docs-dead-end-hunterlisted

# docs-dead-end-hunter Traffic tells you how many arrived. This tells you **how many left with nothing — and which page to fix.** A *dead end* is a visit where the reader demonstrably tried: they searched, asked the assistant, or opened three or more pages — and still produced no sign of having got what they came for. It is the closest thing to direct evidence that documentation is failing, and unlike bounce rate it never counts someone who never engaged in the first place. ## Before you start Read `metrics/metric-dictionary.json` for the metrics listed in `metadata.measures`. Every number below has confounders that will make you give confidently wrong advice if you skip them — in particular: - **A high exit rate is not a problem on its own.** A page people leave from *after succeeding* is a terminal success page. Recommending a rewrite there makes the docs worse. Always read `exit_rate` next to `dead_end_rate`. - **Rates below ~30 visits are noise.** The product withholds the percentage and returns counts instead. When it does, say "not enough data" and report the absolute number — never quote a percentage it declined to give you. - **Visitors are hashed IPs.** Shared networks merge readers, mobile networks split them. Report direction of travel, not headcounts. ## Workflow 1. **Establish the headline.** Get the visit-outcome breakdown for the period: dead-end rate, self-serve resolution rate, and the raw counts behind them. If the sample is too small,