AI Resilience is the only LLM-scored module in the framework. Claude Sonnet evaluates each company across five dimensions using earnings transcripts, thesis narratives, financial data, and peer context. Scores are calibrated against reference anchors to prevent drift.
| Dimension | Weight | What It Measures |
|---|---|---|
| Revenue Catalyst | 25% | Degree to which AI directly drives incremental revenue. Direct AI product sales, AI-enabled upsell, AI-driven customer acquisition. |
| Moat Durability | 25% | Switching costs, data moats, network effects, system-of-record status. How defensible is the business against AI-native competitors? |
| Operating Leverage | 15% | AI's impact on margin improvement, revenue-per-employee, cost structure. Can the company use AI to scale without proportional cost increases? |
| Pricing Resilience | 20% | Vulnerability to seat compression, pricing disruption, commodity risk. Will AI force price reductions or enable premium pricing? |
| Obsolescence Shield | 15% | Protection from AI-native replacement. Regulatory barriers, physical infrastructure requirements, domain expertise moats. |
Claude Sonnet receives a comprehensive data package for each ticker. The context is gathered from two functions: _gather_ai_score_context() (base) and _gather_ai_resilience_context() (extended).
| Data Source | What It Provides | Dimensions Informed |
|---|---|---|
| Company profile | Name, sector, subsectors, market cap, 5Y hist/fwd revenue CAGR | All — establishes business context |
| AI themes | Themes from ticker_themes where category=‘ai’: theme name, source, occurrences, last_seen. Plus AI/total theme ratio. |
Revenue Catalyst, Operating Leverage |
| Competitive themes | Themes where category=‘competitive’: switching costs, moat signals | Moat Durability, Obsolescence Shield |
| Operational themes | Themes where category=‘operational’: efficiency, cost structure | Operating Leverage, Pricing Resilience |
| Transcript summaries | Last 2 quarters: headline, key themes, full summary (1200 chars), mgmt tone, sentiment | All — qualitative evidence for AI narrative |
| Financial metrics | Last 4 quarters: revenue, YoY growth, gross margin, operating margin | Operating Leverage, Revenue Catalyst |
| Margin trend | Up to 8 quarters (chronological): GM, OM, NM trajectory | Operating Leverage, Pricing Resilience |
| Balance sheet | Last 4 quarters from financial_health: cash, debt, net debt, FCF, capex, capex % of revenue |
Operating Leverage (AI investment capacity) |
| Consensus estimates | Forward revenue and EPS from analyst_estimates |
Revenue Catalyst (forward growth trajectory) |
| Investment thesis | Current narrative + risk factors (if thesis exists) | All — qualitative context |
| Peer AI themes | AI themes prevalent in the company’s peer group (from peer_group_themes) |
Moat Durability, Obsolescence Shield (competitive positioning) |
| Analyst notes | User-written notes (last 10) | All — analyst overlay |
| Calibration anchors | Hardcoded reference scores for known tickers (binding ±15 points) | All — prevents LLM score drift |
AI Resilience does not depend on peer_group_memberships, theme_momentum, financial_health (except for balance sheet context), or investment_theses (reads thesis if available, not required). It primarily needs ticker_themes, earnings_summaries, earnings_metrics, and companies — all of which have 100% coverage.
Reference scores are provided in the prompt to ground the LLM's output. On a 1–5 scale (mapped to 0–100):
| Company | Type | Rev Cat | Moat | Op Lev | Pricing | Obsol |
|---|---|---|---|---|---|---|
| PLTR | Software | 5 | 4 | 4 | 3 | 3 |
| U (Unity) | Software | 2 | 2 | 2 | 1 | 1 |
| ORCL | Software | 3 | 5 | 3 | 4 | 4 |
| CEG | Infrastructure | 5 | 5 | 4 | 5 | 5 |
| NEE | Infrastructure | 3 | 4 | 3 | 4 | 5 |
| BE | Infrastructure | 3 | 2 | 2 | 2 | 2 |
Scale mapping: 1→20, 2→40, 3→60, 4→80, 5→100
Soft cap at 92: scores above 92 require identified vulnerabilities in the rationale. Every dimension must cite at least one vulnerability. This prevents score inflation from overly optimistic LLM outputs.
For companies in Energy, Utilities, Materials, Industrials, and Real Estate, the dimensions are reinterpreted:
| Dimension | Infrastructure Meaning |
|---|---|
| Revenue Catalyst | % revenue from AI/data center customers |
| Moat Durability | Contracted backlog depth, supply constraints, switching costs |
| Operating Leverage | Execution track record, margin expansion from AI demand |
| Pricing Resilience | Supply-constraint pricing power |
| Obsolescence Shield | Secular demand duration (10–20yr structural theme) |
US data center power demand is projected to 3x to 134.4 GW by 2030, with $720B in grid spending and 7–15 year capital cycles. Infrastructure companies are scored against this structural backdrop.
| Label | Score Range | Meaning |
|---|---|---|
| Fortress | ≥80 | AI is a major tailwind; business model is strengthened by AI adoption |
| Defensible | ≥60 | Well-positioned with manageable risks; AI is net positive |
| Moderate | ≥40 | Mixed picture; some AI benefit but meaningful vulnerabilities |
| At-Risk | ≥20 | Significant AI disruption risk; business model under pressure |
| Vulnerable | <20 | High probability of AI-driven disruption to core business |
AI Resilience flows into conviction scoring in three ways:
Neutral scores (~50) carry only 0.3 weight — AI is not informative for this company, so theme momentum and peer standing dominate. Extreme scores (fortress/vulnerable) carry up to 1.0 weight — AI clearly matters, so the signal gets full voice. Other thematic components always carry weight 1.0.
Without it, AI resilience dominated the thematic dimension for 40+ tickers that lacked theme momentum and peer ranking data — making thematic = AI resilience alone. A neutral AI score of 50 would anchor thematic at 50 regardless of strong theme momentum or peer signals. The adaptive weight ensures AI resilience speaks loudly when it has something meaningful to say (fortress or vulnerable) and fades when it doesn’t (moderate/neutral).