Every note, by release
Lab notes and engineering notes ship together, each with the rule book and worked examples it relies on. Newest first.
Engineering Note 1, with its rule book and case studies
Marked to market: how we decide our AI is ready to ship
How we evaluate financial AI through the evidence behind its answers, the outcomes of its calls, and the failures we find.
What gets a verdict
The evidence a run must carry, the checks it must clear, and the limits of a passing verdict.
What the record caught
Five cases that explain our evaluation rules, from a grader's false pass to the boundaries of client preferences.
The launch: why the Lab exists, the season plan and the rubric
Mark it to market
Why the Lab exists. How we will be graded. Start with the launch note, then read the season plan. The rubric and worked examples are linked inside.
What we will publish and how we will be graded
The earnings-season test: what we record, what we compare against, and how we separate a useful business judgment from a correct price call.
What gets a grade
The exact clocks, score bands, denominators, baselines and six checks on the business analysis. Open decisions remain labeled.
What each system said
Meta and Equifax, with United Rentals and CBRE as supporting cases. Original expectations beside the outcomes, including the limitations.