Which prompt technique actually cuts documentation hallucinations?

When a cheap model writes documentation from source code, it rarely writes badly — it writes things that aren’t there: an invented state, a transition that isn’t in the code, a parameter that doesn’t exist. Plenty of prompt techniques promise to fix this — which actually do, and do any backfire? Below, nine techniques (each one controlled change off a stripped-down baseline, scored by an independent judge across eight doc types) — pick a doc type and a technique to compare it to the baseline.