Memo lab
A readout a PM will sign
Slide decks are where feature-test decisions go to hide. Variant Readout Desk scores a one-page memo instead. If a designer cannot repeat the decision in a hallway, the page is not done — regardless of how pretty the confidence interval looks.
We did not invent one-pagers. We did insist that app analytics for feature flags end in a signature, not a screenshot. The template below is the one we use in the Phuket room. Steal it. Then put a name on the decision line.
The five blocks
Decision. Ship, hold, or kill — one verb. “Lean ship pending more data” is not a verb we accept. If you need more data, you have not finished the runtime you declared.
Evidence. Primary metric, MDE, observed delta, sample, runtime. Four numbers and a date range. Guardrail result sits on the same line, not below the fold.
Caveats. Joins you lacked, holidays you excluded, segments too small to claim. This block is where honesty lives. Students who leave it blank fail the lab even when the primary moved.
Owner. A person, not “growth.” Someone who will still work here when the holdout ends.
Holdout. What you still owe finance after you ship. Fourteen days is our classroom default, not a law. Write the window. Write what would make you roll back.
Charts stay in the appendix
We are not anti-chart. We are anti-leading with a chart. The first screen a busy PM sees should be the verb and the four numbers. Funnel plots, CUPED notes, and segment cuts can follow. If they contradict the header, the header was wrong.
Remote seats sometimes turn the one-pager into a Notion page with twelve toggles. We ask them to export to a single screen before they submit. Toggles are how caveats disappear.
A mild warning
This structure will make some experiments look smaller than your last all-hands slide. That is the point. A readout a PM will sign is allowed to be boring. Boring is how features stop lying.