/selection-toolkit — the instrument the panel scores in

One job: a selection has to be run, and the panel needs the workbook that runs it. Between publication and award. /tor-procurement wrote the ToR and stopped; procurement-offer-review reads one offer once the bids are in. This builds what sits between them.

The rule that governs everything here

Nothing is invented. Every criterion, every weight, every threshold, every date comes from the source document or is reported as a gap for Ane to resolve. A default weight is indistinguishable from a published one once it is in a cell, and the panel scores against it for six weeks without noticing. validate_spec refuses a spec with a hole in it, and read never fills one.

Two rules to carry into anything you write for a user of this skill:

1. Scoring may only use criteria the source published. A criterion the panel wishes had been published is a lesson for the next ToR, not a column in this workbook.
2. Once distributed or scored in, these files are edited in place, never regenerated. build refuses to overwrite an existing toolkit for exactly this reason. The scores in a returned scorer file exist nowhere else.

Where the code lives

All Office work is in ane_package. This skill contributes judgement, not plumbing.