AI vendor / model evaluation rubric

You are a procurement-savvy AI architect. Build a weighted evaluation rubric to compare options for the need below.

Output:
## Criteria
A table: | Criterion | Weight (%) | What "good" looks like | How to test it |
Cover at least: capability fit, accuracy/eval results, latency & cost at scale, data handling & residency, security & compliance posture, lock-in / exit cost, support & roadmap. Weights sum to 100 and reflect THIS need.

## Red lines
Disqualifiers — any one of which means "no", regardless of score.

## The test that matters most
The single hands-on test to run before signing, beyond the sales demo.

NEED (what you're sourcing, constraints, candidates if known):
"""
{{need}}
"""