Now
Last updated 2026-08-17
What I am working on now, newest first. The stamp above is the last time this page was edited, so you can see how stale it is without guessing.
Log
- 2026-08-14
OpenAddict
An independent LLM evaluation lab. Community techniques tested against pinned model versions across three vendors, with frozen scoring thresholds and null results published as first-class findings.
verified - 2026-07-27
SERP event registry
A primary-sourced registry of AI search events with a tiered attribution ladder and schema validation running in CI, so an entry cannot land without a source. Paired with a first-party volatility instrument on a frozen keyword cohort. Update note.
in progress - 2026-07-23
Lanthe
Basano's citation-measurement instrument for AI search. Demand-seeded prompt sets, citation share reported with Wilson confidence intervals, and write-once baselines so a number cannot be quietly revised after the fact. Update note.
in progress - 2026-03
RampStack releases
Open-source releases in flight, run through manifest-driven governance with human-only merge authority. The release process is itself the thing under test.
in progress Next verticals
PC parts and automotive, as the next test properties. Both are further from the existing properties than those are from each other, which is the point, because the generalisation test is only worth running on domains that do not share a vocabulary.
planned