Antonine
Antonine never asks a language model for a score. It asks the model for judgments, then computes the score from those judgments with a formula that does not vary.
The model reads the deal package and makes four kinds of call: whether the sector is aligned, adjacent, or outside your mandate; whether each financial criterion passes, sits borderline, or fails; whether each diligence question was answered substantively; and whether each named risk is clear or flagged. Every judgment carries its reasoning and its source.
Everything after that is arithmetic. That boundary is the product. It means a given score means the same thing in March and in October, two analysts working the same deal against the same profile reach the same number, and any partner can open a score and see exactly which judgment moved it.
Every deal starts from a neutral position and moves across four dimensions. Nothing else touches the score.
Whether the opportunity sits inside your mandate, adjacent to it, or fundamentally outside it. The single largest swing in either direction.
Each of your financial thresholds, graded pass, borderline or fail against the deal's actual numbers. Missing data counts against a deal, but far less than a clear miss.
Whether your standing questions were actually answered by the materials. A substantive answer helps; a question the package never addresses costs more than an answer gains.
Your named risks, each marked clear or flagged with a severity. A flagged high-severity risk is the heaviest single deduction in the model.
The first matching condition wins. Ambiguity routes to a person by design.
The deal is clearly outside the mandate, misses a threshold by a wide margin, or scores low enough that no combination of the remaining factors would recover it.
The highest bar in the model, and the hardest to clear. Score, sector, risk, diligence coverage and every minimum criterion must all clear simultaneously — any one of them alone is disqualifying.
Everything else — and deliberately the default. A deal that doesn't clearly fail and doesn't clearly clear is a judgment call, and judgment calls go to your team.
No threshold is a hard line. Each one carries a band either side, so a deal that misses your EBITDA floor by a hair isn't silently killed — it's surfaced as a judgment call.
A deal that misses your EBITDA floor by a hair lands in the middle band, not the left one. It costs nothing and gains nothing — it gets surfaced for a human read. Only a clear miss deducts.
A mid-market business-services deal screened against a private equity profile with four financial criteria and six diligence questions.
Every line is a judgment the model made and a weight the platform applied. If you disagree with the verdict, you disagree with a specific line — and each one links back to the page in the deal package it came from.
Sectors, financial thresholds, named risks and diligence questions are set by your firm, per asset class, and apply to every deal from the moment you change them. The relative weight of the four dimensions is tuned with us during onboarding.
The fastest way to judge the method is to point it at a deal you already have an opinion on, and see whether it agrees with you — and why.