(event)
    return "", 200