Verification infrastructure for AI-generated science
Computation is abundant. Evidence is not
Genomarker is the verification layer between scientific computation and scientific claims. It catches methodological failures before they become results, packages every analysis into durable, verifiable evidence — and independently replays the computation so no one has to trust the analyst, the pipeline, or the AI that produced it.
Compute Verify Preserve Replay Prove
Genomarker Assurance StatementGM-2026-08-14-8F29E1Specimen
- Analysis
- Differential expression
- Assurance profile
- GM-BIOMARKER-DE-1.3
- Input integrity
- PASS
- Study design
- PASS
- Statistical methodology
- PASS · DISCLOSED FINDING
- Confounder control & leakage
- PASS
- SAP alignment
- PRE-SPECIFIED
- Execution & provenance
- VERIFIED · COMPLETE
- Independent replay
- REPRODUCED_EQUIVALENT
- Temporal reproducibility
- GRADE A
The problem
The question is no longer “can this analysis be run?”
It is: can this result be trusted, inspected, and independently reproduced after the person, AI agent, software environment, or infrastructure that produced it is gone? Reproducibility fails in specific, recurring ways:
Methods that were never defensible
The wrong statistical test, an unsupported design, a violated assumption — chosen silently and reported confidently.
Confounding and leakage no one caught
Batch tracks treatment; test data leaks into training. The result looks strong precisely because it is wrong.
Analyses that drifted from the plan
What was pre-registered and what was run quietly diverge. Post-hoc becomes indistinguishable from pre-specified.
Provenance that says what, not how
Parameters unrecorded, software versions lost, the exact input data no longer identifiable.
AI decisions nobody can inspect
An agent made an inferential choice mid-analysis. The agent is gone. The justification never existed.
Results that expire with their software
Rerunnable today, unreproducible in five years. The environment decays; the claim stays in the literature.
Scientists, pipelines, and increasingly capable AI agents — producing more analyses, faster and cheaper than ever.
The missing layer: methodology, execution integrity, provenance, preservation, and independent replay.
Publications, biomarker programs, investment decisions, and regulatory filings that depend on computational results being true.
The assurance model
Precise claims, not “certified correct.”
Scientific truth cannot be determined automatically from a workflow. So Genomarker never issues a vague verdict — it evaluates every analysis across independent assurance dimensions, governed by three explicit contracts.
AnalysisSpecWhat exact computation should be performed?
- Content-hashed inputs
- Method, implementation, version
- Parameters, covariates, seeds
- Deterministically hashable
Scientific Analysis PlanWhat did the scientist intend before seeing the result?
- Pre-registration constraints
- Pre-specified vs. post-hoc
- Amendments tracked & signed
- Deviations surfaced, never hidden
AssuranceProfileWhat must be true before a claim can be made?
- Versioned scientific rules
- Evidence & replay requirements
- Permitted exceptions, on record
- e.g. GM-BIOMARKER-DE-1
| Assurance dimension | The question it answers |
|---|---|
| Input integrity | Do we know exactly what data entered the analysis? |
| Study design | Does the experimental design actually support the contrast being tested? |
| Statistical methodology | Were appropriate methodological safeguards applied? |
| Confounder control & leakage | Were detectable confounders and information leakage caught? |
| SAP alignment | Did the analysis follow the pre-registered scientific plan? |
| Execution integrity & provenance | Did the requested computation execute as specified — and can we reconstruct what happened? |
| Independent replay | Can it be re-executed independently of the original execution? |
| Temporal reproducibility | Is enough computational state preserved for long-horizon replay? |
How it works
Assurance across the whole lifecycle.
- Ingest
Content-address every input. Identity before analysis.
- Pre-execution
Versioned assurance rules evaluate the spec. Blocking findings stop bad science before it runs.
- Execution
The requested computation runs exactly as specified, in a recorded environment.
- Evidence
Findings, provenance, environment, and results assemble into a portable, signed package.
