Scheduling Recurring Audits

Setting a schedule

answer_schedule_set takes a project id, a cadence, weekly or monthly, an enabled flag and a timezone. Every field is optional and an omitted one is left alone, so pausing is enabled false on its own. The response's disclosure lists the cadences this account may pick; daily is restricted to internal accounts. Enabling does not start a run now: the first scheduled run fires one cadence from now, at the next_run_at in the response, in the local hour you chose, preserved across daylight-saving changes.

What a scheduled run does and does not do

It runs the project's approved buying questions across the enabled engines exactly as a manual audit does, reserves budget first, and is refused by the same per-run and daily caps. It appears in History like any other run. It does not add prompts, change samples per question or bypass a cap because it was scheduled.

Missed slots are skipped, never replayed as a backlog. If the account was over its daily allowance when a slot came due, that slot is gone; the next one fires on cadence.

Pausing, resuming, and plan changes

Pausing writes only the enabled flag and clears the next slot, so a resumed schedule does not fire the instant it comes back. The cadence and timezone are kept, which is what makes 'paused' and 'never set up' distinguishable and makes resuming one call with enabled true.

When a subscription stops including Answer Radar, the app pauses every schedule on the account the same way, so a churned account cannot keep spending. Winning the account back does not automatically re-enable them, because nothing records who turned a schedule off and restoring blindly would also switch on projects you paused yourself. Resume the ones you want.

Why did my scheduled audit not run?

answer_activity_list records what Answer Radar did on the account and what it refused: scheduled audits that were skipped and why, budget reservations that were denied, runs created and charged. A skipped slot is an action of schedule.dispatch with outcome denied, and evidence mode carries the reason and the slot it was for. Events are attached to what they acted on, so filtering by project returns its scheduled skips.

Absence is not evidence: each action is recorded only from the date its writer shipped, so an empty result does not prove nothing happened. Newest first.

• Denied dispatch: a cap or the kill switch refused the slot.
• No dispatch recorded: the schedule was paused, or the slot predates activity recording.
• Run present but partial: the slot fired and some samples failed.

FAQ