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

# Research companies with Mosaic

[Canonical page](https://portraitresearch.com/documentation/mosaic/research-companies/)

## Applicability

- **Task:** Create and verify company research with Mosaic before expanding it
- **Access:** Mosaic-enabled Portrait account; Easel access; applicable credits and data entitlements
- **Audience:** Analysts with Mosaic access
- **Surfaces:** web
- **Workflow status:** current
- **Availability:** Account-gated
- **Content type:** how-to
- **Last reviewed:** 2026-09-22

Use this workflow to test a research idea on a small company sample, check the evidence and usage, and then investigate the wider universe.

## Before you begin

You need Mosaic and Easel access in the web app, the relevant data entitlements, and sufficient [Mosaic credits](https://portraitresearch.com/documentation/mosaic/usage/). Mosaic is not available on iOS. Prepare the company universe, period, criteria, and the evidence you want to inspect.

<span id="define-the-framework"></span>

## 1. Describe the research

Open **Mosaic** and use its Easel panel to describe the job. For example:

> Create a Mosaic screen for US-listed semiconductor companies. Investigate their exposure to AI data-center spending using recent filings and earnings calls. Show me the proposed quantitative filters, research instructions, and ranking criteria before starting research.

Tell Easel whether you want companies assessed against screening conditions or a **Task** that collects information without ranking. For a Task without quantitative filters, provide an explicit company list; Mosaic does not automatically choose all companies. The saved instructions form a framework, listed in Mosaic as a screen. Open **Mosaic instructions** and check the universe, criteria, sources, and period before asking Easel to proceed.

<span id="run-quant-then-a-trial"></span>

## 2. Test a small sample

1. Ask Easel to run **Quant**. Inspect the company count and filter results. A current Quant execution is required before Research, even when you supply a company list or CSV.
2. Select up to **20 companies** for a trial. Ask Easel to confirm the resolved list before starting. With no explicit selection, Research uses companies that passed Quant; an explicit selection can differ. Companies without suitable documents can be excluded, so check the companies actually investigated.
3. Ask Easel to start **trial Research**. This consumes credits. Follow its status in Mosaic; closing the browser does not cancel it.
4. When it finishes, open several companies, including matches and non-matches. Check the available summaries, research files, and explanations. An execution error or missing evidence is different from a company failing a condition.
5. Review the trial's usage. For screening work, ask Easel to rank the trial if you also want to test the ranking criteria and estimate that stage's cost.

<span id="verify-research-and-ranking"></span>

## 3. Check the results

Open the company's available files, such as `summary.md`, `research.md`, and `fail_reason.md`. Verify material claims against accessible source evidence. Mosaic's research-file view does not provide the citation side panel used by Easel and Deep Research.

The files accumulate within the framework across research executions. An enrichment request can add to earlier work. An old pass's instructions remain recorded, but the company files you inspect can include later work. Downloaded copies are snapshots; editing them does not update Mosaic.

For a ranking, check the company set and rubric as well as the summaries. Elo scores express relative ranking under that rubric; they are not probabilities, valuations, or expected returns. Task work has no ranking.

## 4. Expand the research

Review [Mosaic usage and estimates](https://portraitresearch.com/documentation/mosaic/usage/) before asking Easel to run **full Research**. A current Quant result and a completed trial under the current Research instructions are required. The full-pass estimate can omit ranking if the trial has not been ranked.

Once research has completed, screening work can be ranked. Companies you explicitly select for ranking must already have usable research summaries. Ranking does not itself rerun their research or prove that older evidence satisfies revised instructions.

<span id="what-a-change-requires"></span>

## Change the instructions

Use Easel to revise the saved instructions, then inspect **Mosaic instructions** and its version history.

| What changed | What to run next |
|---|---|
| Quantitative filters | Run Quant again before Research. Changing only Quant does not require another trial. |
| Research instructions | Complete a new trial before the next full Research execution. |
| Ranking criteria | Rank the existing research again if appropriate. Changing only the rubric does not require another research trial. |

Each pass records its instruction versions. A later change does not rewrite those recorded instructions. Ranking can still use older research after a Research revision; check whether you need fresh evidence first.

## Export a ranking

Use **Export to Excel** in the ranking view to download an `.xlsx` file with **Rank**, **Company**, **Ticker**, **ELO**, and **Summary**. It exports the selected ranking's complete result table, not only the companies left visible by a search or filter. The workbook does not include all company research files.

<span id="cancel-reset-and-recover"></span>

## Stop or correct work

| Action | Effect and what to check |
|---|---|
| **Cancel pass** | Requests cancellation. Wait for a final status before retrying. Research already written and usage already incurred can remain. |
| **Pass** on a company that did not pass the screen | Manually changes its screening verdict and retains the previous reasoning in failure history. It does not collect new evidence or repair a failed execution. |
| **Reset research** | Permanently removes the selected companies' generated research and verdicts from this framework's live files, keeping Quant files. It requires a finished trial Research pass and no active pass in the framework. It does not automatically start new research. |
| **Archive screen** | Removes the saved screen from the Mosaic list and makes its detail links unavailable. Results are retained, but there is no restore control in this workflow. Archiving does not cancel a running pass. |

A trial reset acts on the shared live company files, not an isolated trial copy. It can change what you see from other passes. Save any files you need before confirming. A full Research pass cannot itself be reset through this action, and changing instructions does not automatically require a reset.

<span id="read-estimates-and-incomplete-states"></span>

## If research is blocked or incomplete

Read the visible status and check the required Quant result, trial, companies, credits, and any active pass. A queued or running pass is not a completed result. After a failure or cancellation, inspect retained files before requesting another execution.

If you want to stop work before archiving, cancel the active pass and verify that it has stopped first. For an unexplained failure, [contact support](https://portraitresearch.com/documentation/more/contact-us/) with the screen link, selected pass, time, and a redacted error message.

## Historical research

To work at a past date, ask Easel for ready completed month-end snapshots and create a framework with the chosen fixed date. Follow [Run Mosaic at a historical date](https://portraitresearch.com/documentation/mosaic/backtesting/) for snapshot availability, verification, and the current-relationship and authoring-hindsight limitations. A refused date must not silently become a live run.