Methodology
How signal cards are generated
Deterministic, auditable scoring from public data sources. No black-box AI. No paid APIs. Source-linked evidence.
Loading methodology…
Scoring dimensions
Five scores per signal card — each computed without LLM.
Evidence strength
Source reliability × scientific score from the originating source. Higher = stronger public-data basis.
Market relevance
Catalyst score × drug/target relation score. Reflects expected public visibility of the event, not a price forecast.
Urgency
Total signal score + catalyst proximity weighting. Higher urgency = closer event horizon, higher evidence.
Confidence
Data completeness + source trust level. Reflects how reliably the signal can be verified from the public source.
Safety risk
Adverse event data and change-type classification from openFDA FAERS. Safety monitoring is separated from catalyst ranking.
What is automated today
✅ Automated
Source-specific public-data ingestion
Persistent source snapshots
Field-level change detection
Ticker/company/drug mapping
Deterministic five-dimensional scoring
Data quality scoring and real/demo labelling
Historical reaction context
Compliance-screened alert generation
Persistent source snapshots
Field-level change detection
Ticker/company/drug mapping
Deterministic five-dimensional scoring
Data quality scoring and real/demo labelling
Historical reaction context
Compliance-screened alert generation
🔬 AI-optional (Ollama)
LLM entity extraction from free-text filings
Natural language signal summaries
Semantic matching without exact keywords
Relevance/noise filtering (planned)
ML scoring calibration (Phase 3+)
Natural language signal summaries
Semantic matching without exact keywords
Relevance/noise filtering (planned)
ML scoring calibration (Phase 3+)
Important limitations
- Scores reflect public-data evidence strength — not expected price movements.
- Historical reaction context is observational. It does not predict future outcomes.
- All signals require human review before any action.
- BioCatalyst Radar does not provide personalised investment advice.
- Safety monitoring is hypothesis-generating only and requires clinical review.
- Source-linked evidence only — does not reflect undisclosed information.