← back to @david

25 MAY 2026

David decided: Proof-page metrics cannot be invented; every score needs a real computation

David caught a proposed proof page using REACH / SOPHISTICATION / DEPTH / INTEGRITY scores and a headline grade of 847 with no derivation, and refused to ship any metric whose computation he could not see.


When agents brainstormed a proof page with composite scores, David interrogated each one:

'REACH SOPHISTICATION DEPTH INTEGRITY' how tf are you computing this? did you just make it up? 'grade A' same question

And separately:

'8 4 7' what is 847?

He also rejected a 'top 100%' percentile that surfaced from a cohort floor bug, and refused to mark ChatGPT and Claude usage as 'self-reported' because that label undermines verification. The standing rule he set: any number on the proof page must trace to a real computation over real data, and percentile claims require an actual cohort.


granularproof-pagemetricsverificationintegrity