/edge-hint-extractor

Hints Schema

Use this schema as input for downstream concept synthesis and auto detection.

generated_at_utc: "2026-02-22T12:00:00+00:00"
as_of: "2026-02-20"
meta:
  rule_hints: 6
  llm_hints: 3
  total_hints: 9
  regime: RiskOn
hints:
  - title: "Breadth-supported breakout regime"
    observation: "Risk-on regime with pct_above_ma50=0.65"
    hypothesis_type: "breakout"              # optional
    preferred_entry_family: "pivot_breakout"  # optional
    symbols: ["NVDA", "AVGO"]              # optional
    regime_bias: "RiskOn"                    # optional
    mechanism_tag: "behavior"                # optional

Field Notes

• hypothesis_type: optional; if present should be one of: breakout, earnings_drift, news_reaction, futures_trigger, calendar_anomaly, panic_reversal, regime_shift, sector_x_stock. Unrecognized values are first checked against keyword inference from the hint's title and observation; if no match is found, they fall back to research_hypothesis. Used for clustering when --promote-hints is enabled.
• preferred_entry_family: optional; if present must be pivot_breakout or gap_up_continuation.
• symbols: optional focus list. Empty means broad market hint.
• regime_bias: optional regime gate (RiskOn, Neutral, RiskOff).
• mechanism_tag: optional mechanism label (behavior, flow, structure, etc.).

LLM CLI Contract

build_hints.py --llm-ideas-cmd "<command>" sends JSON to stdin:

{
  "as_of": "YYYY-MM-DD",
  "market_summary": {...},
  "anomalies": [...],
  "news_reactions": [...],
  "instruction": "Generate concise edge hints ..."
}

The command must print either:

• [{...}, {...}]
• {"hints": [{...}, {...}]}