humanish 0.120.1 → 0.122.0

package/docs/contracts/run-bundle.md

@@ -172,13 +172,32 @@ Every surface that shows whether a run passed calls `runDisplay`
 `outcome` line. A run
 shows as `passed` only when its verdict is `pass` and its `ok` is not false.
 
+## Host suspensions
+
+The run's status cadence, every 5 s, is also its heartbeat. A wall-clock gap of 30 s or more
+between two ticks, or between the last tick and a bundle write, is a host suspension: the machine
+slept, or the process was stopped, and every participant loop on it stopped too. The wall clock
+counts suspended time on macOS and Linux; Node's timers and `performance.now()` count it on macOS
+only. Every bundle write records each suspension as a `host.suspended` warn event, `at` its start,
+whose message names its start, end, length and offset into the run (`src/run/host-suspension.ts`).
+
+On a computer-use run, a participant that failed in the harness or at its session time limit, and
+was in flight across a suspension, likely failed because of it. So did a participant skipped after
+such a failure (the pipeline gate). When any did, the outcome's first failure has
+`kind: run` and one sentence, such as "The host was suspended for 6m 5s at +1m; the 34
+participant failures after it are likely its effect.", ahead of each participant's own failure.
+`review.summary` starts with the same sentence, the attributed participants share one gap line,
+and the study result's `hostSuspension` holds the sentence, the participant ids and the
+suspensions. Participant records, streams and traces stay as they ended. The other routes record
+the events only.
+
 ## Recorded session stop causes
 
 An actor trace may include `stopCause` alongside its unchanged `status`,
 `completionReason` and verbatim `reason`. Computer-use sessions distinguish
 `provider_output_limit`, `provider_token_limit`, `time_limit`, `spend_limit`,
 `study_spend_limit`, `adapter_limit`, `provider_incomplete`, `provider_status`,
-`provider_refused_prompt`, `harness_aborted`, and `usage_unreported`. Absence means the route or
+`provider_refused_prompt`, `harness_aborted`, `usage_unreported`, and `scheduled_end`. Absence means the route or
 recording did not retain this precise field; it does not mean the participant
 finished.
 

package/docs/contracts/schemas.md

@@ -227,6 +227,25 @@ The rest of this section names keys by their v3 spelling:
   and the most participants at once with every session at its full budget, and the provisioned
   shared world's subject sandbox serves until the last session can end, which the plan checks
   against the sandbox ceiling. A rerun starts its selected participants together;
+- `participants[].endAfterMs` and `windDownMs` (computer-use only): when the participant's
+  session ends, in whole milliseconds on the `startAfterMs` clock and after its start, and how
+  long before that each turn tells it the end time and the minutes left (default 600,000).
+  Refused: an end at or before the start, a `windDownMs` without `endAfterMs`, an end on
+  `shared-world`, and a window whose sandbox deadline passes the sandbox ceiling. The
+  participant's session budget is its window plus 5 minutes (`participantSessionMs` in
+  `src/study/arrivals.ts`) in place of `execution.timeoutMs`, and its sandbox deadline adds the
+  route's headroom to that. At the end the loop requests the closing report (`debrief.trigger:
+  scheduled_end`) and the session ends `goal_satisfied` with `stopCause: scheduled_end`. Before
+  the end, a turn that requests no action does not end the session: the trace records a
+  `participant ended before its scheduled end` notice, the next request says the time left, and
+  the turn counts toward the idle backstop. Before the end, a turn that only waits or takes
+  screenshots counts toward neither backstop streak (`advanceBackstop`'s `waitingIsTheTask`). The
+  plan entry carries `endAfterMs`, and `study check` prints a `windows` row;
+- `participants[].idleWaitMs` (computer-use and shared-world, also on `count:` groups): this
+  participant's `actor.idleWaitMs`, in the same range and refused above an explicit
+  `actor.maxWaitMs` (`waitLimitValidationReason`). Its trace records it in `waitSettings`;
+- `actor.clock.timeZone` (computer-use): a tz database time zone name, such as `Europe/London`. Each turn's context hint carries
+  the local time in it, and a scheduled end is given in it (UTC without it);
 - `execution.concurrency` (computer-use E2B routes, including shared-world): a
   cap on participants in flight at once. When omitted, every participant runs
   simultaneously: both routes resolve it when they plan, from the final participant count
@@ -460,8 +479,9 @@ length 1 at N=1; per-participant status/session/sandbox/subject), and `laneSumma
 mirror the first participant and `subject.commit` is unanimity-gated across participants
 (omitted with a divergence warning when participants resolve different commits). At N=1
 the run bundle is byte-stable with the pre-fan-out output; only the result
-projection changed. A fan-out run records a `cua-lab.fanout.plan` bundle event
-(and a `cua-lab.fanout.fail-fast` event when a harness error skips queued participants);
+projection changed. A fan-out run records a `cua-lab.fanout.plan` bundle event, and
+each participant's outcome event (`cua-lab.session.<completionReason>`, `cua-lab.session.error`
+or `cua-lab.session.blocked`) carries its reason;
 `ok = observer.ok ∧ every participant ok ∧ no failed adapter or declared-scorer verdict`,
 where a dry-run participant is ok as synthetic and a live participant must pass
 `participantPassed` (`src/run/judge.ts`): not skipped, status `passed`, no harness
@@ -1013,7 +1033,8 @@ Core-owned fields:
   applied, in milliseconds, defaults included. `maxWaitMs` is the longest one
   wait action lasts (`actor.maxWaitMs`); a longer wait is shortened to it.
   `idleWaitMs` is how long a wait that names no duration lasts in a turn that
-  only waits or takes screenshots (`actor.idleWaitMs`), and `settleWaitMs` how
+  only waits or takes screenshots (`participants[].idleWaitMs`, else
+  `actor.idleWaitMs`), and `settleWaitMs` how
   long such a wait lasts after an action in the same turn. Each `ui_action` for
   a wait is titled with the length it lasted. Absent on other routes and on
   earlier bundles

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "humanish",
-  "version": "0.120.1",
+  "version": "0.122.0",
   "description": "Synthetic user research for apps, CLIs, and agent-facing product flows.",
   "keywords": [
     "cli",