FINALIST
HUBAI MODEL PILOTTest the same work.
Test the same work.
Choose with proof.
Record what happened—not which brand you prefer. HubAI converts accepted outputs, human correction, speed, cost and failures into a transparent provisional decision.
One task. One acceptance rule.
Device copies stay in this browser. Saving to your account sends these fields to HubAI. Free text is not sent to analytics. Do not paste confidential source material here.
02 · RECORD OBSERVATIONS
Score the result, not the demo.
Use complete-task cost, including retries. A critical failure blocks a recommendation even when the numerical score is high.
FINALIST
Claude
FINALIST
Gemini
03 · PROVISIONAL DECISION
—/100INSUFFICIENT EVIDENCE
Enter valid observations for every finalist. Counts cannot exceed the sample. Use at least five representative cases; eight or more raises confidence.
ChatGPT
0% accepted without rework · 0.0 min average correction · £0.00 total observed cost · Governance evidence not reviewed · No observations recorded
Claude
0% accepted without rework · 0.0 min average correction · £0.00 total observed cost · Governance evidence not reviewed · No observations recorded
Gemini
0% accepted without rework · 0.0 min average correction · £0.00 total observed cost · Governance evidence not reviewed · No observations recorded
VISIBLE MATH · METHOD v1.0
What the score rewards.
Accepted output carries 50 points, correction burden 20, observed total cost 15, speed 5 and reliability 10. Governance is a separate eligibility gate; evidence confidence depends on sample size and review status. A leader also needs accepted outputs, reviewed governance and zero critical failures; empty or invalid observations cannot produce a leader.
- ACCEPTED OUTPUT
- Can a person use it without material correction?
- COMPLETE COST
- Include retries, long outputs and every model call.
- CRITICAL FAILURE
- A material safety, accuracy or policy failure blocks the route.