Skip to main content

Overview

The bigdata_get_monitor_runs tool reads the runs of a search monitor and the events each run found. It reads one of three things, depending on what you pass:
  • History (no simulation_id): the monitor’s past runs
  • Flash simulation (simulation_id and mode: "flash"): the result of a flash preview
  • Full simulation (simulation_id and mode: "full"): the replayed runs of a full simulation
Start simulations with bigdata_simulate_monitor.

When to Use

Get Monitor Runs is ideal for:
  • Reviewing a preview: Reading a flash simulation before activating a monitor
  • Backtesting: Reading the runs of a full simulation over a past period
  • Checking alerts: Reading what an active monitor found on its recent runs

How It Works

  1. Start a simulation with bigdata_simulate_monitor, or skip this step to read history
  2. Call bigdata_get_monitor_runs with the monitor_id, plus the simulation_id and mode it returned
  3. Poll a full simulation every 10 to 15 seconds until simulation.phase is completed or failed. A flash read waits inside the call

Parameters

What each read accepts: Flash simulation
History

Important Notes

  • Reading with the wrong mode returns an error that names the right mode, once a worker has had about 60 seconds to pick up the job
  • A flash read waits up to 25 seconds inside the call, checking every 3 seconds
  • An unknown or expired flash simulation_id returns a not-found error. Previews expire after 7 days, or after 30 minutes if they never started
  • A flash simulation that failed before reading any news returns an error, not an empty page
  • History includes full-simulation runs, marked is_simulation: true. They are replays, not alerts that were sent
  • If the page of runs is too large, the call returns an error. Lower limit or set include_events to false

Data Returned

Each run has: Each event has: simulation.phase goes accepted, seeding, running, and ends at completed or failed. Stop polling at either of the last two. A flash preview of the Anthropic – Regulatory Watch monitor, which found two events in the latest four-hour window. The second event’s grounding is cut to two of its eight sources:

Practical Tips

Leave out novelty for flash

Flash previews do not judge novelty, so leave novel out when showing their events. A full simulation includes events rejected as already seen, marked novel: false.

Simulated runs are not alerts

When showing history, separate runs with is_simulation: true. They were replays and were not delivered.