# Core Frameworks, Methodologies & Operating Systems

## The 12-Dimension AI Moat Scorecard
Every opportunity is scored 1–10 on each dimension. Weighted total with minimum thresholds by tier (Core ≥ 85/120, no dimension < 5).

1. Data Advantage Durability (15%) — flywheel strength, exclusivity, labeling cost, feedback velocity.
2. Talent Density & Retention (12%) — researcher quality, founder judgment, recruiting power against labs and Big Tech.
3. Compute Efficiency & Access (10%) — training/inference cost curves, hardware relationships, frontier cluster access.
4. Technical Differentiation (10%) — novel architecture, training paradigm, evaluation methodology, or deployment approach.
5. Go-to-Market & Distribution (10%) — sales complexity, customer lock-in, platform vs. point-solution nature.
6. Unit Economics at Scale (8%) — gross margins under realistic inference loads and token volumes.
7. IP & Legal Defensibility (7%) — patents, trade secrets, FTO, publication strategy.
8. Regulatory & Geopolitical Positioning (7%) — export control resilience, policy alignment, data sovereignty.
9. Founder-Market Obsession & Execution (7%) — domain depth, willingness to make hard tradeoffs, operational tempo.
10. Capital Efficiency (6%) — burn multiple relative to value creation, path to next inflection without excessive dilution.
11. Network/Platform Effects (5%) — data, developer, or user effects that compound advantage.
12. Exit Optionality (3%) — quality of strategic acquirers, public comps, secondary paths.

## Portfolio Architecture
- Core Compounders (60–65%): 10–14 positions, $40–120M checks, 6–10 year horizon, target 5x+ MOIC. Require 85+ Moat Score and category leadership potential.
- Tactical Accelerators (20–25%): 6–9 positions, $15–35M checks, 3–5 year horizon, 3–4x target. Higher risk tolerated for asymmetric payoff.
- Optionality Bets (10–15%): 4–8 small checks, pure shots on 10x+ outcomes. High failure rate accepted for learning and convexity.

Rebalancing triggers: position >15% of AUM, conviction drop >20 points, sub-sector >40%.

## Due Diligence Operating System
Stage 1 — Triage (30–90 min): founder background + pubs, product demo + technical artifacts, competitive 2x2 (AI-native vs incumbent+AI), first-principles TAM.
Stage 2 — Deep Dive (8–25 hrs): 4–7 customer/lost-deal calls, independent technical expert review, detailed compute cost model, full team references, four-outcome scenario tree with probabilities and payoffs.
Stage 3 — IC Memo: 8–12 page structured document, explicit probability-weighted return distribution, adversarial Red Team section, sizing recommendation with regret analysis.

## Valuation & Sizing
Primary: Probability-Weighted Expected Value using multi-state outcome trees.
Secondary: Quality-adjusted comparables (higher multiple for superior Moat Score and capital efficiency).
Sizing: Fractional Kelly (f=0.25–0.45) subject to 12% single-name cap, drawdown limits, and regret-minimization overlay.

## AI-Era Risk Models (Living)
- Inference economics shock (rapid cost deflation)
- Open-weights commoditization
- Regulatory phase change (training data or deployment restrictions)
- Talent arms race escalation
- Paradigm discontinuity (test-time compute, new architectures, agentic workflows)

Every Core position maintains a living Risk Memo updated quarterly.