# MCP Builder Lab assessment rubric
Score each dimension 0, 1 or 2. A high total does not excuse a missing core behavior.

| Dimension | 0: missing | 1: developing | 2: demonstrated |
|---|---|---|---|
| Contract | Inputs/outcomes unclear | Main path described | Inputs, bounds, empty results and errors specified |
| Discovery | Tool unavailable | Tool listed without inspection | Name, description and schema verified |
| Correctness | Wrong filtering | Normal example works | Normal, empty and boundary cases pass |
| Validation | Invalid input accepted unintentionally | Some failures handled | Blank, missing, wrong-type and out-of-range cases tested |
| Reproducibility | Cannot run | Needs undocumented help | Setup and test work from documented instructions |
| Explanation | Repeats code without meaning | Explains normal flow | Explains design decisions and distinguishes model vs tool errors |

Total: __ / 12
Evidence links or terminal records:
Strongest behavior:
Most important correction:
Retest result:

## Required completion gate
Discovery, correct filtering and invalid-input handling must each be demonstrated before the project is called complete. Score is a coaching aid, not a certification.

## Peer questions
1. Why can an empty array be a successful result?
2. What changes if max_minutes is omitted?
3. Why is "20" different from 20 in this contract?
4. What does a read-only hint describe, and what does the implementation actually do?
5. Which new responsibilities appear if the catalog comes from a real user database?
