# Review research and export results

Read results from the summary to the underlying evidence. A completed run can contain findings, open questions, failed experiments, missing sources, or no final report.

## Follow the project reading path

| View | Use it to |
| --- | --- |
| **Overview** | See the current assessment, objectives, metrics, and items that need review. |
| **Threads** | Follow research questions and their findings, experiments, and decisions across iterations. Open Methods records when you need the recorded account of how work was performed. |
| **Reports** | Read the final report with **Time Machine** set to **all**, or the selected iteration's summary when an iteration is selected. |
| **Project Explorer** | Inspect the graph, iteration summary, findings, hypotheses, experiments, ruled-out directions, decisions, and VERA reports. |
| **Literature** | Search papers in the project, inspect identifiers and abstracts, and see which findings cite a paper. |
| **Data** | Browse experiment files, research documents, and uploaded datasets. See [Use datasets and project files](/docs/data-and-files). |
| **Activity** | Read live events or earlier run history. |

Some views can be empty for a valid project. **Reports** says when no final report exists. **Literature** is shown for Literature and Full projects. A loading or unavailable message is different from an empty result; retry it or use [Troubleshooting and support](/docs/troubleshooting).

To review notes or check project papers for corrections and retractions, see [Review transferred papers in a project](/docs/atlas#review-transferred-papers-in-a-project).

## Follow a source from a report

In **Reports**, select a finding reference such as `[F050_01]`. Marvin opens the recorded finding beside the report so you can inspect its evidence without losing your place. From the finding, follow linked literature or use **Ask Marvin** to discuss it in chat.

If Marvin shows **No recorded source for this reference**, return to the report and review the surrounding evidence. If Marvin shows **This reference could not be read from the project**, retry, then use [Troubleshooting and support](/docs/troubleshooting) if it still fails.

## Review a finding

Open **Project Explorer > Findings**, then select a finding.

Check:

1. **Claim:** what the finding actually says.
2. **Confidence:** Established, Suggested, Speculative, Inconclusive, Refuted, or Unlabeled.
3. **Quality:** the quality tier shown for the finding, when present.
4. **Full finding:** the detail and qualifications behind the short claim.
5. **Source experiment:** the experiment associated with the finding, when present.
6. **Cited literature:** the sources linked from the finding or its related paper records.

An Unlabeled or missing field is not a positive assessment. Follow the available source links, compare the finding with conflicting results, and check what evidence is absent.

### Follow evidence links

When available, a finding groups its links under **Supports**, **Supported by**, **Extends**, **Extended by**, **Refutes**, **Refuted by**, or **Motivated by**. Select a linked record to read it.

The heading states the recorded relationship. It does not establish that the evidence is sufficient. **Related** lists other connected records without specifying that relationship. Some findings show only **Related**.

## Review hypotheses, experiments, and decisions

Use the other **Project Explorer** tabs to reconstruct what happened:

- **Hypotheses** shows Pending, Confirmed, Refuted, and Superseded states.
- **Experiments** shows plans and experiments with Passed, Failed, Inconclusive, or Pending outcomes. Open an experiment for its brief, design review, code, preflight plan, results, challenge, predictions, plots, and recorded gaps.
- **Ruled Out** keeps dead ends and lessons visible.
- **Decisions** records active project decisions.
- **Graph** shows relationships among project records.

Do not treat a passed experiment as proof of every claim it touches. Read its design, inputs, result, and limitations, then compare those with the finding's wording and confidence.

Use **Preflight** to review the checks planned for the experiment. If the plan or plots fail to load, retry, then use [Troubleshooting and support](/docs/troubleshooting) if the failure continues.

## Read Threads and Methods

**Threads** groups the project's research records by question and iteration. Use it to see which lines of inquiry continued, crossed, or stopped. The selected **Time Machine** iteration brings that iteration into view; it does not delete the other thread history.

Methods records describe work captured for a record or iteration. Read them as project output. They can be absent or marked as needing review. The view does not fill in missing methods.

## Read VERA reports

Open **Project Explorer > VERA** to read available report files. Filter by **Dream**, **Contrarian**, or **Session**, then select a report to open it.

VERA reports are review material. Use them to find challenges, alternate interpretations, or follow-up questions, then check those points against findings and evidence. Do not treat the report type as a confidence rating.

## Compare iterations with Time Machine

Use **Time Machine** in the project header to choose **all** or a numbered iteration. The selection changes the project views that support iteration filtering:

- **Reports** shows the final report for **all** and an iteration summary for a numbered iteration.
- **Project Explorer** and **Overview** show the records available for the selected scope.
- **Threads** keeps the full history visible and brings the selected iteration into view.

Time Machine is a reading filter. It does not restore files, undo later work, or restart an earlier run.

## Review learned skills

Open **Settings > Learned skills**. You can browse skills and expand one to read its description and guidance.

Each skill shows a confidence label such as **Established**, **Suggested**, or **Speculative**. The label describes the recorded basis for the skill. Review the content before reusing it.

To remove a skill from the active list, select **Archive**. Archived skills appear under **Archived** with a **Restore** action. Archiving and restoring require write access and are unavailable while a run is active. These actions do not change saved workflows.

## Download the final report as Word

1. Open **Reports**.
2. Set **Time Machine** to **all**.
3. Select **Download as Word** when it is available.

The `.docx` contains the saved final report and an export note with the download time and report status. Figures referenced by the report are included when they are available project files in PNG, JPEG, or GIF format. Missing, inaccessible, external, or unsupported figures are marked as omitted in the document.

The download does not start research or rewrite the report. Word export applies to the final report; iteration summaries do not offer it. For failed downloads, see [Troubleshooting and support](/docs/troubleshooting#a-source-file-or-export-is-unavailable).

## Export a finding citation

Open a finding and select **Cite**. You can copy its BibTeX entry or select **Download .bib**.

This is a citation for the finding, not an export of the papers that support it. Open the linked literature records to cite those sources separately.

## Download an experiment bundle

Open an experiment and select **Download bundle**. The browser creates `<experiment-id>_bundle.zip` containing:

- readable files fetched from that experiment's folder, including subfolders
- available project settings
- `BUNDLE_README.md` with the experiment ID, iteration, generation time, and a list of included files

Files that cannot be read are left out. The download message reports how many were left out, and `BUNDLE_README.md` lists each one with the reason. Text fetched at the file-content limit is marked as truncated in the file list. Compare the bundle with **View output files** before treating it as complete. Use the **Data** view to download or inspect other project files.