Commercial services

Four ways to work with us.

One engine. Four engagements shaped around what you're trying to defend, an FDA submission, a journal revision, an investor memo, or a proprietary cohort. Priced to match scope. Scoped to match evidence requirements.

Data posture. We never share our data and never resell yours. Each engagement validates the client's own data against the published literature, using public reference datasets (MIMIC-IV, eICU-CRD) only as the literature comparator.

Service 01

Discovery Engine License

Engine-only review · same protocol · same standard

A dedicated instance of the RocSite Discovery Engine running against your proprietary patient data. The engine automatically detects contradictions between your outcomes and published literature, and generates a monthly report with evidence hashes and OSF pre-registration for every finding.

Scoped per engagement

Pharma · health systems · clinical AI companies with proprietary data.

What you get

  • Dedicated engine instance, sized for your cohort
  • Your data stays in your environment, or in ours under a DUA
  • Automated contradiction detection vs. 51M indexed documents
  • Monthly discovery reports with novelty scoring
  • Evidence hashes and OSF pre-registration for every finding
  • Publication support, co-authored or ghost, your choice

Four-week onboarding

Week 1

Infrastructure

Data use agreement, environment provisioning, access controls.

Week 2

Cohort baseline

Cohort definition, literature baseline sweep, contradiction thresholds.

Week 3

Discovery run

Engine executes. RocStars consensus validation. Novelty scoring.

Week 4

Findings + OSF

Report delivery, pre-registration, audit trail handoff.

Request a licensing conversation

Adam replies personally. No sales sequence.

Service 02

Adversarial Validation

Engine-only review · same protocol · same standard

Independent falsification testing of your model or claim before FDA submission or hospital procurement. This is the only independent adversarial validation service with publicly documented methodology and a pre-registered protocol. Not proprietary. Not a black box. Citable.

Scoped per engagement

Clinical AI companies · hospital systems · pharma outcomes teams.

Six-layer validation framework

01
Cohort definition audit
02
Feature leakage detection
03
Label stability testing
04
Cross-dataset validation
05
Subgroup performance
06
Temporal stability

What you get

  • Independent falsification report suitable for FDA submission
  • Care-process leakage detection (the one that fails benchmarks)
  • Institution-type bias analysis (eICU-CRD spans ~208 US hospitals; MIMIC-IV is Beth Israel Deaconess)
  • Dataset shift + temporal drift assessment
  • Written validation report with cryptographic evidence chain
  • Ed25519-signed attestation, timestamped on OSF

FDA AI/ML guidance, readiness checklist

  • Good Machine Learning Practice (GMLP) evidence trail
  • Predetermined Change Control Plan (PCCP) compatible
  • Representative data sampling documented
  • Performance monitoring framework included
  • Transparency report generated automatically
  • Bias evaluation across demographic subgroups

Request an adversarial audit

Adam replies personally.

Service 03

Clinical Discovery Partnership

Engine-only review · same protocol · same standard

We run the engine against your proprietary cohort. Joint analysis, joint interpretation, co-authored publication with RocSite methodology credited. All code published openly. OSF pre-registration before any analysis runs.

Scoped per engagement · co-authorship

Academic medical centers · mid-size pharma · contract research organizations.

What's included

  • RocSite engineers run the engine against your cohort
  • OSF pre-registration before analysis, priority claim secured for your team
  • Joint interpretation sessions with Adam and our clinical informatics lead
  • Co-authored manuscript, open methods, open code
  • All evidence hashes delivered with the manuscript
  • Your institution retains data custody throughout

Propose a partnership

Adam replies personally.

Service 04 · New

Evidence Integrity Score

Engine-only review · same protocol · same standard

Run any paper, clinical study, FDA submission, or outcomes claim through the RocSite contradiction engine. Receive a scored report signed with Ed25519 cryptographic attestation, suitable for regulatory submission, investor due diligence, or a journal peer-review response.

Designed to be the first audit a regulator, journal, or investor wants to see.

$2,495 – $24,995 per document

Three tiers · Standard $2,495 · Extended $7,995 · Portfolio $24,995. Volume pricing for portfolios.

The Integrity Score

A 0–100 composite built from five independently-scored axes. Each axis runs against the full engine, not a spot-check.

Example report

Literature consistency
72
Internal consistency
88
Cross-dataset replication
54
Temporal stability
81
Subgroup fairness
76

COMPOSITE INTEGRITY SCORE
74

What the report includes

  • Full breakdown of each axis with contradicting literature citations
  • Specific claims flagged as unsupported or fragile
  • Recommended remediation steps before re-submission
  • Ed25519-signed PDF with cryptographic evidence chain
  • OSF timestamp and Merkle-chain audit trail
  • Suitable for FDA response, journal revision letter, investor due diligence

Run a document

Upload first → instant cost estimate. Nothing charged until you approve.

Pay after you see the quote. Larger uploads routed directly to R2 (up to 2 GB).

Domains

The same framework, in four domains.

The four service tiers above are domain-agnostic. The published track record is in clinical AI; the methodology, pre-registration, cross-dataset replication, adversarial debate, cryptographic attestation, applies anywhere a quantitative claim has high consequences. Engagements outside clinical AI are scoped on request.

Primary credential

Clinical AI

Audits of clinical prediction models, ICU mortality estimates, sepsis benchmarks, institution-type bias, and outcome miscalibration. Three published studies on MIMIC-IV and eICU-CRD; FDA-track work in progress. Cohort definitions and analysis code are peer-reviewable; confirmed findings ship with reproducible bundles.

To engage, pick a tier above based on scope: Engine License for ongoing surveillance, Adversarial Validation for pre-FDA falsification testing, Clinical Discovery Partnership for joint cohort work, or Evidence Integrity Score for a single-paper audit. Contact for scope conversation.

By engagement

Financial Models & Fraud Detection

Audits of backtests, risk models, fraud-detection classifiers, and signal-generation pipelines. The same falsification framework surfaces lookahead bias, label drift, survivorship, trade-execution leakage, and out-of-distribution failure modes that collapse model performance once a strategy is live or once an adversary adapts.

To engage, the work begins with a scope conversation, then a pre-registered protocol (hypothesis, data definition, falsification criteria, planned analyses), an independent audit on held-out or client data, and a citable report with cryptographic attestation. Available by engagement. Contact for scope conversation.

By engagement

Applied Research Validation

Pre-registration, replication, and methodology audit for any quantitative claim where the cost of being wrong is high, published-paper integrity scoring, pre-submission review for journals or institutions, replication of headline findings on independent data, and methodology audits of forthcoming work before public release.

To engage, scope a conversation, then a pre-registered protocol with falsification criteria, an independent run on the relevant data (or on equivalent data when the original is private), and a citable report. Available by engagement. Contact for scope conversation.