Working principles
Claims, with their limits.
A clear record makes the work easier to inspect and improve.
Scope the claim
Experiments identify the environment, versions and scenario. An observed result is distinguished from an interpretation or an untested possibility.
Show the source
Where possible I link code, releases and evidence. A hash establishes byte integrity; a passing check supports its own assertions. Neither by itself establishes every property of a system.
Credit the contribution
Project pages describe my role and identify relevant AI assistance and ownership. A readback through my own account is not described as an independent third-party audit.
Keep corrections visible
A material correction receives a dated notice on the affected article. Its original publication date is retained. Historical failed results remain part of the research record where they are relevant.
Protect private material
Evidence is reviewed before publication. Credentials, private customer details and account-only records are not published just because an agent could access them.