Runtime
Continuous or one-shot cycles, configurable interval, timezone-aware scheduling, blocked local hours, optional dry run with paper cash, environment kill switch.
Feature matrix
A straight shot through what the stack can do today—grouped the way you’d configure it in YAML and the GUI.
Strategy
Resolution window, min time-to-close, min confidence, edge floors, price bands, targets per cycle, adaptive relaxed pass, hourly P&L goal hooks, self-adaptation bias.
Risk
Daily loss vs start-of-day cash, per-cycle loss framing, spread and volume gates, book depth, per-market allocation cap, correlated event limits, risky-tier deploy ceiling.
Execution
Limit or market orders, time-in-force, post-only / maker-style joins, retries, pre-submit REST book refresh for top picks.
Models
Heuristic ranker or OpenAI / Gemini with temperature and token caps, optional external context blend, anti-anchoring prompt split for LLMs.
Consensus
Require heuristic second opinion on LLM paths, minimum primary edge after cross-venue adjustments—reduces single-model whipsaw.
Hard gates
Bundle-sum deviation reject, absolute max minutes-to-resolution enforced in scan, gates, and post-refresh—no late surprises.
Cross-venue
Optional Polymarket comparison by curated condition IDs—skip or downscore when venues materially disagree.
Tape-aware rank
Public Kalshi trades feed summarized into flow nudges—align / oppose boosts scaled by dominance on busy tickers, with enforce-goal depth mode.
Sizing
Fractional Kelly weights with min-edge floor and equal-split fallback, optional confidence-scaled deployment boost within caps.
Calibration
Resolved prediction buckets nudge minimum confidence when enough telemetry exists—auto-tune can adjust runtime deltas on a cadence.
Regimes
Optional local-hour windows override slices of strategy/risk—for example tighter spreads overnight or fewer concurrent bets intraday.
See it in narrative form
The about page translates this matrix into what actually happens minute by minute inside a cycle.