- Preservation
Every replay dependency classified and retained. Graded, not assumed.
- Replay
An isolated environment re-executes the analysis and attests to the outcome.
External proof
Illustrative figures — audit in progressWe audit published science we had no role in preparing.
The Public-Corpus Reproducibility Audit re-executes published computational biology analyses through the Genomarker assurance model — an adversarial test of the platform on work it did not produce, publishing as a preprint and an interactive public benchmark.
7
Published analyses independently re-executed
of 9 papers audited · 1 refused with stated reason · 1 unauditable by construction
57
Structured assurance findings emitted
1 hard refuse · 11 block with override · 39 warn · 6 ok
12
Material issues capable of changing interpretation
block-with-override + hard-refuse findings
Illustrative figures — audit in progress71%
Detectable before execution, from the spec alone
Illustrative figures — audit in progress- REPRODUCED_EXACT38%
- REPRODUCED_EQUIVALENT33%
- DIVERGED17%
- NOT_REPLAYABLE12%
The long horizon
Results should not expire when their software does
The north-star demonstration: a computation sealed in 2026 carries enough preserved state that an independent researcher in 2036 — without the original scientist, agent, or application — can verify, reconstruct, and replay it.
2026
Original execution
- AnalysisSpec proposed — by a scientist or an AI agent
- Methodological problem detected; execution blocked
- Machine-readable finding returned; plan corrected
- Analysis executed; signed evidence generated
- Environment and dependencies preserved — Grade A
- Evidence published as GM-2026-08-14-8F29E1
2036
Ten years later — original researcher not required
- Independent researcher retrieves GM-2026-08-14-8F29E1
- Cryptographically verifies the evidence package
- Inspects the original scientific decisions
- Reconstructs the computational environment
- Replays the computation against preserved evidence
- REPRODUCED_EQUIVALENT
One substrate, three surfaces
Wherever computation happens, the evidence is the same.
For scientists & research teams
Genomarker Runtime
Supported computational analyses with methodological assurance built in. Every run produces provenance, a signed evidence package, and a replay path — through the web, API, SDK, or CLI.
- Bioinformaticians
- Translational research groups
- Core facilities
For platforms & AI agents
Genomarker Assurance API
The external system orchestrates; Genomarker constrains, records, and verifies. Agents that cannot be perfect scientists can still operate through a layer that makes their computation inspectable.
- AI scientific platforms
- Pharma internal tools
- Workflow systems
For evaluators of scientific claims
Genomarker Diligence
Service-assisted evidence review: assurance statements, structured findings, replay attestations, and a verdict report for computational claims you did not produce.
- Life-science investors
- BD & pharma diligence
- Translational reviewers
Company
Built by people who have lived the reproducibility problem.
Salar Sayyad
Founder & CEO
Architect of the entire Genomarker platform — the assurance kernel, evidence format, and replay system — built solo, full-time since Feb 2026. 10+ years as an AI engineer, data scientist, and product manager; BSc Computer Science, University of Toronto.
Dr. Saed Sayad
Scientific advisor · AI & bioinformatics
PhD in biochemistry & bioinformatics; adjunct professor at the University of Toronto with 25+ years in machine learning and predictive modeling, focused on biomarker discovery and precision medicine. Inventor of the Real Time Learning Machine and author of the widely used An Introduction to Data Science. Advises on statistical methodology and the assurance rule library.
Dr. Mark Hiatt, MD, MBA, MS
Scientific advisor · clinical & regulatory
Stanford-trained physician executive in precision medicine — former VP of Medical Affairs at Guardant Health, SVP of Market Access & Strategy at BostonGene, and CMO at RadSite; 75+ articles and book chapters, 150+ conference presentations. Advises on clinical, regulatory, and market-access context for evidence standards.
Turn your computation into evidence.
Genomarker is in controlled early access while the public corpus audit completes. Access is reviewed — for research teams, platform partners, and diligence engagements.