CleanstoneDevelopmental and independent evidence for machine minds.

Adversarial registry: direction questions

Q1 open Neutrality of the executor

An institution that executes an entry while holding claims of its own can be accused of bias. Default: for the first pairing we ask Eleos to host or co-host and to name a reviewer. The institution executes only entries on which it has made no prior claim. Its own claims go in as ordinary entries, executed by others where possible.

Q2 open Camps may decline to commit to numeric thresholds

Cogitate took years of negotiation to agree its format. Default: if a camp declines to name a threshold, the template accepts one proposed by the executor and records it as such. A camp that declines both is listed as "did not commit".

Q3 open IIT cannot be paired on software claims

IIT's position (Findlay, Koch, Tononi 2025) is about the physical substrate. It yields no divergent prediction about a language model's behaviour or activations. Default: document it in the registry as an open problem and do not force a pairing.

None blocking as of 2026-09-30.