Frequently Asked Questions

Can I reproduce a finding myself?

Yes. Every finding names the source, the exact bytes loaded by sha256, the source version, and the contributing row numbers. A human reviewer must be able to rerun the computation by hand.

What happens if a supporting document is missing?

The claim becomes UNVERIFIED, never a red flag. Removing a document cannot produce a finding. Missing data is never guessed.

Does it use an AI model?

Not in the core path. The engine is deterministic and standard-library only, with no clock, no randomness, no network and no model call, so the same input yields byte-identical output every time.

Is this an audit?

No. PreDealCheck is an arithmetic reconciliation of stated claims against the source table each claim cites, prepared for a qualified human reviewer. It is not an accounting certification, a legal opinion, or an investment assessment.

Does it tell me whether to do the deal?

No. PreDealCheck does not provide investment advice: no buy or sell recommendation, no valuation, no statement that a deal is safe, and no fee contingent on a deal closing or failing. A human reviewer retains responsibility for assessing legal, technical, and business materiality.

Why does it refuse to compute churn?

Period churn needs a starting cohort plus period events. See methodology for the full explanation.

Can it read a PDF or a data room?

Not today. PreDealCheck takes a bounded CSV or JSON manifest—claim records linked to the tables that support them. Document extraction is deliberately not built yet.

My number was right but it was not confirmed. Why?

Because the cited table measured a different period, basis or currency. Matching numbers across different definitions prove nothing, so PreDealCheck asks the question rather than confirms, marking the gap as DEFINITION_MISMATCH.

Is it better than an analyst with a spreadsheet, or than ChatGPT?

Unknown; that comparison has not been run. The blind comparison that would settle it is: the same case reviewed by a manual analyst, by ChatGPT plus spreadsheet, and by PreDealCheck, with independent measurement of errors, time, corrections, and checkability. This test has not been executed.

Where does my data go?

The engine makes no network call and runs locally on your machine. This limits data transfer but does not by itself make a deployment compliant; data-handling terms must be agreed before any real material is processed.

Who is using it?

Nobody yet. PreDealCheck has no customers, no paid pilots, no revenue. Tests are the project's own technical checks on fixtures, not independent acceptance or certification from any customer.

What does it cost?

Visit the pilot program for the first price experiment. That figure is a hypothesis under test, not a market rate.