ChatMaestro · Scenarios · Experiment lifecycle

Monitor a Live Run

Experimenter  Watch a run as it happens, drill into one enrollee, and pause, resume, or abort if you must.
Screens  Live-monitor experimenter-monitor.htmlWatch-drawer watch-drawer.html
Reads  the run experiment_runwho's in it run_enrollmenteach exchange chat_round
Key fields  status experiment_run.statepresence run_enrollment.last_activity_atwithdrawn withdrawn_atlive scores chat_round.score_composite

1 · Given / When / Then

Human wording first; the exact table/field in mono.
Given
A run exists with state = running (or paused), and enrollees have begun chatting.
When
The experimenter opens the Live-monitor Screen, picks the run, and watches progress per enrollee. The roster shows two counts: everyone enrolled, and the attendees — those taking part right now, meaning not withdrawn and recently active, derived live from run_enrollment.last_activity_at rather than stored. It also shows how many exchanges each has produced and the scores landing on chat_round. They can open the Watch-drawer Screen to follow one enrollee's transcript, or intervene by pausing, resuming, or aborting the run.
Then
Reading changes nothing. Intervening moves experiment_run.state: running → paused (and back), or → aborted. Letting it finish takes it to done, which is what triggers the run's scorecard (run_result) to be built.
reads experiment_run, run_enrollment, chat_round  writes experiment_run.state (only when you intervene)
Variations and exceptions
If…Then…
They try to clear a run that is running or pausedRefused: clearing is allowed only once the run is done or aborted (Clear a Run).
They remove an enrollee from the runrun_enrollment.withdrawn_at is set. The enrollee stops receiving the condition; what they already produced is kept and still counts.
They pause the run and try to launch another onto the same peopleRefused: paused counts as live for the disjoint-cohort rule.
They abort the runstate = aborted; its exchanges remain until the run is explicitly cleared.
Acceptance test — Given a running run, When the experimenter pauses then resumes it, Then experiment_run.state goes running → paused → running, no exchange data is lost, and clearing remains blocked throughout.

2 · The Journey, Screen by Screen

Watching is read-only; intervening changes the run's status.
Open the live monitor Live-monitor Screen Pick a live run state = running See progress per enrollee run_enrollment · chat_round Zoom into one enrollee Watch-drawer Screen Need to intervene? Pause / resume / abort state changes Let it finish state → done scorecard is built observe yes no

3 · In Plain English

While a study is live you mostly just watch. The monitor shows who is attending right now versus merely enrolled, how much they have done, and how the answers are scoring as they land.

If you want detail on one person, you open their drawer and read the actual conversation. If something is wrong, you can pause the whole thing, resume it later, or abort it outright.

Two things are deliberately not possible while a run is live: you cannot delete it, and you cannot start another study on the same people. Both protect the data you are in the middle of collecting.

Fits the schema cleanly. Every status change is a move on experiment_run.state — the full state machine is in Run a nudge-comparison series.