a builder's codex
codex · patterns · Measure with absolute counts and correlated short signals, not stage rates and long loops

Measure with absolute counts and correlated short signals, not stage rates and long loops

Convergence

Three operators converge on a measurement reset: stop optimising stage conversion rates, stop pretending you have long feedback loops, and stop running experiments that won't reach sample size. Optimise absolute counts at each stage. Find correlated short signals for the long outcome you care about. Use pre/post when sample size won't cooperate.

Operators

Variation

Implication

For every experiment you run, ask: (1) will it reach sample size in 30 days? if not, run pre/post. (2) am I optimising rate or count? optimise count. (3) is the outcome I care about months out? find a correlated 7-day proxy and validate it. Audit one quarter of "wins" against a 1-year holdout and expect 30-40% evaporation.

Sources

Open the interactive view →