pro
Ishaan Ghosh
AI governance — policy enforcement, audit, claims review
New to Contra
Ishaan is ready for their next project!
AI governance claims review. Every technical claim in a system, and in the
materials going to a regulator or an investor, put through four examinations:
what would you observe if this were false; is the architecture as described the
architecture as built; what is enforced versus what is merely recorded; and what
the materials do not say. One report — what was examined, against what criteria,
what I found with the evidence quoted, what I did NOT examine, and what the
report does not establish. Interim findings the day I find them.
0
16
Execution Coverage Reconciliation — a specification for reconciling what a
governance system decided against what actually executed. Jointly authored,
CC BY 4.0. Arrived at by argument; neither author has implemented it, and the
paper says so. DOI 10.5281/zenodo.21843543
0
27
Keeping Score — on measuring which of your own claims turned out to be wrong,
and who found them. Published as a range and a floor rather than a single
figure, because the classification underneath it will not carry a point
estimate. DOI 10.5281/zenodo.21977571
0
24
Shango MID — the published transparency record. The evidence page for my own
write-governance layer: what it does, what it does not establish, and every
defect found in it, dated, with who found it and what changed. There are more
things on it the software cannot do than things it can. That is deliberate — a
control whose limits you cannot inspect is a reassurance, not a control.
shango.in/transparency
0
33