Track record

Most ideas at a small investment firm aren’t lost because they were wrong. They’re lost because nobody wrote them down. When one works out, the only record is a memory: someone read about it once.

That costs more than the missed trade. Without a record of what you believed and when, you can’t tell skill from luck. You have no track record to show anyone, including yourself.

What keeping score takes

Four things, each harder than it sounds:

• A written claim. A thesis, an owner and a time horizon. A vague hunch can’t be wrong, so it can’t teach anything.
• A dated belief. Conviction recorded as it changes, not as one number. Otherwise every review is done with hindsight.
• A decision with reasons, including the ideas you pass on. Rejections are half the record.
• A review of every idea, whatever happened to it.

None of this is new. The hard part is that each step asks for effort at the moment people least want to spend it.

Making it cheap

At Munoth Capital I built this into the firm’s internal system. An idea moves through five stages:

The design choices that make it hold:

Recording costs people nothing.

A system that has to be fed by hand gets abandoned. AI does the digging, the transcription and the sorting. People only have to talk.

History can’t be edited.

Closing an idea needs a written reason, and a closed idea can’t be reopened. If someone acts against the vote, the vote they overrode is saved with the decision.

A veto and a vote do different jobs.

Anyone can stop an idea. The vote still runs, because knowing where the team stands is worth having even when nobody objects.

The rules live in the database.

Voting windows, closing rules and access are enforced in Postgres, not in the interface, so they hold however the data is reached.

What it changes

Every idea now has an owner, a dated record of belief, a decision with its reasons and a review. Over time that adds up to a record of judgment: which kinds of conviction to trust, and whose.