Research Lab

Research Lab builds feature datasets, runs local or cloud-assisted research jobs, compares results, and can pass selected work into strategy-building and validation workflows.

Preview

Research Lab is a Preview feature. Its dataset formats, Python environment, reports, cloud workflow, and retention behavior can change between releases. Do not use it as the only copy of research data.

Research Lab is not supported on a fresh public installation until the exact clean, signed installer passes validation with its separately packaged runtime prerequisites. Candidate research-module defaults alone do not prove that the Python/runtime payload is present or usable. Use Support And Diagnostics to confirm the installed version. Do not install an untrusted runtime or custom module as a workaround.

Availability

Research Lab opens only when the current license includes the Research Lab or Premium Models entitlement. A visible navigation command does not by itself mean that the license and packaged runtime are available.

Create A Dataset

1. Open Research Lab.
2. Select Create dataset.
3. Choose the instrument, timeframe, historical range, and feature configuration.
4. Review the target-label and lookahead settings.
5. Start the dataset build.
6. Review the dataset summary before starting a research job.

The default research model can include multiple future lookaheads and triple-barrier-style target settings. These are research labels, not trading signals or performance guarantees.

Custom feature providers can add columns to a dataset. A provider that fails while calculating a feature can cause that feature to be omitted, so verify the final feature list before comparing runs.

Run Local Research

Use Run local when the selected workflow supports execution through the managed Python environment on the same computer.

If the runtime is unavailable:

1. Open the Python/runtime actions in Research Lab.
2. Select Repair Python.
3. Wait for verification and extraction to finish.
4. Retry the local job.
5. Export diagnostics if repair continues to fail.

The managed runtime and Research Worker are application components. Do not replace them with downloads from an unofficial source.

Run Cloud ML

The ML scan action is a cloud-assisted workflow.

Data upload

Starting a cloud ML scan uploads the full generated feature dataset as a CSV, not only aggregate metrics. The request also includes research metadata such as instrument/timeframe context, pipeline and label settings, cost-model settings, selected feature names, and filters.

Use cloud ML only when you are authorized to send that market-derived dataset and metadata to the HyperionX cloud service. Remove proprietary or restricted features before starting the scan. Cancelling after upload begins may not recall data already transmitted.

The public documentation does not currently promise a fixed server-side retention period or immediate deletion of uploaded research data. Confirm the current service and privacy terms before uploading sensitive or licensed datasets.

Storage And Encryption Scope

Research Lab stores local data below Documents\HyperionX\ResearchLab.

Data	Current local protection
Saved Research Lab dataset files	Encrypted at rest. The encryption key is protected for the current Windows user.
Protected dataset summary sidecar	Protected with the dataset storage.
CSV and JSON exports	Plaintext.
Worker dataset-build request files containing raw bars	Plaintext JSON.
Job records, generated JSON, Markdown reports, and logs	Do not assume they are encrypted.

Dataset encryption does not encrypt the original provider cache, exports, screenshots, support archives, backups, or files copied out of Research Lab.

Files protected for one Windows user may not be recoverable from another Windows account or after losing that user's Windows key material. Keep an appropriate backup of data you must retain.

Retention And Cleanup

Research Lab does not currently provide a documented automatic retention guarantee for every artifact.

In particular, worker request files can remain in the Research Lab Worker folder after a dataset build. They can contain raw bar data. CSV/JSON exports also remain plaintext until they are removed.

Periodically:

1. Close Research Lab and wait for its worker to stop.
2. Review datasets, exports, reports, logs, and the Worker folder.
3. Back up the items you must keep.
4. Remove only files you recognize and no longer need.
5. Review synced folders, backups, and support archives separately.

Deleting a dataset from Research Lab can also delete matching exports. Confirm that needed exports are backed up before using dataset-delete or clear-all actions.

Review And Promote Research

Research Lab includes actions for prescans, reruns, comparisons, promotion, report review, feature research, feature-filter generation, and Strategy Builder.

Before promoting a result:

• Verify the final feature list and target settings.
• Separate training and later evaluation dates.
• Include realistic costs where the workflow supports them.
• Check for leakage from future data.
• Compare multiple date ranges and instruments.
• Re-run the resulting strategy in Validator.
• Test playback and simulation with an explicitly selected LocalPaper account. Playback does not select it automatically; the current release-candidate build rejects external-account order mutations while Playback is active.

Generated filters or strategies are starting points for review. They are not approved for live trading merely because a research job completed.

Export Safely

Research exports are plaintext and can contain derived market data, labels, feature names, and research output.

Before sharing an export:

• Confirm the underlying data license permits sharing.
• Inspect it for proprietary columns, account information, paths, or notes.
• Use an approved secure-transfer method.
• Delete temporary copies when they are no longer required.

Troubleshooting

When a dataset or job fails:

1. Confirm the entitlement is active.
2. Confirm the historical input has enough rows for every lookahead.
3. Review the final feature list for missing provider outputs.
4. Run Python repair for local-runtime failures.
5. Verify that security software has not quarantined the signed worker/runtime.
6. Export diagnostics and include the job time and action that failed.

See Support And Diagnostics.

Related Pages

• Backtesting
• Validation
• Performance Database
• Install, Update, And Uninstall
• Risk And Security