• Use venue

  result

  for settlement — never self-computed truth
• Use

  close_time

  (or settlement date from the ticker), not date-of-crawl
• Decision features cut at the city/event's own local decision time — no UTC mismatch
• Entry prices from a decision-time snapshot, not near-close prices (which have already converged)
• Walk the real ladder to your size; cap by top-of-book depth; deduplicate fills
• Fee-deducted PnL only — never notional
• Temporal holdout or expanding-window walk-forward only — no shuffled CV
• Forward paper-trading before live capital: log model p, decision-time ask, and outcome; compare to the backtest Brier/accuracy

1. Markets aggregate information. By decision time, the market price already reflects NWS model output, recent actuals, and whatever edge the sharp accounts have extracted. Your forecast must be measurably better than the market — not just accurate.
2. Brier score is not edge. A 0.07 OOS Brier score (better than climatology) is consistent with zero net edge if the market is priced at 0.07 too.
3. The actual edge is behavioral. The longshot overpricing is not informational — it's behavioral (retail overconfidence in cheap contracts). You capture it by being the liquidity provider on the tail, regardless of your forecast quality in those brackets.
4. Forecast-driven YES entries work only at the margin. When your calibrated model says a bracket is materially underpriced (> θ net edge after fees), buying YES is valid. But this is a small subset of decision points, and the fill rate on thin tails constrains capacity.

• Whelan, Makers and Takers: The Economics of the Kalshi Prediction Market (CEPR VoxEU; GWU/UCD working papers). 300k+ Kalshi contracts. Average pre-fee return ≈ −20%, concentrated in takers and longshot buyers. Makers earn positive returns. The foundational result: winners provide liquidity, losers take it.
• Large-N Polymarket maker-taker study (588M+ trades, SSRN). Top ~1% of accounts capture ~76.5% of profit, predominantly by resting limit orders. Confirms the maker edge generalizes across venues.
• Gupta, Who Profits in Binary Prediction Markets? Maker–Taker Dynamics, Behavioral Bias, and Sentiment Arbitrage on Kalshi (SSRN). Microstructure + behavioral-bias treatment of who wins and why.

• Classic betting-market literature (Thaler, Ziemba): longshots win less often than their price implies across most wagering markets.
• Kalshi-specific: ~$0.05 contract historically wins ~2%; sub-$0.10 contracts lose ~60% of stake to buyers. Favorites are fairly- to slightly-underpriced.
• The tradeable expression: NO-side maker bid on cheap brackets, diversified.

• Fee-inclusive rotating-CV backtest on Kalshi weather markets: ~+1.7–3.3% taker ROI / +7.9–9% maker ROI on the longshot fade.
• Realized hit-rate on brackets priced ~10%: ~3–7% → ~+6.7¢/contract gross; maker economics improve this further.

• That a good weather or event forecast beats the market at decision time (redundant with the price — see Lessons above).
• That retail-visible cross-venue or intra-venue arbitrage is repeatably profitable net of fees/geo/lockup.
• That the edge concentrates where your forecast is most accurate (it tracks market thinness/retail-ness instead).

Verify specific figures against primary sources before sizing — magnitudes vary by sample window and venue.

• references/strategy-catalog.md

  — Full strategy verdicts with evidence and catches

• references/sizing-and-edge-gates.md

  — Complete fee-aware sizing rules and gate derivations

• references/backtesting-methodology.md

  — Settle-on-venue-result rule, phantom-edge hall of fame, method checklist

• references/evidence-and-literature.md

  — Primary sources: Whelan, Polymarket 588M, Gupta, favorite-longshot magnitudes

• scripts/sizing.py

  —

  kalshi_fee

  ,

  net_edge

  ,

  theta_for_account

  ,

  limit_price_cents

  ,

  kelly_contracts

  ,

  slippage_ok

  — pure stdlib, runs offline