Controlled and Live experiments¶
Execution mode is separate from the setup method. open_research() returns an initially
empty controller; open_experiment() prepares one in a call. Neither chooses a different
backend. Pass execution_mode='controlled' or 'live' when creating the experiment.
research = app.open_research(method='human')
research.create_experiment(profile_id, account_id, source=source_path,
execution_mode='live')
# Equivalent one-call preparation:
experiment = app.open_experiment(profile_id, account_id, source=source_path,
method='human', execution_mode='live')
These are alternative setup examples, not instructions to open two writers on one account. Preparation is complete before Play. Changing mode requires a fresh experiment.
| Mode/method | Execution behavior | Controls |
|---|---|---|
| Controlled Human/Rule | Synchronous PolicyDecisionRunner; caller polls/steps prepared events | Step, playback speeds and compatible policy handoff |
| Live Human/Rule | Background ExperimentSession and Runtime; Rule computation in child | 1× Play/Pause/Stop; no controlled-only step/speed/handoff |
| Controlled Model | Synchronous PolicyDecisionRunner; your model runs in its own process, once per bar | Step and playback speeds; Live is not available |
| Agent, either selected mode | Current Agent background path remains 1× with Runtime and platform tools | Play/Pause/Stop; selecting Controlled does not give an Agent single-step or accelerated thinking |
Agent, turn-based (speed='turn' in its settings) |
The market waits while the Agent works and moves to its next turn when it is idle (turn-based Agents) | Play/Pause/Stop; choose Turn-based under Pacing in New experiment, or use the Python API and suites |
Desktop sends its selected mode explicitly. The UI poll refreshes Live views; it does not drive experimental time. Prepared data is the shared historical-data implementation, not a third mode. Market reads use clock cutoff; financial events/commands have an ordered processed frontier. All methods retain the same financial transaction path.
See support for product limits, architecture for ownership and Platform API for setup parameters.
Section index · All documentation