019 min

What judges actually score

The seven official criteria, the AI that reads you before a human does, and what they explicitly don't grade.

The seven criteria

Colosseum publishes what it evaluates. Almost nobody reads it, and it explains most losses.

Founder-market fit and motivation: why you, specifically, for this problem.
Unique insight and technology: what you know that the room doesn't.
Product execution quality: does it work, and does it feel built.
Market size: is the ceiling worth an investor's time.
Founder communication: can you explain it in two minutes.
Business viability: how it makes money.
Traction and demand: proof someone wants it.

The official rules, in full

Something reads you before a human does

Colosseum runs an internal AI evaluation engine that filters submissions and raises flags before their team works through them. It scores founder-market fit, product velocity, user demand, market timing, and what they call evidence of grit in your past.

Humans still review afterwards. But something is reading your history before they reach you, and your history is not something you can build in four weeks, which is an argument for entering with the track record you already have, written down and linkable, rather than a clean slate.

After panel judging, a smaller group gets a fifteen-minute call. That call is about velocity and commitment, not a second demo. Expect to be asked what you shipped last week and whether you are doing this full time.

Two published criteria lists, and they disagree

The seven above are what Colosseum publishes in its own guidance. Section 8 of the official rules lists six different ones: functionality, potential impact, novelty, UX, open-source and composability, and business plan. Both are real, and where they conflict the rules are the document that settles an argument.

The sharpest conflict is about code. Colosseum's FAQ says they are not looking for a particular language or framework, or for design patterns, best practices, or code-quality checks, which is often quoted as "they don't grade your code". The rules say the opposite in the first criterion: "How well does this Project Submission work? What is the quality of the code?"

The rules also reward open-sourcing and composing with existing protocols directly, which the seven criteria never mention. If your repo is public and your product builds on other people's primitives, say so explicitly in the submission. It is a scored criterion, not a footnote.

Are you even eligible

Narrower than most people assume. This is for new startups that have not raised significant outside capital, and you are judged on work done inside the window. If you have already raised a round, read the rules before you spend four weeks on it.

Pitch and deck tactics

The first ten seconds decide the rest. Judges review hundreds of submissions, so open with the problem and your solution, not with your team's origin story.
People remember people, not PDFs. Record yourself. Let them feel the team.
Slides support you, they don't speak for you. Less text, more clarity.
Structure: Problem → Solution → Demo → Team. Deviating from this costs you clarity and buys you nothing.
Scored does not mean it belongs in the video. Write the business model down where the form asks for it, and keep the three minutes for the problem, the product and the team. Revenue models and roadmaps read as padding at this length.

Show why your team matters

Don't list names and titles. Say why each person is the right one to execute this specific thing.
Link proof of execution: shipped products, repos, portfolios. Builders, not thinkers.

Product and audience

If it's for everyone, it's for no one. A specific problem for a narrow audience beats a general product every time.
Ship early, polish later. You don't have to be perfect to win, but you do have to be functional.