fix(docs): correct sampling and restart-cost claims to match implementation

docs/manual/runs.md

@@ -36,7 +36,7 @@ By default the installed app stays in place between runs, so whatever the previo
 
 sanderling does not restart the app every few steps. Restarting throws away two things.
 
-**Exploration history.** Action selection favors states the run has not seen. A restart resets that history and the explorer re-treads the same early screens.
+**Accumulated data.** Accounts created, items added, caches warmed, settings changed. Interesting bugs live in apps with history, and a restart wipes it.
 
 **Deep app states.** Many bugs live in states that take many actions to reach: nested settings, a loaded cart, the screen after the third transaction. A 50-step path to "cart with 3 items" never happens if every run starts cold.
 

docs/manual/writing-specs.md

@@ -317,7 +317,7 @@ Weights declare testing intent:
 - `doubleTaps` gets explicit weight because rapid double submission is a failure mode these forms must survive. This is the line that flushes out the double-submit bug.
 - `defaultActions` keeps a quarter of the budget on random exploration, so the explorer still wanders everywhere and types edge-case values into every field. Without it, the spec only tests the flows you thought of, which defeats the point.
 
-A generator that returns an empty list drops out of the sampling for that step, so weights only ever distribute across what is currently possible.
+When a sampled generator returns an empty list, the runner re-draws, up to 16 times per step. So weights state preference, and the step still lands on a generator that currently has something to offer.
 
 ## The finished spec