Stars are bought. Shipping isn't.

Find the builder before
the database does.

Filter builders by un-fakeable build telemetry — tokens, build time, languages, what they actually ship — captured at the source and tied to the person, often months before they have a deck or a domain. Set a thesis, surface who matches, and reach out first. The same signal YC built Paxel on — open to you.

Not affiliated with or endorsed by Y Combinator. Paxel referenced as public category validation.

0
tokens tracked
0
builds shipped
0
builders
The signal

Every signal you scout on is gamed or late.

Every legacy proxy for builder ability broke at once. Build telemetry — how someone actually ships with AI — is the one signal that didn't. Here's why the rest fail you.

GitHub stars

A diligence liability.

~$85 buys the 2,850-star seed-stage median — a 3,500× return on fraud. ~6 million fake stars across 18,617 repos, AI/LLM repos the largest category.

Carnegie Mellon University, StarScout · ICSE 2026

The pitch

Rewards charisma.

A deck rewards whoever presents best, not whoever ships best. By the time there's a deck, the edge is gone — and so is the price.

Theo Browne: “a $50 problem with a $50 million consequence.”

Sourcing tools

Index companies after they form.

Harmonic, Evertrace and gitdealflow scrape external, gameable signals — stars, commit velocity, registries — attributed to a company. You see the builder once everyone else can.

Evertrace: “detect founders before they show up in any startup database.”

The moat

Un-fakeable build telemetry, tied to the person.

Vibe coding exposed the gap between generating code and shipping. Promptbook shows who actually ships — tokens, build time, real shipped work, captured continuously from the terminal. Stars cost three cents. A 2-million-token build can't be bought at any price.

Polywork raised $44.5M and died — self-reported data, and no buyer to give the profile a job. Ours is captured at the source, so it can't be faked, and you are the buyer — the supply compounds on its own.

GitHub stars$0.03 each · 6M faked
Commit velocityscraped · gameable
LinkedIn / résuméself-reported
A 2M-token buildcaptured at the source — un-fakeable

How it works for you

From thesis to first check.

No data-science team, no scraping pipeline. Set a thesis on real build axes and reach the builder directly — often before there's a company to scrape.

STEP 01

Filter to your thesis

Filter the whole pool on real axes — tokens, build time, builds, languages, the tools they live in. You set the bar; we never sell you a ranked match-score.

STEP 02

Surface the outliers

See exactly who clears it — the builders shipping hardest with AI, captured at the source, tied to the person, before they're on anyone else's radar.

STEP 03

Reach out first

Message the builder directly, while they're still just shipping. Get the warm intro before there's a deck, a domain, or a competing term sheet.

Founding access

Find them before
the database does.

We're onboarding a first cohort of scouts and funds that source on proof, not stars. Leave your details and we'll set up your access.

Un-fakeable build telemetry, tied to the individual
Builders captured months before a deck or a domain
The same signal YC keeps for Paxel — open to you

Builders are free, always. The data is theirs — never pay-to-rank.

Not affiliated with or endorsed by Y Combinator. Paxel and the GitHub fake-star figures (Carnegie Mellon University, ICSE 2026) are referenced as public category validation.