- tradingagents/dataflows/universe.py: single source of truth for ticker
universe; all scanners now call load_universe(config) instead of
duplicating the 3-level fallback chain with hardcoded "data/tickers.txt"
- scripts/prefetch_ohlcv.py: nightly script using existing ohlcv_cache.py
incremental logic; first run downloads 1y history, subsequent runs append
only new trading days
- .github/workflows/prefetch.yml: runs at 01:00 UTC daily, before all other
workflows; commits updated parquet to repo
- Updated 6 scanners: minervini, high_52w_breakout, ml_signal, options_flow,
sector_rotation, technical_breakout — removed duplicate DEFAULT_TICKER_FILE
constants and _load_tickers_from_file() functions
- minervini, high_52w_breakout, technical_breakout: replace yf.download()
with download_ohlcv_cached() — reads from prefetched cache instead of
hitting yfinance at discovery time
- default_config.py: added discovery.ohlcv_cache_dir config key
- data/ohlcv_cache/: initial 1y backfill (588 tickers, 5.4MB parquet)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously the scanner stopped as soon as self.limit candidates were found
from as_completed() futures. Since futures complete in non-deterministic
network-latency order, this was equivalent to random sampling — fast-to-
respond tickers won regardless of how strong their options signal was.
Fix: collect all candidates from the full universe, then sort by options_score
(unusual strike count weighted 1.5x for calls to favor bullish flow) before
applying the limit. The top-N strongest signals are now always returned.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add GitHub Actions workflow for daily discovery (8:30 AM ET, weekdays)
- Add headless run_daily_discovery.py script for scheduling
- Expand options_flow scanner to use tickers.txt with parallel execution
- Add recommendation history section to Performance page with filters and charts
- Fix strategy name normalization (momentum/Momentum/Momentum-Hype → momentum)
- Fix strategy metrics to count all recs, not just evaluated ones
- Add error handling to Streamlit page rendering
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Major additions:
- ML win probability scanner: scans ticker universe using trained
LightGBM/TabPFN model, surfaces candidates with P(WIN) above threshold
- 30-feature engineering pipeline (20 base + 10 interaction features)
computed from OHLCV data via stockstats + pandas
- Triple-barrier labeling for training data generation
- Dataset builder and training script with calibration analysis
- Discovery enrichment: confluence scoring, short interest extraction,
earnings estimates, options signal normalization, quant pre-score
- Configurable prompt logging (log_prompts_console flag)
- Enhanced ranker investment thesis (4-6 sentence reasoning)
- Typed DiscoveryConfig dataclass for all discovery settings
- Console price charts for visual ticker analysis
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Youssef Aitousarrah