Spectralgraph
The lab

An independent measurement laboratory, built to be checked.

Spectralgraph LLC is an independent AI-testing laboratory in Enid, Oklahoma. It measures fabrication risk, the confidently wrong answers, in large language models, and reports what the instrument saw with the discipline of nondestructive testing: measure without altering, state the limits, publish the failures. This page is written to be forwarded to your reviewers.

US PATENT PENDING APP. NO. 19/724,790 · METHOD PUBLISHED (DOI 10.5281/ZENODO.19283569) · 18 MODELS · 15 DOMAINS
The instrument

Single-pass measurement of the model's own signals.

The measurement reads a model's token-probability geometry in a single forward pass, on the exact configuration the client deploys, same model, same adapters or fine-tunes, same quantization, supplemented by a documented behavioral protocol scored against ground truth on the client's real question domain. Validated performance on fabricated-entity detection is approximately 0.85 AUC, leave-one-model-out across 18 models; that figure is provisional until the work is published. The founder came to AI measurement from ultrasonic nondestructive testing, and the lab applies the same discipline: test the integrity of a system without opening it.

Evidence discipline

The failures are printed with the findings.

The lab publishes its methodology, its calibration approach, and its negative results. The validation record includes a pre-registered, hash-frozen experiment whose thresholds were not met, reported in full, because a measurement you cannot see fail is not one you can trust to pass. The public 18-model reliability index demonstrates the method; client reports are confidential and never published. Every report states what was measured, how, with what error rates and limitations, and quotes the specific flagged outputs so a reviewer can verify each finding independently.

Independence

Fixed fees. No stake in outcomes. Client-owned reports.

Fees are published, fixed, and never contingent on results. The lab holds no interest in any client's product, sales, coverage, or certification outcome, and declines remediation consulting for systems it measures. Reports belong to the client, are held confidential under written agreement, and are never published or attributed without written permission. For engagements that include reviewer support, technical questions from a client's designated reviewers are answered in writing, on the record, at contact@spectralgraph.ai.

Straight answers

What reviewers ask about the lab

Is Spectralgraph an auditor or certification body?
No, and it will not describe itself as one. Deliverables are screening measurements: quantitative, reproducible, bounded, and stated with their limitations. They identify where a model fabricates for human verification. They are not audits, attestations, certifications, or endorsements.
How is independence maintained?
Fees are fixed and never contingent on the results reported. The lab holds no financial interest in any client's product, sales outcome, or insurance placement, and does not provide remediation consulting to systems it measures. Reports are client-owned; the lab never names a client or publishes an engagement without written permission.
What does the lab need access to?
A client-designated model endpoint and a de-identified question domain. No customer, member, policyholder, or patient data at any point, and nothing installed in anyone's environment. A due-diligence packet with the W-9, no-data-access statement, and security one-pager is available on request.
Why should anyone trust a one-person laboratory?
Judge the work: the method is published with DOIs, the calibration spans 18 models, the validation record includes a pre-registered experiment whose thresholds were not met, reported in full, and every report quotes the exact outputs it flags so reviewers can verify findings independently. Rigor, not headcount, is what a measurement stands on.

Questions from a security review? Send them.

Written technical questions about the method, scope, or a specific engagement are answered on the record, same day where possible.

contact@spectralgraph.ai
SEE ALSO: THE SAMPLE REPORT · THE PUBLIC INDEX