Our Own Brands, Measured in Public

The rules every case follows

Written before the first measurement, applied to every brand the same way.

• Real numbers only. Every figure on these pages is read from the product, and a zero is shown as a zero.
• Headline numbers come from 10 run tests: each prompt is asked 10 times on each engine, and the share of runs that name the brand is the mention rate. One-off spot checks are shown apart and never as a percentage.
• Baselines measured before a brand was in the product are imported with the engine version of every run and drawn as their own segment. When an engine changes its model, or the control prompts move by more than 10 points, that period is marked as an engine change and not credited to any fix.
• Every fix gets a dated entry in the fix log, with the page link and commit where there is one. The case claims all changes since day 0 as a set, never one fix at a time.
• These are our own brands. No customer is named here without their written consent.

The numbers refresh from the product every time a page loads. The method behind them is on the methodology page, and the how it works page shows what you get when you run the same measurement on your own brand.