Built bottom-up from actual experience + verified knowledge, NOT aspiration. Buffett’s “20-punch card” idea: out-of-circle names should be fast-skipped — don’t consume attention budget on them.

Trigger mechanism: when a sector / industry is mentioned in conversation, auto-check this file:

  • In-circle → full thesis analysis + Interconnections framework
  • Edge of circle → explicitly tag “learning area, apply Interconnections cautiously + open questions”
  • Out-of-circle → fast-skip, suggest stay out

Status: Starter version 2026-05-18, awaiting calibration + ongoing expansion.

In-circle (Strong edge — full thesis authority)

AI hands-on capability assessment

  • Edge source: Deep first-hand use of mainstream AI tools → can judge whether the market is over-optimistic about AI’s actual deployment capability
  • Confirmed applications:
    • ACN AI-substitution thesis skepticism (based on real AI consulting capability assessment, judge that AI cannot capture 30–40% share)
    • DUOL DET application-rate read (AI substitute/augment in language-learning context)
    • SPGI AI-threat-narrative-mispriced judgment (field research, SP500 corporate governance ratings, etc. — AI cannot do these)
  • Limits: Not in-circle on AI model architecture / training economics

Language learning consumer apps

  • Edge source: DET as first-hand test taker + DUOL as user
  • Confirmed applications: DUOL DET strategic role analysis (brand pawn, not standalone revenue)
  • Limits: K12 education / corporate training are out-of-circle

Mag 7 cross-comparison (relative risk positioning)

  • Edge source: Continuous tracking + Interconnections framework applied along the AI vector
  • Confirmed applications: AMZN → MSFT swap (within Mag 7, MSFT carries the lowest AI-exposure risk); GOOG/META capital-allocation relative positioning
  • Limits: Deep operational analysis of a single Mag 7 company (e.g., Azure margin breakdown) is edge-of-circle

Edge of circle (Learning — apply Interconnections cautiously)

US Defense sector

  • Status: 2026-05-17 LDOS 5-round research established the faction + ultimate-beneficiary + EO discretionary-buyback framework
  • Open questions:
    • ITA basket economics vs single-name picking
    • Does CACI’s premium valuation reflect a faction advantage (vs LDOS being mispriced)?
    • PLTR / AXON / Booz / SAIC relative faction positioning
  • Promote criteria: Complete ≥ 2 defense-name comparative studies + 1 government-budget cycle tracked

US Telecom (incumbent carriers)

  • Status: 2026-05-18 VZ research established cable MVNO threat + FWA + capex cycle second-order library
  • Open questions:
    • TMUS upgrade path vs VZ / T defense
    • Long-term equilibrium cable wireless share (will 45% → 50%+ stop somewhere?)
    • Timing of the 6G capex cycle and its FCF impact
  • Promote criteria: Complete 1 cycle tracked + 1 TMUS vs T comparative study

Financial data infrastructure

  • Status: Hold SPGI + V + IBKR + AXP, but the thesis rests more on “upstream/downstream + defensive” than on a deep sector model
  • Open questions:
    • SPGI ratings business vs index business vs Platts data business — relative growth drivers
    • V vs MA fee-structure differences under new payment rails (RTP, stablecoin) and how they evolve
  • Promote criteria: Complete SPGI three-segment analysis + detailed V/MA comparison

AI-power / data center electricity (IPP)

  • Status: 2026-05 TLN hold + sell crystallized the framework “AI-power is a component of the AI bubble”
  • Edge source: AI hands-on edge → recognize AI-power narrative as an extension of the AI bubble (not an independent utility thesis)
  • Current view: Short-term bearish — pure AI-capex beneficiary, AI-bubble-late-stage priced-in, high leverage amplifies downside
  • Long-term: Uncertain — depends on whether AI genuinely produces new incremental demand (vs converting existing demand) → after AI-capex peaks, the IPP investment logic will reset
  • Representative names: TLN (sold), VST, CEG, and similar nuclear / IPP names with hyperscaler PPAs
  • Open questions:
    • Real economics of PPA structures with hyperscaler long-term contracts
    • Nuclear regulatory path (Three Mile Island restart, SMR rollout)
    • IPP leverage sensitivity through rate-hike cycles (the TLN case is the lesson)
  • Promote criteria: Wait for AI-bubble situation to clarify / complete VST + CEG + TLN three-IPP comparative study

Out-of-circle (Fast-skip — do not form a thesis)

Sectors / strategies explicitly not participated in:

  • Biotech / pharma early-stage — FDA approval / pipeline probabilities not reliably estimable
  • Semis manufacturing process detail — TSMC / Samsung fab process minutiae (different from NVDA application-layer judgment)
  • Early startup / VC private — not participated in
  • Forex trading — not a thesis-driven school
  • Crypto / digital assets — not participated in (unless KB is recalibrated)
  • Commodities active trading (not the GLDM fiat-debasement long-term thesis hedge) — not participated in
  • Options / derivatives strategies — not participated in (holdings only simple equity + ETF)
  • Distressed / bankruptcy restructuring — Klarman 1980s style is not suited to the current environment

Maintenance

  • New sector library added → assess whether the sector belongs in-circle / edge / outside
  • Quarterly review: should any edge-of-circle sectors be promoted / demoted?
  • Hands-on experience (e.g., new product use, on-the-ground industry observation) → can push a sector up into the inner circle
  • “I suddenly feel like looking at [out-of-circle sector]” → should trigger a discipline check: is real edge actually forming, or is this narrative-driven curiosity?