Name the figure you are questioning and where you saw it. Every number here has a source, so we start from the same one. No attachments, except the two named below.
Two ways to reach us, email and a public issue
Write to contact@crusetra.com: the direct channel. For anything worth deciding in the open (a disputed figure, a wrong line, a question others will have), open an issue at github.com/ArslaneSempai-ui/crusetra-routing. An issue is public, so keep confidential material out of it: no client names, no records, no internal figures. Either way, attach no file, except the list you send for a Screening report, or the measured record of a Routing free test. For anything else, a channel for records is opened in the engagement letter, before signature, like every other term.
What to bring
The page you are on and the figure you dispute: each one here has a declared source, so we can meet at it. If you disagree with a measurement, bring your own rerun. The record is frozen and the command is public, so you can reproduce it. If you measured your own cases, bring the report the tool wrote beside your CSV: it contains rates and their confidence intervals, and no values from your documents.
What to keep out
We receive none of your documents, and we do not want a sample. Two things do reach us by email when you ask for a report: the list of names for a Screening report, and the measured record of a Routing free test, which holds verdicts and counts and no value read from a document. The tool exists so your documents never leave your machine. A sample attached to a public issue is published, and a clone, once made, cannot be taken back. If your question needs your data, the answer is a command that runs where the data lives, and we will point you to it.
Before writing
Two shortcuts. On a fresh clone, node src/premiere-reponse.mjs prints the conclusion in under a second, with what would flip it. And the eight objections a bank's reviewers actually raise are already answered, mechanism by mechanism, in Appendix C, Questions below.