docs/research/2026-07-08-forecastex-probe-findings.md

ForecastEx read-only probe — findings (#2216)

ADDENDUM 2026-07-10 — the follow-up 2026-07-10-forecastex-uhlga-settlement-and-klga-fit.md confirms the LGA station call from primary sources (CFTC U-contract terms + the venue's own product list) and adds two facts this probe couldn't see: (1) the U-series settles on Weather Underground (≡ round(max METAR tmpf), measured 14/14) — not the NWS CLI; (2) the full board incl. settlements is public at forecastex.com/api/download, so the Observe leg does not need the EC entitlement (orders still do). The KLGA re-fit this note called for is done and committed (weather-oracle-params-klga.json, 30% OOS gain, no measurable ≥100 ceiling).

Date: 2026-07-08 · Account: DUR193395 (IBKR paper) · Method: live read-only CPAPI calls via experiments/forecastex_probe.js (OAuth1, local-owner creds). No order code was run.

TL;DR

ForecastEx is reachable from this account via /iserver/secdef/search, and it lists NYC

  • Chicago/Denver/Miami temperature contracts and CPI/Unemployment/GDP/Fed-Funds econ

contracts. But the headline premise of the port is wrong: ForecastEx NYC daily-high settles on LaGuardia (symbol UHLGA), not Central Park / KNYC. Kalshi KXHIGHNY settles on Central Park. They are different underlyings, so the "same contract, cheaper fee" and "identical-settlement arb" framings do not hold as written.

The four gating questions

1. Are ForecastEx contracts listed via secdef/search, and what shape? — YES.

  • Only surface with name: true (search by economic NAME, not ticker — the old

KNYC/HIGHNY guesses returned stocks). Matches carry exchange FORECASTX.

  • Underlying = secType IND, type: "Event", hasOptions: true, price increment

0.01 (1¢ tick). The tradeable buckets are secType EC (Event Contract).

  • NYC daily high: symbol UHLGA, conid 853400786, "New York City Daily

Temperature High".FORECASTX underlyings resolved across the searched names.

2. Fill depth at the ~1¢ spread? — NOT RETRIEVABLE FROM THIS ACCOUNT — entitlement gap (RESOLVED 2026-07-08). The blocker is NOT the EC month parameter. Diagnosis:

  • /iserver/secdef/info?conid=…&secType=EC&month=… returns a persistent HTTP 503 (3/3

retries, every month format) — the earlier "status 0" was just thes client timeout truncating the 503.

  • /iserver/contract/853400786/info-and-rules → has_related_contracts: false.
  • /iserver/marketdata/snapshot on the underlying returns no bid/ask/last fields.
  • /portfolio/accounts → tradingType: "STKNOPT" (DEMO, IB-CAN): this account is

entitled to Stocks + Options only, not ForecastEx event contracts.

Together these are conclusive: DUR193395 lacks ForecastEx (EC) entitlement, so IBKR won't serve the EC contract ladder or its market data from this session — no month string fixes that. The 1¢ tick is confirmed (from the underlying trsrv/secdef increment rule); per-bucket depth needs a ForecastEx-permissioned account. Note ForecastEx is a US CFTC-regulated venue; an IB-CAN demo may not be eligible — an eligible US-entity account with the ForecastEx permission enabled is the prerequisite to enumerate the ladder + depth.

3. Does the account carry ForecastEx / EC trading permission? — NO (confirmed 2026-07-08). /portfolio/accounts reports tradingType: "STKNOPT" (Stocks + Options only) for DUR193395, and every EC contract-detail endpoint returns a persistentwith no market data — search visibility of the contracts does not imply trade/data entitlement. A ForecastEx-permissioned account is required before the port can proceed.

4. Do the strikes line up with the oracle's bucket ladder? — CANNOT CONFIRM (blocked by #2, no EC ladder retrieved) — and moot until the settlement-station issue is resolved, because a KLGA ladder is not what the KNYC-calibrated oracle expects.

The decisive finding: settlement station mismatch

The symbols encode the settlement station: Chicago UHMDW (Midway), Miami UHMIA, and NYC UHLGA — LaGuardia. NYC has only UHLGA/ULLGA (high/low); there is no Central Park / KNYC ForecastEx series. Kalshi KXHIGHNY settles on Central Park (KNYC).

Consequences for the epic:

  • #2221 (cross-venue arb) premise fails. Kalshi (KNYC) and ForecastEx (KLGA) settle on

different stations — LGA typically runs warmer than Central Park by a systematic 1–3°F. A position spanning them is not market-neutral; it carries KLGA↔KNYC basis risk. The monitor code stands, but "identical KNYC settlement" must be struck; reframe as a basis monitor, not an arb.

  • #2217 (port) needs a re-fit, not a fee swap. The oracle's fitted params (coolBias

−1.43, σ) are calibrated to KNYC (kalshi-mos.js station=KNYC). Trading them on a KLGA-settled contract is a train/serve station skew. ForecastEx NYC is a new market that needs its own KLGA fit (its own MOS station + settlement pairs) before any edge is claimed.

  • #2220 (multi-city) is corroborated and needs station codes. Chicago/Denver/Miami all

list — but as airport-settled series (MDW/…); the city registry's station must match the ForecastEx settlement, per venue.

Net

The probe did its job: it de-risked the port before any order code by surfacing that ForecastEx NYC ≠ Kalshi NYC. Recommended next steps, in order: (a) read the ForecastEx contract spec to confirm KLGA settlement in writing; (b) resolve the EC month param to get the ladder + depth; (c) confirm EC trading permission; (d) only then decide whether to fit a separate KLGA oracle. The 1¢ rail is real and attractive — but on a different underlying than we thought.