AUG 14, 2026 · 3 MIN READ
The Result That Inverted My Own Leaderboard
A clean safety result turned out to be a bug in my harness. Finding it flipped the rankings and cost me a published conclusion.
Stories
(sorted by newest)5 posts
JUL 2, 2026 · 2 MIN READ
Zero-Shot Beat Everything, and I Published It
Eleven conditions of chain-of-thought prompting and fine-tuning on table reasoning. The baseline won. Here is why that is worth reporting.
JUN 10, 2026 · 3 MIN READ
What "Compliant Agent Behavior" Actually Means
Turning EU AI Act obligations written for organizations into three dimensions a system can be scored against.
MAY 20, 2026 · 3 MIN READ
Building an HTAP Storage Engine From Scratch
Transactional and analytical workloads want opposite storage layouts. Lineage-based columnar storage serves both without keeping two copies of the truth.
APR 18, 2026 · 3 MIN READ
Taking API Cost Out of the Iteration Loop
When every experiment costs money, trial count becomes a budget decision instead of a scientific one. Moving the stack offline fixed that.