Run history.
Every comparison run is published as its own page and never edited after publishing. New runs are added on top; older runs stay exactly as they were. This is the performance-over-time log — the comparison overview reflects the latest run, while everything below remains as history.
Runs differ in scope (number of tests, synthetic vs real account, which setups). Each page states its own scope and caveats, so a number from one run is never mistaken for another.
Runs
Bare Claude vs AdTAO — same live account
Scope · First fair head-to-head · per-test evidence
Both setups reason over the SAME real Google Ads account. AdTAO: 4 wins / 1 tie / 0 losses. Every test on this page shows the prompt, both setups' actual answers, the scoring rubric, and the per-test winner — including the tie where bare Claude got it right too.
AdTAO read suite on a live account
Scope · 28-test capability log · per-test prompt + tools + answer
AdTAO's actual work on every test in the read suite: 22 grounded, 4 asked for specifics, 1 no-tools, 1 timeout. Each entry shows the exact prompt, the live tools called, and the real answer (withheld bodies for cross-MCC tests preserve client privacy; tool calls still shown).
Cross-bucket synthetic smoke
Scope · Synthetic fixtures, 4 setups, 5 buckets
First cross-bucket smoke — small, synthetic, kept as history. The fabrication test (niche-segment benchmark) is the durable contrast that remains evergreen on the comparison overview.
Newest first. Each entry is immutable — when results change, a new run is added on top.