A · The 59, scoredtap for all 59 ⊕
Where the 59 landed — and the tail
Each scored 0–12 on how firmly it keeps a human accountable for what a machine can't prove. Most land high; a short tail breaks rank.
B · The linetap for their words ⊕
The line every institution drew
Every one lets AI fix your grammar. Not one lets AI sign the work. A readout of what they permit — in their own words.
Cheap to check →where they let machines in
- grammar & copy-editing
- translation
- formatting & references
- search & summary
- writing code
Expensive to check →where they kept a human
- authorship
- judgment & interpretation
- original reporting
- the byline
- accountability
C · Segment gradienttap to expand ⊕
Trust relocates as proof gets costly
proofprocess
publisherssocietiesjournals
universitieslibrariesed-techintergov
Judge a checkable artifact — a paper, a citation — and proof is cheap. Trust rides on the work.
Judge a diffuse outcome — access, equity, governance — and there's nothing to inspect. Trust rides on who's accountable.
AI widens the gap. Once a machine does the work, the reasoning is obscured inside the model — so trust slides from proof to process.
D · Dissent & deviationtap to expand ⊕
The honest edges
46 drew the line · 8 drew it partly · 5 drew a different line
EDUCAUSE CA 3 · the lowest score in the corpus
Found AI-detection doesn't work — so it moved trust out of proof and into process: who's in the room when the rule gets written. — the Index's reading of the policy, not a quote
When proof becomes impossible, trust doesn't vanish — it becomes governance.