Pin what is actually being tested.
One public repository, paper artifact, model, or benchmark target and one explicit claim.
Target receipt ✓48-Hour GPU Repro CheckFounding price · $495
Give us one public, non-sensitive AI or ML target. We freeze the environment, attempt one bounded reproduction, and hand back exact evidence—whether it works or fails.
founding pricefixed scope
target turnaroundafter fit and access
public targetone stated claim
GPU ceilingaggregate maximum
Exact scopeEvidence over theater
The deliverable records the source revision, environment, commands, runtime, peak VRAM, observed result versus the stated result, material deviations, and a bounded diagnosis when the run does not reproduce.
One public repository, paper artifact, model, or benchmark target and one explicit claim.
Target receipt ✓Up to eight aggregate GPU-hours on the declared hardware, with no quiet scope expansion.
Runtime evidence ✓Numbers, logs, deviations, and uncertainty stay beside the conclusion.
Claim matrix ✓You receive the environment and command path plus a concise failure diagnosis when applicable.
Reproduction handoff ✓BoundariesPublic and non-sensitive only
This is not certification, peer review, security testing, production deployment, a full paper replication, or a guarantee that the original claim is true or false. We accept only public, non-sensitive targets—never credentials, customer data, malware, attachments, non-public datasets, or unpublished confidential code.
Confirm one safe public target and one checkable claim.
Record source, environment, commands, limit, and abort conditions.
Retain exact outputs, runtime, VRAM, failures, and deviations.
Report observed versus stated without upgrading weak evidence.
Deliver the receipt and a concrete next move.
Start smallNo attachments
Include a public URL and the one result you care about. Do not include secrets, private data, accounts, attachments, or confidential material.
Email hello@cybernative.ai