PORTFOLIO GENESIS / LOG-136
How Experiment 00 was actually assembled.
One origin record, with a hard line between what was sealed on 5 September 2026 and what was reconstructed later.
ORIGIN RECORD · VERSIONED
What is sealed
These facts existed in the machine-readable mandate before any forward result was known.
- 26 lines · 25 equities + SHV
- $100,000 initial paper capital
- 100% exact target weights
- 0× leverage
What is reconstructed
The exact 26 lines and weights are real. The full point-in-time candidate universe, rejection log and contemporaneous valuation workbook were not preserved at inception, so the narrative below is explicitly forensic.
- Point-in-time global candidate universe
- Per-candidate inclusion and rejection scores
- Contemporaneous valuation snapshots
- Signed investment-committee minutes
08 LAYERS / 100%
The starting architecture
A thesis-led map of the AGI industrial stack—not a capitalization-weighted index and not an optimizer output.
Models + cloud
Frontier models, distribution and enterprise demand
Compute
Accelerators, custom silicon and software ecosystems
Semicap + memory
Fabs, EUV, yield, packaging and HBM
EDA + IP
Critical software and IP for complex chip design
Networking
AI fabrics, optics and high-speed interconnect
Power + grid
Electricity, cooling, generation and construction
Data centers
Interconnection-rich locations and deliverable power
Reserve
Liquidity for drawdowns, costs and opportunities
CADENCE / GOVERNANCE
Three clocks, not one quarterly blackout
Daily research does not imply daily trading. It means a material event can reach the decision gate on any day.
Evidence is collected every day. A headline alone cannot change capital.
A corroborated earnings, policy, technology or thesis event can trigger an off-cycle decision immediately.
Every position is reviewed at least quarterly even if no event crossed the gate. Execution remains human-authorized.
STARTING LEDGER / 100%
The complete starting ledger
Twenty-five equities plus a short-duration Treasury reserve. Target weights sum to 100%.
Role in the thesis
LayerTargetFrontier models and global distribution
Models + cloud10%Enterprise distribution, cloud and model economics
Models + cloud9%AWS, custom silicon and AI platform
Models + cloud8%Accelerated-compute platform
Compute8%Frontier models, distribution and monetization
Models + cloud7%Custom accelerators and networking
Compute7%Leading-edge foundry bottleneck
Semicap + memory7%EUV lithography bottleneck
Semicap + memory5%Electrical distribution for data centers
Power + grid4%Liquidity reserve and dry powder
Reserve3.5%AI cloud capacity, data and enterprise reach
Models + cloud3%High-speed Ethernet AI fabrics
Networking3%Power and thermal management
Power + grid3%Carbon-free baseload generation
Power + grid2.5%Generation and grid equipment
Power + grid2.5%Second-source accelerators and CPUs
Compute2%Process control and yield management
Semicap + memory2%Custom silicon and optical interconnect
Networking2%Transmission and mission-critical construction
Power + grid2%Materials engineering and advanced packaging
Semicap + memory1.5%HBM and advanced memory
Semicap + memory1.5%EDA and system-design software
EDA + IP1.5%EDA, IP and verification
EDA + IP1.5%Interconnection-rich data centers
Data centers1.5%Etch, deposition and advanced packaging
Semicap + memory1%Space-compute optionality
Models + cloud1%What this record cannot prove
The forward chart begins only after the mandate was sealed.
Reconstruction is labelled; missing provenance is not invented.
Research may propose. A human capital decision is still required.