Instruction-like passages
Supported text patterns that appear to redirect the AI, alter the task or influence the resulting analysis.
Research document security
Scan research documents before they shape AI-assisted analysis.
Check reports, papers and source packs before they become context for research agents and analysis workflows. Retrieve the current Approved copy and check revisions as your source set changes.
- Outside exact versionResearch source
- Organise the recordChoose a Collection for the workflow
- Completed Standard ScanCheck and route
Three destinations
- ApprovedLibrary → current eligible version → AI analysis
- Needs reviewInvestigate, replace or recheck
- QuarantineKeep this exact version out
The Collection keeps all three outcomes organised. Membership does not grant approval or retrieval.
No decision: The scan could not complete, so no security decision was made. Retry is available and no new scan credit is consumed.
The research problem
A source can influence both evidence and instructions.
An AI-assisted research project may treat uploaded documents as evidence while also responding to instruction-like text inside them. That creates a risk when sources come from publishers, downloads, partners or shared folders.
Once a source is retained in project context, it may shape later summaries, comparisons and drafts—not only the first response.
Learn how document instructions can persistResearch intake
Protect the analysis before you expand the source set.
The goal is a consistent decision for every external source, even when the material and publisher change.
Mixed document provenance
Downloads, attachments and shared files need the same intake discipline even when they look authoritative.
Retained project context
A source added once may be consulted across later questions, summaries and research outputs.
Your document workflow
Separate source collection from project admission.
Collect broadly if the work requires it, but make entry into the AI project a deliberate step.
- 1Record the source and why it is relevant.
- 2Explicitly choose the Collection for the research project; .mdSiren does not classify the source automatically.
- 3Check the supported exact version before extraction or analysis.
- 4Follow Approved, Needs review or Quarantine alongside the evidence and coverage.
- 5Keep every source record organised in its Collection while pending, Approved, Needs review or in Quarantine; membership does not grant retrieval.
- 6Retrieve only the current eligible Approved exact version from the Library for later research questions.
When an updated source is accepted as a replacement, it becomes current immediately and needs a new scan; there is no fallback to the earlier version while the replacement is pending or held. Pro API access supports document submission, status checks and retrieval of eligible current Approved exact-version copies under the same rules.
Before analysis
Keep security and research quality as separate decisions.
.mdSiren is designed to support the intake boundary. It does not determine whether a source deserves evidential weight.
- Verify the source, author, version and publication context.
- Choose the Collection for the research project.
- Scan the actual file that will enter the project.
- Review supported findings before extraction or synthesis.
- Check scan coverage instead of relying on status alone.
- Keep a record of which Library version was retrieved for use.
This checkpoint does not determine source accuracy, provenance, permission or evidential quality.
Related resources
Before documents become AI context
Review the source before it becomes part of the analysis.
Check incoming research sources, follow each version’s decision and retrieve current Approved copies for your research workflow.