Scheduled Oscar Queries
Ask Oscar at /assistant to repeat a research question on a schedule.
Date: 2026-09-30 Purpose, access and entry Ask Oscar at /assistant to repeat a research question on a schedule. Signed-in users manage their own schedules through conversation; the administrative schedule dashboard is not a general-user screen. This differs from technical Watch Agents and evidence-based thesis monitoring. Inputs and workflow Specify the question, daily/weekday/weekly cadence, local time, timezone and (for weekly runs) day. State the desired language and optional notification condition. Inspect Oscar's creation receipt: question, interpreted schedule, condition and next run. A conversational request alone is not confirmation that creation succeeded. Ask to list schedules; use the returned ID to edit, pause/resume or delete the intended schedule. Ask to run a schedule now when an immediate background run is needed. “Started” means queued/running, not that an answer exists; the recurring next run is unchanged. Read delivered results and available run history. Delete only if you also intend to remove that schedule's run history. Manage in Settings Open Settings → Notifications → My scheduled queries at /settings?tab=notifications. This is the user-facing management card, separate from the admin dashboard. Review the question, cadence, time/timezone (for non-hourly schedules), notification condition, run/notification counts, last run and any last error. Use the switch to pause/resume. It is disabled while an update is pending; successful updates refresh the list and failures show an error toast. Select Show recent results to expand the latest five runs, or hide them again. Inspect completed/failed/running status, condition-met status, answer preview and error text. Only one schedule is expanded at a time. Use the trash control only when ready to delete: it sends deletion directly, without a separate confirmation dialog. It also removes run history. Wait for the success/error notification. Return to Oscar to create a schedule, change the question/timing/condition or request Run now; this card has no create/edit/run-now form. The card shows skeletons while schedules load. An empty list invites creation in chat; no runs means no assessment yet. The current card does not provide a dedicated list/run-fetch error panel, so an apparent empty result after a fetch failure is not proof that no schedules/runs exist. Outputs and limits Hourly cadence is also supported, at the selected minute each hour; it is not limited to market hours. Up to 10 schedules per user; questions are limited to 2,000 characters. The default timezone is America/NewYork, so explicitly specify yours. Daily/weekday/weekly timing is not the same as a trading-calendar rule. Scheduler availability and workload can delay execution; this is not an exact-second guarantee. Results are evaluated against the optional notification condition; without one, runs notify by default. The in-app bell and preference-controlled push/email are delivery channels. A condition not met is not a failed research run; delivery failure is distinct from assessment. An unreadable/failed condition evaluation does not establish that the condition was met. Headless runs cannot create/list/edit/delete/trigger schedules or invoke research operations. Navigation, language/theme changes, interactive watchlist operations and several podcast/UI tools are excluded. Do not use unattended prompts as a substitute for interactive approvals or assume Committee availability matches interactive chat. Loading, errors and empty states No schedules means there is nothing recurring yet. Running is not completed; errors are not empty answers. Verify status after creation/update errors and use IDs to resolve ambiguity. Paused schedules do not execute. Deployment, scheduler enablement and provider health were not exercised in this audit.