Commit Graph

3 Commits

Author SHA1 Message Date
Codex
ba938c7bd3 feat: complete event catalogue (160 events, 12 categories)
Per SENTIMENT_ANALYSIS_ENGINE_SPEC.md Section 5.2:
- Tokenomics: 16 (unlock, burn, mint, inflation, staking, bridge, liquidity, treasury, vesting, migration)
- Security: 13 (hack, exploit, audit, rug pull, exit scam, bridge hack, phishing, ddos, key compromise, upgrade fail, oracle, governance attack)
- Technology: 15 (mainnet, testnet, upgrade, fork, api deprecation, bug fix, performance, feature, sdk, mobile, layer2, cross-chain, deployment, audit)
- Governance: 9 (proposal new/passed/failed, whale vote, attack, treasury, delay, parameter, emergency)
- Financial: 18 (earnings, guidance, dividend, revenue, analyst, insider, buyback, ipo, spinoff, split)
- Market Structure: 16 (listing, delisting, halt, withdrawal suspend, liquid staking, mm program, whale, etf, liquidity, order book, funding, open interest)
- Regulatory: 15 (ban, clampdown, clarity, SEC, CFTC, MiCA, tax, CBDC, sanctions, legal)
- Media: 10 (mainstream, breaking, rumor, celebrity, influencer, viral, narrative)
- Social: 12 (viral, attack, community vote, AMA, dev quit, proposal, pump coord x3, split, debate)
- DeFi: 11 (yield, liquid staking, insolvency, liquidations, collateral, borrow rate, dex, impermanent loss, optimizer, vault)
- Macro: 15 (Fed, CPI, GDP, employment, inflation, central bank, geopolitical, PMI, retail, confidence)
- M&A: 10 (announcement, acquisition, tender, partnership, strategic investment, spinoff, bankruptcy, default, JV, licensing)

Total: 160 events across 12 categories matching spec.
2026-09-18 00:59:23 +02:00
Codex
f883d5851f refactor: unified weighted lexicon + centroid layer
- CryptoSentimentCalibrator: 2,086-term weighted lexicon (-100 to +100)
  * Priority-based span matching (longest-first, no double-count)
  * Whale phrases ±50, compounds ±30, dot-separated ±20, singles ±10-25
- Calibration logic: lexicon wins on disagreement, amplifies on agreement
- CentroidManager: 6 params × 1024-dim built from lexicon via e5-large-v2
- ScoringEngine._refine_with_centroids: fixed attribute access bug
- config/centroids/*.npy: padded to 1024-dim (e5-large-v2 output)
- lexicon_weights.json: generated unified lexicon
- Unit tests: 46/46 core NLP tests pass
- Labeling pipeline: 22 samples processed
- All 19 critical crypto semantic tests + 6 calibration scenarios pass
2026-09-17 21:17:47 +02:00
Codex
a276aeaded Add sentiment_engine with CryptoSentimentCalibrator fixes - improved keyword lists, lowered FinBERT threshold, added neutral handling 2026-09-14 13:30:05 +02:00