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

# Qualitative document filters

[Canonical page](https://portraitresearch.com/documentation/screening/portrait-textual-reasoning/)

## Applicability

- **Task:** Understand and review company-level textual filters
- **Access:** Portrait account; Normal usage and concurrency limits may apply; source entitlements can apply
- **Audience:** Portrait users discovering companies with explicit criteria
- **Surfaces:** web; ios
- **Availability:** Production
- **Content type:** reference
- **Last reviewed:** 2026-09-22

## What the filter searches

A qualitative filter searches supported company documents, Portrait Primers, and Portrait Case Studies for a condition written in plain language. Available documents vary by company, period, provider, and firm entitlement.

Portrait expands the query into document searches and returns company-level matches with a rationale. Source attribution and passage display depend on the provider's rights; verify a match against the available source before using it.


## Evidence dates and reruns

On the web, choose an **Evidence window** in the filter editor and check the resulting start and end dates. On iOS, describe the period to the [AI Assistant](https://portraitresearch.com/documentation/screening/build-with-ai-assistant/#build-or-edit-a-screen-on-ios), then review the dates in the suggested filter before confirming it. The saved qualitative filter stores those dates; it does not retain a separate fixed-versus-relative preference.

When you run a saved screen again from the web Screening list or result page, or through its automatic schedule, Portrait moves the end of each qualitative window to the run's current date and preserves the number of days between its saved start and end. This also applies to a window originally entered as fixed dates. For example, a 30-day window becomes the 30 days ending on the rerun date.

The new-screen and edit-screen **Run screen** flows on web and iOS use the saved filter dates without advancing them. If you need a historical boundary, review the dates in the builder immediately before running, then verify the new run's filter snapshot. Use the web editor for manual date entry. A web list/result rerun or automatic schedule refreshes the window instead of preserving fixed qualitative dates.

Date scope does not freeze the source corpus or model output. Results can still change with coverage, context, or execution. See [Run and schedule screens](https://portraitresearch.com/documentation/screening/run-and-schedule-screens/) for the distinction between a saved screen and a run.

## Primers and Case Studies

Supported Primers and Case Studies can contribute company context alongside underlying documents. Do not infer that every company has the same coverage or that a generated summary replaces its source material.

## Example uses

- find companies matching a stated investment framework or catalyst;

- find companies exposed to a named development or factor;

- find a defined datapoint or signpost within an ecosystem.

## Query guidance

- A screen supports at most three qualitative filters. If the builder refuses another, remove one before continuing.
- Write a condition that a reviewer could judge from one document.
- Avoid undefined concepts such as *red flags*; name the event and boundary.
- Example: *Find companies that recently replaced their auditor where the new auditor is not KPMG, EY, PwC, or Deloitte.*
- Review false positives and missing evidence before scheduling the screen.