Skoor for funds
Who is building, measured.
Founders and early companies, Skoored on your thesis from the public record.
GitHub, arXiv, and the Y Combinator directory, read every day and turned into a Skoor from 0 to 100 with its n, its confidence, and the evidence attached. Your weights. Your data. Unscored until measured.
Live boards, refreshed daily.
The problem
The record is public. Reading it is the job.
Every week a few thousand people push code, post a paper, or start hiring in the sectors a fund cares about. Almost all of it is visible to anyone. Almost none of it is read, because reading it means a person, a spreadsheet, and a month. Skoor reads all of it, every day, and keeps the n.
How a Skoor is made
Four steps, none of them hidden.
Signals are observed, weights are the fund’s, the Skoor adds them up, and the memo says where every line came from. A partner can read the whole chain in a minute.
Repos created and pushed, commits in ninety days, contributors, code that a paper points to, first authorship, batch, team size, open roles. Each one dated, weighted, and linked to where it was seen.
The fund's thesis is a table: one row per signal, one number per row. Change a weight and every Skoor recomputes. Add a sector by adding its vocabulary. Nothing is hidden in a model.
Additive, 0 to 100, with n and confidence. Fewer than three observations is null, and a null row stays off the board. The score says what the record shows, not who is good.
One page per surfaced founder or company, written by Claude from the pipeline's evidence and nothing else. Every sentence cites its source. Unknowns are listed as unknown. It ends with the questions to ask in the room.
The boards
One board per sector. Every row shows its evidence.
Four sectors run today on a public thesis. A fund’s own boards run on its own thesis, with its own vocabularies, on its own account.
Manipulation, locomotion, autonomy, simulation, the tools around them.
Open the board →DefenseSensing, autonomy, space, the dual-use work that reads as both.
Open the board →EnergyStorage, grid software, fusion, the hardware and the control loops.
Open the board →SemiconductorsDesign tools, open hardware, photonics, the chips and what runs on them.
Open the board →Two Skoors sit on each board. The Founder Skoor reads a person: what they own, what they ship, what they publish, and what their own page says. The Breakout Skoor reads a company: its repos, its contributors, its batch, its hiring. Both use the same scale, the same null rule, and the same evidence lines.
The memo
One page, every sentence sourced.
For anyone who reaches the board, the research agent writes a memo from the evidence the pipeline holds and nothing else: what they built, the record with its n, why now, what is unknown, three questions for the room, and what would move the Skoor either way.
From the repos and the papers, in plain words, each claim linked to the page it came from.
The measured signals with n, separated from anything inferred. Inferences are labeled as inferences.
Funding, team, customers, legal name: if the evidence does not say, the memo says so instead of guessing.
Three questions, each testing one specific claim above, written for the first meeting.
What it is not
A number from the record, not a verdict.
Not a ranking of who is good.
A ranking of who the public record says is building, measured the same way for everyone, with the record attached.
Not reconstructed history.
GitHub no longer publishes star timelines, so velocity comes from commits, contributions, and daily snapshots the service takes itself.
Not scraped from behind a login.
Public APIs and public pages only. A person's page says what they chose to publish, and the memo says where each line came from.
Not a score of people you already know.
Most of the board is names no one has emailed yet. That is the point of reading everything instead of the inbox.
Fit
Built to run inside the fund, not beside it.
A fund that wants its own read of the market does not want another vendor’s list. Skoor is installed on the fund’s account and left there, with the people who built it on call.
Your GitHub token, your Claude key, your database, your weights. The pipeline is installed for you and belongs to you; nothing routes through Skoor.
Sectors are vocabularies, weights are a table, both in version control. The partner who changes a weight is the partner who owns the ranking.
Every sector refreshes once a day. Any founder or company can be re-read and re-memoed in a minute before a meeting.
Every board is a page and a table. Every memo is Markdown. Everything the agents write lands in the tools you already keep deals in.
Memos are drafted by Claude from the pipeline’s evidence and read by a person before anyone acts on them. Claude carries SOC 2 Type 2, ISO 27001, HIPAA, and GDPR.
The pipeline runs under the fund’s own Anthropic account. Commercial Claude usage is billed by Anthropic, and by default Anthropic does not use data from commercial deployments to train its models.
Those are Anthropic's, for Claude. Anthropic Trust Center →
Your thesis, as a board, by next week.
A walkthrough is forty minutes: your sectors as vocabularies, your thesis as weights, and the first board rendered on your account. You leave with names you have not emailed yet, each with its evidence.
Claude is a product of Anthropic, PBC. Company listings come from Y Combinator’s public directory. Skoor is an independent company and is not affiliated with or endorsed by Anthropic, Y Combinator, GitHub, arXiv, or any fund.