Before conviction becomes capital, put the deal on trial.

A second opinion you control is not a second opinion.

Every investor now has a research chatbot open, and it is very good at building the case for a deal you already like. The obvious fix is to prompt it for the bear case. But a bear case you commission from your own AI answers to you: if it stings, you can soften the prompt, run it again, or close the tab, and no one will ever know.

Dissent only counts when it has standing: rules that cannot be bent mid-argument, a checker who does not report to you, and a record that cannot be edited after the fact. Nothing in the current stack provides that.

After the deal has momentum. Before it has money.

Use it after a deal has a sponsor and genuine interest inside your firm, and before the investment committee turns that interest into conviction. That is the last cheap moment to be wrong. Every week after it, the same doubts get more expensive: diligence dollars go out, the founder's story hardens, and momentum starts doing the arguing.

The point is not to replace diligence. The point is to aim it before you spend it.

A seat is a reviewer with a fixed mandate.

Six sit at every table:

Each seat reads the same evidence file without seeing the others' notes. Then they argue, on the record, and every seat must name the opposing argument it cannot beat. A managing adjudication weighs the surviving arguments, never the votes.

Eight questions the deal must survive.

Together the seats must answer them on the record, each question owned by the seat whose mandate is to press it:

Then the memo has to survive an audit it did not write: a second AI from a different company checks every source, every date, and every quote before the memo can ship. Every factual claim carries a link and a date. What survives is not an answer. It is a verdict.

Why this is not another research tool →

Not a summary. A posture.

Read one end to end: the verdict, the dissent printed in full, and every claim carrying a dated source. Both are method demonstrations, not predictions.

The memo is the output. The desk is the product.

For retained firms we build a private evidence desk: your thesis, your process, your past deals and passes, the lines you have decided you will not cross, indexed into a siloed library that never touches anyone else's. Over time the table stops arguing against the public record alone and starts arguing against your own history and your own rules: the pass that got away, the pattern you swore you would not repeat.

That library compounds with every memo. And it is yours: if you leave, it leaves with you.