▚ STOUT STREET · PAPERSMMXXVI
WHITE PAPER №02 · 2026 EDITION

LEVCAP

The lab-embedded record is real, overlapping, and concentrated. Deduplicated, it still justifies a national network · at 7.7-11.5 : 1, not 44 : 1.

Stout Street Capital · July 2026 · PDF · 20 pages · 8,651 words · 216 KB
$ read the pdf → ▚ save levcap-white-paper.pdf
≤ 375
companies, deduplicated
≤ $6.6B
follow-on funding, dedup ceiling
29%
of that total is one company
$7.8M
steady-state node cost / yr
[ abstract ]

What it argues.

A Lab-Embedded Venture Capital Access Program: entrepreneurial ventures embedded inside research laboratories, connected to capital and industry at the same time. This edition reports the evidence deduplicated · Cyclotron Road sits inside both the Activate and LEEP totals · traces Fervo Energy from fellowship to power purchase agreement, answers the equity question a venture-firm author is uniquely obliged to answer, prices a twenty-node federated network a decision-maker can dispute, and states where the model loses to I-Corps, ARPA-E, gap funds and Fraunhofer.

[ figures ]

The evidence, drawn.

fig. 01 · the evidence, deduplicated
Activate, as published$5.5BDOE LEEP, as published$6.0BStacked, as earlier read$11.5Bthe same $4.9B, twiceDeduplicated ceiling≤ $6.6B▨ Fervo Energy $1.89B · 29% of the total
Cyclotron Road is simultaneously a LEEP node and an Activate program, so its $4.9B sits inside both parent totals. Removing the overlap once costs the record 43% of its apparent funding · and one company still carries 29% of what remains.
fig. 02 · a ratio is its denominator
Lab-embedded (2024)44 : 1denominator excludes the selection apparatus and the failuresNSF I-Corps10.5 : 1whole program cost, all 2,500+ teamsARPA-E3.4 : 1every award dollar, incl. technical failuresThe 20-node network7.7-11.5 : 1fully loaded: overhead and failures inside the denominator
Read as published, lab embedding looks four times more efficient than I-Corps. The hatched bar measures the quality of a screen, not the return on a program. Nobody can act on a ratio whose denominator is undefined; 7.7-11.5:1 is the number this paper will defend.
fig. 03 · what the network costs, and what it should produce
Research allocations20 fellows × $200k$4.0MStipends + benefits20 fellows × $150k$3.0MProgram staffdirector + 3$0.8MONE NODE, STEADY STATE$7.8M / yr× 20 nodes$156M / yr× 5 years$780M→ companies, at 5 per node-year≈ 500→ jobs, at the programs' ~20 per company≈ 10,000→ follow-on funding, at the dedup record$6-9B
Stated so a decision-maker can dispute them. The earlier 50,000-job ambition fails its own arithmetic · the reference programs create ~11-20 jobs per company, so 500 companies means ~10,000 jobs, not 50,000. Facility access is contributed in kind and counted, not paid.
[ contents ]

Inside the paper.