Memory-only de-identification  ·  Clinical & forensic documents
Elider
Process the document.    Protect the person.

The original document never persists. PII stripped in memory, against the protocol you specify, with a full audit trail on every decision.

The layer that makes clinical data usable.

See How It WorksSee the Adversarial Demo →
Scroll

The Constraint

Clinical and forensic data is high-value.
It is also, in most hands, unusable.

You cannot reuse it.

Forensic reports, neuropsychological evaluations, and clinical records contain identifying information that makes them legally untouchable for research, training, or aggregate analysis. The data exists. You cannot use it.

You cannot share it safely.

Attorneys, research teams, and institutional partners need de-identified data. Manual redaction is slow, inconsistent, and indefensible under scrutiny. One missed combination identifier is a liability.

You cannot build from it.

Training datasets, published methodology, quality assurance pipelines — all require de-identified source material. Without a traceable process, none of it survives IRB review or cross-examination.

Before Elider

—Manual redaction — inconsistent, slow, not documented
—No audit trail — indefensible if challenged
—Safe Harbor-only analysis — misses combination and linkage risk
—Data locked — cannot be reused, shared, or published
—Every workflow reinvented per project

After Elider

→Pipeline-driven redaction — consistent, fast, documented
→Full audit trail — every decision traceable, IRB and deposition-ready
→Expert Determination analysis — catches combination, contextual, and linkage risk
→Data reusable — for research, training, QA, and publication
→Protocol library — institutional standards pre-loaded and versioned
Who This Is For

Built for professionals operating under adversarial scrutiny

If your work is reviewed under a microscope, your de-identification process needs to be too.

Forensic Psychologists

Share reports with attorneys, courts, and research teams. Build training datasets for expert witness preparation. Publish methodology with IRB-defensible documentation.

Research Institutions

Convert clinical datasets into publishable material. Satisfy IRB requirements with a documented audit trail. Contribute to institutional protocol library.

Litigation Support

De-identify expert reports for discovery and disclosure. Produce audit logs for opposing counsel review. Process bulk document sets without data retention risk.

Neuropsychologists

Protect evaluee identity in shared test data. Reuse battery results for normative research. Document redaction process for peer review.

Government Agencies

Apply CJIS and agency-specific protocols. Use custom checklists for classified or proprietary requirements. Maintain chain-of-custody documentation.

Training Programs

Create de-identified case libraries for clinical training. Run ForensicPrep and similar simulation platforms on real-but-clean source material.

Protocol Selection

01 / Configure

Select your redaction standard before the document enters the pipeline.

Every institution has its own requirements. Elider meets you there.

Path 01

Institutional Library

Search by institution or jurisdiction. Your IRB protocol is pre-loaded. Harvard, Mayo Clinic, Texas A&M, federal frameworks. Community contributed, editorially reviewed, versioned.

Harvard IRB Minimal Risk Protocol
Texas Department of Criminal Justice
NIH Certificate of Confidentiality
VA Research Program Standards
Select path  →
Path 02

Standard Frameworks

Pre-built, maintained by Elider, always current. HIPAA Safe Harbor. HIPAA Expert Determination. IRB Minimal Risk. CJIS Security Policy. Versioned and fully documented.

HIPAA Safe Harbor (18 identifiers)
HIPAA Expert Determination
IRB Minimal Risk Standard
CJIS Security Policy 5.9
Select path  →
Path 03

Custom Checklist

Define exactly what gets redacted, to what confidence threshold, with what elision flags. Saved to your account. Shareable with your team. Proprietary requirements stay private.

Configure PII categories and depth
Set confidence thresholds per field
Define elision flag behavior
Save and share across your team
Select path  →
Five-Stage Pipeline

02 / Process

The original document never persists. Everything happens in memory.

01
Ingest
Document received in memory.
Never written to disk. Original file not persisted at any point in the pipeline.
02
Parse
Structure extracted.
Text isolated from metadata. Document sections identified. Language and encoding normalized.
03
Extract
Two-pass identification.
Pass 1: pattern recognition flags direct identifiers. Pass 2: contextual NER evaluates combination and linkage risk. Each candidate assigned type and preliminary confidence score.
04
Elide
Decision table applied.
≥85%: auto-elide. 70–84%: human review queue. <70%: retain with note. Every decision logged with rule, confidence, and source.
05
Output
De-identified JSON returned.
Audit trail appended. Protocol version, threshold, and timestamp recorded. Log is exportable for IRB, deposition, or peer review.
Principles

03 / Architecture

Built to answer the black box criticism directly.

Transparency

Clear box, not black box.

Every elision decision is logged with its confidence score and the rule that triggered it. The audit trail is a dataset. Methodology is publishable. IRB-defensible by design.

Privacy

Memory-only. No persistence.

The original document is never written to disk. PII is stripped in memory. No cloud storage of sensitive content. The person in the document is protected at the architectural level.

Standards

Your institution's requirements, not ours.

HIPAA Safe Harbor. Expert Determination. IRB minimal risk. CJIS. Or your own checklist. Elider applies what you specify and documents every decision against the standard you chose.

Research

The pipeline is the methodology.

The audit trail becomes your methodology section. Configurable redaction profiles support expert determination under HIPAA. Designed to produce work that survives IRB review and peer review alike.

Defensibility

04 / Stress Test

Here is where it almost fails. Here is why it doesn't.

A live adversarial demonstration on a complex forensic passage. Contextual identifiers. Relational references. Combination risks. Adjust the threshold. Inspect every rule. Download the audit log.

"her brother John, who works at Dell in Austin"
Combination risk: employer + location = re-identification pathway
Confidence scored. Rule documented. Human review flagged.
Adversarial Methodology
Demonstrating containment of failure modes.
Not a polished demo. A stress test run in public, with a downloadable audit log at the end.
—Live confidence threshold control
—Click-to-inspect every elision rule
—Human review queue for flagged items
—Downloadable JSON audit log
Run the Stress Test →