- data_provider: new fetch_historical() that paginates /time_series in
5000-bar chunks (1 credit/chunk) for arbitrary date ranges. Tolerates
the API's "no data on specified dates" boundary error gracefully when
prior chunks succeeded — that error is just the marker that we walked
past the symbol's earliest available bar.
- seed_history.py: one-shot CLI that downloads OHLCV for the configured
pairs/TFs into history/<PAIR>_<TF>.csv. Skips existing files unless
--force.
- backtest.py: walk-forward replay through SignalAggregator with strict
no-look-ahead semantics. Each bar is only visible to the engine after
it has fully closed (`df.index + interval <= as_of`). Trade simulation
is first-touch (SL vs TP1) with worst-case (SL first if both touched
same bar), spread cost deducted from R, max-hold timeout, no-overlap.
Reports per-grade win-rate / avg-R / profit-factor / max-DD, plus
optional matplotlib equity + by-grade plots.
- requirements: add matplotlib (lazy-imported, only needed for --plot).
- .gitignore: history/ and backtest_results/ regenerable, kept untracked.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- Migrate news_engine to google-genai SDK with gemini-2.5-flash-lite
- Batch all headlines into one API call per scan (30 calls -> 1)
- In-memory cache keyed by headline hash; fallback to keyword scoring
- Update requirements.txt: add google-genai, python-dotenv; remove transformers
- Add .gitignore; update CLAUDE.md with Gemini live section
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Harwinder