Week 2 · after our feedback

What we added. Each change answers a problem from week 1:

Problem in week 1	What we added	What happened in week 2
No threshold	config/thresholds.json (63%)	She flagged both weak nights: Monday 60.4% and Saturday 62.7%
She did the math herself	Scripts to normalize, calculate and validate, and the rule "never calculate a number yourself"	Every number matches the reference, Friday included
No evidence	The rule "every insight cites the file and the row"	All 5 insights show where they come from
Two sources don't match	The rule "when two sources disagree, say so"	She warned that the group grid covers 1 of 25 blocks, that the export says 29 groups but lists 25, and that the event dates don't match
No fixed format	A format for the report and the insights, and a check before saving	Both files pass the check

She also got her map ("Where things live") and her first how-to, the skill normalize-pms.

Cost of this run: about 0.83 dollars.