Memos 02

Companies, read closely.

Investment memos on early-stage companies, one at a time. Only public sources, every claim linked, the bear case written by me, and a dated prediction I can be held to.

Who writes this. I'm a student at Minerva University, not a professional investor or analyst. I write these memos to learn how to evaluate companies, in public. Everything comes from public sources, and every claim links to where it came from. I hold no position in the companies I write about, and no one pays for these memos. I will get things wrong: corrections are welcome at [email protected]. Not investment advice.

The map is a commodity. The planning desk might not be.

2026.11 · Series A · San Francisco · City & utility infrastructure

CivilGrid

The map is a commodity. The planning desk might not be.

Key idea

Everyone calls CivilGrid “Google Maps for the underground.” But only 2.16% of US utility damages trace to a bad record or map. The value PG&E reported was paving money, found by screening planned projects against city plans. If that screen repeats every planning cycle, CivilGrid is a planning company with a map attached.

The company

CivilGrid collects records of what is buried under US streets directly from the asset owners and puts them on one map engineers use before they design. $26M Series A led by Spark Capital, August 2026.

In numbers

5/10: confidence it becomes a venture-scale planning company
$60M+: paving conflicts PG&E reports, from a one-time screen
2.16%: of 2024 utility damages caused by a bad record or map

Read the memo

PG&E newsroom · 2026-08-27“$60M+ in paving conflicts eliminated.”CivilGrid case study“Identified $60M in Avoidable Costs.” Same number, two verbs. The customer claims more than the vendor.

A run rate built on inventory that expires.

2026.11 · Series A · San Francisco · RL environments for finance AI

Halluminate

A run rate built on inventory that expires.

Key idea

Halluminate's CEO says environment complexity must roughly double every six to eight months, and tasks across the category may be retired near a 70% pass rate. So the useful question about its company-reported run rate is what a dollar is worth a year later. That depends on what's kept after experts and how much work repeats, and neither is public.

The company

Halluminate builds finance simulations and expert-written tasks that AI labs use to train agents. $30M Series A led by Oak HC/FT, October 2026.

In numbers

4/10: confidence it becomes a durable, high-margin business
2.8x–4x: complexity a year at the CEO's stated pace
143%: repeat needed to keep 50 cents, on my assumptions

Read the memo

Fortune · 2026-10-01“based on quarterly revenue from work already delivered and paid for.”Epoch AI · 2026-01-12Tasks may be retired near a 70% pass rate. A real revenue basis, and a clock on what was sold.

The next memo is on the way.