Runs and review

A run is one Scribe job on one source file. You find runs in the Project's Scribe tab, which lists every run with its source file, workflow, routing rule, status, review state, model, and when it was triggered. Use the search box (by source file name) and the status and workflow filters to narrow it down.

Permissions

Viewing runs needs agent_run.view_results (all Project roles by default). Triggering, rerunning, commenting, and approving need agent_run.trigger (Project Editor or Admin).

Run statuses

Status	Meaning
Pending	Created and waiting for a worker to pick it up.
Running	Scribe is reading, drafting, and checking.
Completed	The draft and the internal check agreed. The output is in your output folder.
Needs review	The internal check couldn't fully settle after its maximum rounds. Output may still be attached; a person should look at it. See below.
Failed	The run couldn't produce output. The run page shows the reason.

The summary cards above the table show totals, runs in progress, completed runs, and runs that need attention (needs review plus failed).

How a run works

1. Read — Scribe reads the source file (and, for workflows that compare, the comparison baseline).
2. Draft — it fills in your output template, or builds a comparison, with a citation for every value pointing to where in the source it came from.
3. Check — a separate reviewer step checks the draft against the source and flags errors. Scribe corrects and the loop repeats, up to three rounds, until the reviewer has nothing left to flag.
4. Write — the result is scanned and saved to the rule's output folder in your storage, named after the source file with _scribe_output added.

Only the clean result reaches you. The back-and-forth is kept in the run's history if you want to audit it.

The output

• Word templates produce a Word document where Scribe's entries are tracked changes, each with a comment citing the exact place in the source document. Accept, reject, or edit them in Word like any other tracked change.
• PDF templates produce a PDF where added content is drawn in a distinct color on your template, with citation annotations. PDFs don't support accept/reject, so Word is the better choice if your reviewers want to work through entries one by one.

Nothing is final until a person signs off. Scribe suggests; you decide.

Open a run

Click a row to open the run page. It shows:

• Status, workflow, and review state at the top, plus an Org Managed model tag if the run used one of your own models.
• Routing rule (or who triggered it, for a manual run), Model, Output format, Source revision, Duration, and timestamps.
• Source file, Output folder, and Output file paths.
• Output preview — the generated document, rendered in place. It loads using your own Dropbox access, so you need access to that folder. If it can't load, use Open in Dropbox.
• Detail results for the workflow, collapsed until you open them: the reconciliation results for Certificate of Analysis Review, the prescreening checklist for Batch Record Review, and the draft/check iteration history.

Review and approve

At the bottom of the run page is the Conversation, a running thread for your team. Everyone with trigger permission can Comment, using Markdown if they like. Once the run has finished, they can also:

• Approve — record that the output is fit to use.
• Reject — record that it is not.

The decision is shown as a Review badge (Unreviewed, Approved, or Rejected) with who made it, and it is separate from the run status. A run can be Completed and still Unreviewed, or Needs review and later Approved.

When a run needs review

A Needs review run has an amber note explaining what the checker was still unhappy about. Open the output, work through the items, then use Mark converged if you're satisfied with it or Mark failed if it isn't usable. Both are recorded in the thread.

Start a run by hand

You don't need a routing rule to process a single file.

1. In the Scribe tab, click Trigger a run.
2. Choose the Workflow, the Model, and the reference documents it needs.
3. In the folder browser, click Use as input file on the source file, and Use as output on the destination folder.
4. Click Trigger run.

Rerun

On a finished run's page, click Rerun to open the same form prefilled from that run. Rerun is handy after you fix a reference document, change the model, or recover a missing folder. It replays the exact file revision from the original run rather than fetching a newer one. If the run's routing rule has since been deleted, create a manual run instead.

Troubleshooting

A file was dropped but no run appeared. Work through the checklist in What triggers a run: file type, filename pattern, folder, enabled workflow, Scribe enabled for the organization, and the input folder still attached with its path intact.

Failed: "output folder path no longer exists". The rule's output folder was deleted or renamed. Recreate it with the same path and rerun.

Failed: "reference file no longer exists". A reference document the rule uses was archived. Upload a replacement and create a new rule.

Failed: "generated output failed malware scan and was not written". The output was blocked before it reached your storage. Rerun; if it repeats, contact support.

Failed with a model message. The model chosen on the rule is no longer available to your organization (a provider was disconnected, or access was revoked). Create a rule with an available model.

A source file is rejected as too large or invalid. Source files and reference documents are limited to 25 MB and checked for malformed archives and malware before processing.

Failed: "an internal error occurred". Something unexpected on PharmFindr's side. Rerun, and contact support with the run's source file name and time if it persists.

Output looks wrong or incomplete. Open Reconciliation results or Prescreening checklist on the run page for per-item notes, check the citations against the source, and leave a comment. Large documents on smaller, cheaper models are the most common cause; try a stronger model on a new rule.