Event-Driven NLP Nowcasting
Generate fast event nowcasts from text streams and route through LLM risk filters.
Every strategy MooBot runs or plans to run. The machine-readable source is lib/v2/strategies/catalog.ts โ the operator-readable source of truth is STRATEGIES.md. Activity counts below come from live v2 tables.
5 tracks
Generate fast event nowcasts from text streams and route through LLM risk filters.
Use LLM probability estimation to find YES/NO mispricing vs market price. Generates both YES and NO signals.
Detect multi-outcome markets where probability sums deviate from 100% and trade the mispriced legs.
Fade sharp price moves when volume/speed suggests overreaction, targeting partial mean reversion.
Buy NO contracts when YES is overpriced or at momentum exhaustion levels with time remaining.
4 tracks
Capture basis dislocations between prediction and crypto venues under strict inventory limits.
Optimize quote placement and inventory with queue-reactive microstructure features.
Rank instruments with embedding-aware models for risk-adjusted long/short tilts.
Exploit crowd decomposition and maker/taker flow asymmetries with bounded downside.
Tracked separately โ Robinhood intents are keyed by strategy_version on robinhood_strategy_intents, not through strategy_registry.
Paste/forward takes become structured ideas in research_ideas. Not yet auto-promoted to signal_scores โ planned as operator_research_v1.
For the full per-strategy spec (state, venue, gates, ownership, and known gotchas), see the canonical STRATEGIES.md. Agents modifying any strategy must update that file in the same PR.