THE LIGHT COMPANY — PHYSICAL ROBOT EVALUATION PLANNING SHEET
Version: 2026-11-01
Source: https://lightcompany.ai/resources/robot-evaluation
Planning aid. Not a Prism API contract or a claim of sufficient statistical power.

DECISION
What decision should this evaluation support?
What evidence would change that decision?

TASK AND RULES
Task name and version:
Observable success criteria:
Timeout and termination conditions:
Allowed assistance and contact:
Invalid-episode rules:
Review process for uncertain outcomes:

SYSTEM
Robot / controller / gripper:
Policy / checkpoint / configuration:
Camera arrangement and relevant settings:
Calibration reference and date checked:
Guidance visible during policy recording? When does it switch off?

SCENE
Condition identifier and revision:
Object identities and material changes:
Target position and orientation:
Clutter / lighting / surface / robot start state:
What stays fixed? What changes deliberately?
Observed setup and evidence:
Acceptance criteria and unresolved uncertainty:

TRIAL
Episode identifier:
Start / end time:
Policy and scene reference:
Outcome: success / failure / invalid / unresolved
Evidence or recording reference:
Interventions and reasons:
Reviewer and rubric version:
Score changes and reasons:

REPORT
Trials and successes per condition:
Invalid or excluded episodes and reasons:
Uncertainty appropriate to the experiment:
Failure modes worth investigating:
Untested conditions and scope limits:
Next experiment and scenes to revisit: