Evidence and trust
Simulation is only as good as its track record
We treat every prediction as a claim to be checked. This page explains how we validate models, how we handle patient data and where in-silico evidence fits in regulatory work.
Principles
What we hold ourselves to
Validation
Back-test before you trust
No model is used for a decision until it has been compared with outcomes it did not see.
Honesty
Say what the model cannot do
Every report lists the questions the evidence does not support.
People
Humans approve decisions
Agents propose. Named reviewers approve anything that leaves the platform.
Method
How a model earns credibility
Context of use
Write down the decision the model will inform and what is at stake if it is wrong.
Verification
Check the code and surrogates against reference mechanistic models.
Validation
Compare predictions with held-out clinical outcomes and report calibration.
Applicability
State the populations and doses where the validation holds.
Standards context
Frameworks our evidence is designed to support
Sanenza supports your evidence work. We are not a certification body and do not hold certifications listed here.
| Framework | What it covers | How Sanenza relates |
|---|---|---|
| ICH M15 (MIDD general principles) | Planning, documentation and assessment of model-informed drug development | Reports are structured around context of use and model credibility |
| FDA MIDD paired meeting programme | Early regulator discussion of MIDD approaches | Evidence packages are designed to support sponsor submissions |
| ASME V&V 40 | Risk-informed credibility of computational models | Our method follows its context-of-use and credibility steps |
| ICH E9 and E9(R1) | Statistical principles and estimands | Endpoints and estimands are defined before simulation |
| 21 CFR Part 11 and EU Annex 11 | Electronic records and signatures | Evidra is designed to support these controls; assessment planned |
| HIPAA and GDPR | Protection of health and personal data | Data processing designed to meet both; DPAs offered for pilots |
Audit trail
Every result can be rebuilt
- Each run stores its cohort version, model versions, seed, inputs and approver.
- Ledger entries are append-only and signed.
- Corrections are new entries that reference the original.
- Evidence packages export with full lineage for inspection.
Security
Protected health data, protected
Compute
Confidential computing
Training and simulation on protected data run on H100 confidential-computing GPUs.
Tenancy
Single-tenant by default
Each sponsor's data, cohorts and models are isolated.
Programme
Audits planned
A SOC 2 Type II audit and an ISO/IEC 27001 certification are planned, not held.
Questions
About evidence and data
Is Sanenza validated for regulatory use?
No product is validated for a regulatory decision in general. Validation is specific to a context of use, and each report states the context it covers.
Do you train shared models on our data?
Not without your written agreement. By default your data only calibrates models in your own tenant.
Can regulators see the evidence?
If you choose. Evidence packages are designed to be shared with regulators as part of your submission.
Review our validation approach with your team
A briefing walks your modellers and statisticians through a back-test on a public example.