PoQ
Give quality shape.
See every facet.
Define the facets that matter.
Route uncertainty with agents.
Let expertise examine the detail.
Keep each result open to inspection.
Generation became effortless.
Evaluation stayed complicated.
Judgment needs more than a single score.
Useful evaluation needs a clear structure.
Beyond one answer
Open-ended work needs criteria that explain what quality means.
Beyond one viewpoint
A single perspective leaves too much unseen.
Separate the criteria
Give reviewers distinct dimensions to examine consistently.
Inspect the differences
Agreement is useful. Disagreement tells you where to look.
See quality from more than one side.
Give subjective quality a clear structure.
For outputs, prompts and decisions with more than one right answer.
Step
01
Rubric
Break the standard into the dimensions the work must satisfy.
( A )
Define each facet
Choose the dimensions, weights and thresholds for evaluation.
( B )
Bring the evidence
Collect the outputs and context reviewers need.
( C )
Choose perspectives
Bring qualified experts to review the work.
02
Consensus
Review each dimension before combining the scores.
( D )
Examine independently
Agents surface uncertainty. Experts examine its significance.
03
Record
Trace a conclusion back to the judgments that shaped it.
( E )
Quality Score
See how each dimension contributes to the overall result.
( F )
Consensus Score
Locate the differences that an average could hide.
( G )
Proof Report + PoQ Record
Keep the conclusion and its supporting detail together.
Look at the whole system
Data quality
Measure whether labels and datasets meet the standard you set.
Model outputs
Evaluate answers, recs, and labels.
Agent behavior
Review agent decisions where correctness is a matter of judgment.
Prompt quality
Compare prompts and evaluations that shape production behavior.
Read the detail before relying on the result.
→ What needs fixing
→ Where reviewers differ
→ What supports the result
Find the unclear edges.
Examine the important facets.
Make a better judgment.
Many perspectives. One inspectable record.
One clear structure.
Quality measured by
expert consensus.
Quality measured by
expert consensus.
A useful score comes with an explanation.