Methodology
How Silicon Analysis Ranks AI & Physical-AI
Live scores blend real-time news analysis, price momentum, and data confidence metrics. All calculations are deterministic and explainable.
Alpha Score (Explainable)
Alpha is now a weighted composite of three explainable components. Each company page shows the breakdown.
Aggregated news impact, sentiment, and severity from the Intel Wire. More high-impact coverage = higher score.
24-hour price momentum mapped to 0–100. Neutral at 50, positive movement scales up, negative scales down.
Data completeness factor. Penalizes stale prices, missing news, or data gaps. High confidence = reliable signal.
Divergence Signal
Divergence measures how Alpha movement differs from price movement over 24 hours. Range: -100 to +100.
Alpha is rising faster than price. May indicate undervaluation — news and fundamentals ahead of market pricing.
Price is rising faster than Alpha. May indicate overvaluation — market pricing ahead of fundamentals.
Legacy Sub-scores
Still available on company pages for reference.
- Financial Strength: balance sheet resilience & capital velocity.
- AI Intensity: percentage of R&D, CAPEX, and roadmap tied to AI.
- Hardware Quality: node, bandwidth, memory, and performance proxies.
- Transparency: disclosure quality and governance signals.
- Sentiment: near-real-time tone derived from the Intel Wire.
Intel Wire Severity
CRITICAL/BREAKING items imply near-term risk or execution shifts. CONFIRMED reflects validated disclosures. RUMOR carries lower reliability and is marked accordingly. ANALYSIS adds human/LLM context on strategic impact.
Hardware Assumptions
Specs (node, bandwidth, yield) are sourced from vendor docs and supply-chain chatter. Deal/shipment volumes are directional and updated as new PO or qualification intel lands.
Alerts
The system generates alerts for significant changes: rank jumps (3+ positions), alpha spikes (10+ points), critical news, pipeline staleness, and divergence signals (20+ points).
Technical Notes
- All scoring uses deterministic math — no LLM in the scoring path.
- Price data is cached; no external API calls on page loads.
- Score recalculation runs hourly via background agents.
- Model version tracked per snapshot (currently
alpha_v2.0).