docs(manual): plain-language rewrite of runs page

docs/manual/runs.md

@@ -2,11 +2,11 @@
 title: Runs
 ---
 
-# What is a run?
+# Runs
 
-One `sanderling test` invocation. Fresh install, spec-driven exploration, then the trace lands in `runs/<timestamp>/`. Typically minutes to hours, not seconds.
+A run is one `sanderling test` invocation: launch the app, explore it under the spec for a fixed duration, write a trace. Runs typically last minutes to hours, not seconds.
 
-A run is not analogous to a unit test. A closer framing is: boot a fuzzer for an hour and see what breaks. Violations are recorded in the trace and exploration continues, so one run can surface many bugs.
+A run is not a unit test. The closer picture is: boot a fuzzer for an hour and see what breaks. A violated property is recorded in the trace and exploration continues, so one run can surface many bugs.
 
 ## Lifecycle
 
@@ -14,56 +14,53 @@ A run is not analogous to a unit test. A closer framing is: boot a fuzzer for an
 sanderling test --spec spec.ts --bundle-id com.example.app --duration 30m
   │
   ├── launch the app under test (pass --clear-data to wipe app data first)
-  ├── boot the sidecar, connect the agent socket
+  ├── boot the sidecar (or connect to Chrome on web)
-  ├── bundle the spec, load it into goja
+  ├── bundle the spec, load it into the JS runtime
   │
-  ├── step 0..N:  pause, capture state, evaluate properties, pick action, resume, dispatch
+  ├── step 0..N:  read state, check properties, pick and perform an action
   │
-  └── terminate when --duration elapses (or SIGINT)
+  └── stop when --duration elapses (or on Ctrl+C)
         └── trace written to ./runs/<timestamp>/
               ├── trace.jsonl
               ├── screenshots/
               └── meta.json
 ```
 
+The trace is written incrementally. An interrupted run is complete up to the step where it stopped.
+
 ## App state across runs
 
-By default the installed app is left in place between runs. Whatever state the previous run left behind (account, cached responses, onboarding completion) carries over. Pass `--clear-data` to wipe app data before launch and start cold every run. See [CLI reference](./cli/#sanderling-test) for the flag.
+By default the installed app stays in place between runs, so whatever the previous run left behind (an account, cached responses, completed onboarding) carries over. Pass `--clear-data` to wipe app data before launch and start cold every time. See the [CLI reference](./cli/#sanderling-test).
 
-## Why runs are long and linear
+## Why runs are long
 
-sanderling does not restart the app every N steps. Each restart throws away two things.
+sanderling does not restart the app every few steps. Restarting throws away two things.
 
-**Novelty and coverage signal.** The exploration strategy weights actions by whether they reach previously unseen state. Restarting resets that history.
+**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.
 
-**Deep app states.** Many screens take many actions to reach: nested settings, a loaded cart, post-checkout flows. A 50-step prefix to reach "cart with 3 items" does not happen if every run starts cold.
+**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.
 
-Long-linear trajectories find bugs that restart-based testing structurally cannot.
+Long runs reach states that restart-per-test approaches structurally cannot.
 
-## Setup cost amortizes
+## Setup cost is paid once
 
-Preconditions (login, onboarding, consent dialogs) are written as weighted action generators gated on extractors. See [writing specs](./writing-specs/#pattern-preconditions-login-onboarding). They fire only when applicable, so login happens once per run, not per step.
+Preconditions like login run through the spec's `setup` export (see [writing specs](./writing-specs/#getting-past-login)). They fire when their condition is unmet and go quiet after, so login costs a few seconds once per run, not once per test case.
 
-| Run length | Login cost | % of run |
+| Run length | Login cost | Share of run |
 |---|---|---|
 | 5 min | ~15s | 5% |
 | 30 min | ~15s | 0.8% |
 | 1 hour | ~15s | 0.4% |
-| CI: 3 seeds × 10 min | ~45s total | 2.5% |
-
-At any non-trivial run length, preconditions are a rounding error.
 
 ## Session state
 
-Session tokens, keychain, shared prefs, cookies, and other app-managed persistence survive the full run. If the app logs the user out mid-run, the `doLogin` generator re-fires automatically because its gating extractor (`onLoginScreen`) becomes true again. No retry logic. No special-casing.
+Session tokens, keychain entries, shared preferences, and cookies survive the whole run. If the app logs the user out mid-run, the gating extractor flips, `setup` re-engages, and the run logs back in. No retry logic needed in the spec.
 
 ## Termination
 
-A run ends when either of these happens.
+A run ends when:
 
-- `--duration` elapses.
+- `--duration` elapses, or
-- The process is interrupted (SIGINT).
+- the process is interrupted (Ctrl+C).
 
-The trace is written incrementally, so an interrupted run is still fully inspectable.
+Additional conditions (`--max-steps`, `--exit-on-violation`, hard crash handling) land in [v0.1.0](https://github.com/priyanshujain/sanderling/issues/4).
-
-Additional termination conditions (`--max-steps`, `--exit-on-violation`, hard crash handling) land in [v0.1.0](https://github.com/priyanshujain/sanderling/issues/4).