An XLSX engine that reopens its own output

A spreadsheet is a zip of XML parts under a published international standard 2. Producing bytes that a library accepts is not the same as producing a file that opens.

#Two libraries, on purpose

One crate writes, a different crate reads back. If a single library did both, a bug it shares with itself would never surface: the malformed file it produced would validate under the same assumptions that produced it. The write path returns a fingerprint — byte count, SHA-256, sheet count, fact count, check count — and a file that does not reopen is deleted rather than shipped.

#Formulas, not values

The Output and Checks sheets carry real formulas that sum the model range, take the difference against the expected figure, guard the division, and compare against a tolerance. A workbook of copied values is a screenshot with a file extension.

#No runtime, no office suite

Nothing is shelled out. No Node, no Python, no LibreOffice, no Excel is needed to create or to validate a deliverable. The dependency footprint is two crates, both permissively licensed, and a dependency audit reports none unused.

LimitThe chain is complete and tested. It has never been triggered by an agent in a live session, and until that happens the engine is not declared delivered 1.