> For the complete user guide index, see [llms.txt](https://portraitresearch.com/documentation/llms.txt).

# Create a scheduled research task

[Canonical page](https://portraitresearch.com/documentation/scheduled-research/create-a-task/)

## Applicability

- **Task:** Save a research instruction, choose a cadence and timezone, authorize it, and verify activation.
- **Access:** Scheduled Research enabled; eligible usage-based account; sufficient credits
- **Audience:** Portrait users and support teams
- **Surfaces:** web
- **Availability:** Account-dependent; see requirements and limits
- **Content type:** how-to
- **Last reviewed:** 2026-10-08

Create a scheduled task when you want Easel to repeat a defined research job. For example: “Every Monday at 09:00 Europe/London, review these companies for material developments since the previous run; cite the sources and email me a concise report.” A task can also run just once at a later time, such as “in four hours, check this screen and send me the results”.

## Create and activate

1. Open **Scheduled** in the web app and choose **New scheduled task**. Describe the job and choose **Set it up with Easel**, select a starting template, or open **Prefer to build it manually?**. A template opens setup; it does not activate a task. State the company scope, sources, period, result, and cadence explicitly.
2. Review the saved task name and instruction. Use a distinct name and specify how to handle an empty result or unavailable source.
3. Under **Runs**, choose the trigger from the table below and review its company scope. For earnings or document events, use a fixed list of up to 25 companies or follow a Project's current companies. For calendar tasks, name the intended companies or Project in the instruction. Check the local time and timezone; new tasks default to your browser's zone.
4. Choose **Email me**: **Never** keeps successful results in run history; **When it matters** notifies only when Easel finds something worth your attention; **Every run** requests a notification even when nothing changed. Failures, cancellations, and one-offs have separate notification rules. The choice governs both Inbox and email. Earnings and event tasks also offer **Deliver → Each company separately** or **One summary per day**, with an optional **Summary brief**. Company runs then stay quiet and the day's summary handles delivery; a day with one run delivers that run directly. With **Never**, the summary stays in run history.
5. Review the proposed work and usage. **Create scheduled task** saves and enables the task in one step, which books its credit commitment. If authorization is required, complete it when prompted; on the task page, use **Enable this research** or **Verify to enable**, or **Restart this research** after a revoked connection. A task whose enable did not complete stays a **Draft**; use its **Enable** action in the list.
6. Return to the task and confirm that its status is **Active** (**Pending** for a one-off) and the next scheduled time matches your intention. If authorization completed but the task is still **Draft** or **Paused**, use **Enable** or **Resume**; a connected account alone does not prove that this task is active.

**Settings → Connections** shows the standing Scheduled Research connection for eligible accounts. Start or restart authorization from the task when prompted. A missing connection card can reflect account access; the support assistant cannot inspect or enable it for you.

| Trigger | What starts a run |
|---|---|
| **On a schedule** | Daily, weekdays, or weekly at a saved local time |
| **Around earnings** | A separate company run on, before, or after its earnings date; disable estimated dates when confirmed dates are required |
| **When something happens** | A new earnings-call transcript becomes readable. Automatic runs support `Conference_Calls.Earnings_Call`; other document types, Monitor-update events, and external events remain draft-only |
| **Once** | One chosen date and time, at most 60 days ahead |

## Timezones and clock changes

The schedule uses local wall-clock time in its named IANA timezone, such as `Europe/London`, rather than a fixed UTC offset. Check both the zone and the next due time when travelling or changing a schedule.

For a calendar schedule, if the selected time does not exist when clocks move forward, that occurrence is skipped; an earnings run at that time runs an hour later instead. If it occurs twice when clocks move back, it runs once at the first occurrence. A daily 02:30 task therefore does not run twice solely because the clock repeats that hour.

A transcript's arrival and a date after earnings are different triggers. Choose the one matching your intention; enabling both can duplicate work.

## One-off runs

A one-off starts at the time you choose, at most 60 days ahead, and results arrive once the research finishes. It always notifies you when it finishes, and it has no **Run now**. Until then it shows as **Pending**. After its run starts, the task shows as **Completed**; to run it again, open it and pick a new start in its settings, which reserves credits for the new run the same way enabling did. Write its instruction so it stands on its own, naming the screen, companies, or documents it needs, because the run does not see the conversation in which you asked for it. Deleting a one-off returns the credits reserved for a run that has not started, and a one-off that reaches its time but cannot start returns them on its own; runs that already happened stay charged, as do credits reserved in an earlier usage period.

## Common task shapes

Starting templates include **Morning note**, **Earnings preview**, **Earnings post-mortem**, and **Weekly note**. Review their proposed coverage, timing, sources, and deliverable before activation. A post-mortem can use a matching pre-earnings packet when available; missing evidence can leave the answer partial.

To keep a file current, open it in **My Library** and choose **Schedule research** where available. Describe the changes and preservation requirements. This path follows the same item's latest accessible version across renames; a missing or inaccessible file is a blocker, not permission to substitute another item. A pending browser-local workbook adjustment must be saved before a scheduled run can read it. Verify the saved file and task's run history after execution.

A run can read supported filings, transcripts, news, consensus, prices, your library files and skills, named Projects and Monitors, and its earlier runs. It cannot ask you questions, send ordinary email, or use connected apps such as Outlook. Notifications use the task's delivery policy. Specify enough scope and recovery instructions for the work to proceed without you.

## Verify the result

Open the task's run history after its due time, or use **Run now** on a recurring task to preview the output before the first scheduled run. Review the run state and answer, its effective research period, and its citations. A manual run performs work and can consume usage; do not use repeated runs just to test whether email arrives.

## If activation fails

Read the visible message before retrying. Missing feature access, ineligible usage plans, missing authorization, incomplete instructions, and insufficient credits need different fixes. See [Notifications and troubleshooting](https://portraitresearch.com/documentation/scheduled-research/notifications-and-troubleshooting/).