Changelog

New 12 programmatic Kalshi market category guides

One landing page per Kalshi market category — weather, sports, crypto, financials, politics, economics, culture — plus high-value sub-pages for temperature, rain, NBA, NFL, and Bitcoin. Each page includes a long-form strategy explainer, settlement rules, where the edge actually is, a live current-markets table, and a category-specific FAQ. Browse the index.

Also this week

• FAQ page with 20 Q&As on Kalshi, prediction markets, and the copilot's behavior. Read the FAQ.
• How it works page — long-form walkthrough of AI analysis, cross-platform arbitrage, NO-side pricing, and backtesting. Read it.
• Dedicated pricing page at /pricing with full transparency on the $14/month metered model.

Fix Stop endorsing losing strategy drafts

When the AI drafts a strategy and the projected backtest comes back losing, the artifact now carries a top-level _health: "losing" flag and a warning string. The system prompt now forbids "this is the strategy I'm building" framing on losing drafts — the AI must acknowledge the loss directly and offer concrete iteration paths instead. The strategy card also renders an inline red warning banner so the verdict stands on its own.

Fix Volume-prioritized candle backfill

Backtester candle data coverage was stuck at 4.2% because the daily ingest was processing markets chronologically — burning the rate-limit budget on freshest tiny markets before reaching high-volume ones the strategies actually trade against. Now ordered by volume_24h DESC so the markets a backtester cares about get filled first. Coverage of the tradeable universe should climb significantly over the next 24-72 hours.

Fix Whitelist keyword asymmetry

The live scanner checked keyword whitelists against both market title AND ticker; the backtester only checked the title. So AI-drafted strategies with whitelistKeywords: ["nba", "nhl"] worked live (matched KXNBAGAME-* tickers) but returned 0 trades in backtest. Both call sites now use the same logic.

New Decision-first cards with A-F grades

Every analyze_market result now leads with a letter-grade verdict (A through F), suggested position size, expected value, max gain, max loss, and the win rate source. The AI's bridge sentence quotes the grade + the most load-bearing single number, in that order. Less judgmental than BUY/PASS, more honest about confidence calibration.

Data 90-day rolling retention

Historical market data (market_candles, settled_markets) now follows a 90-day rolling retention window instead of keep-all. Trade-off: backtests beyond 90 days are no longer possible, but storage and query performance scale linearly with active trading activity rather than with all-time history. User account data is never pruned.

New AI accuracy scoreboard

Every BUY-verdict recommendation the copilot has emitted is now logged and scored against the actual market settlement. Ask "how accurate are you?" or "show me your calibration" and the copilot returns a card with hit rate vs predicted win rate, by category, with sample size. Calibration honesty is the feature — we don't hide losses.

Content Strategy guide library

Published nine educational guides covering the major Kalshi trading strategies: favorite-longshot bias, liquidity-first, spread capture, weather markets, fast turnover, event-driven, mean reversion, correlation arbitrage, and broad scanning. Each is research-backed with cited sources.

New Product pivot: scanner + copilot

Pivoted from "trading bot" to AI scanner + copilot. The platform no longer places trades automatically — every order requires the user's explicit tap. Backend engine and risk-management code remain intact for future opt-in automated execution, but the default product surface is decision support, not automation.