Evaluation Planner

Evaluation Planner

Evaluation is where claims become marks. Load ../../references/assessment-rubric.md and ../../references/report-writing.md before planning or rewriting evaluation.

Workflow

1. List the report's contribution claims.
2. For each claim, define the evidence needed to make it defensible.
3. Choose metrics, baselines, datasets, scenarios, users, proofs, or qualitative evidence.
4. Turn raw results into figures/tables with captions that state the finding.
5. Write limitations as controlled scope, not excuses.
6. Tie future work to observed limits, not random feature ideas.

Evaluation Matrix

Use this table:

| Claim | Evidence | Metric/method | Baseline/comparator | Figure/table | Risk if missing |
| --- | --- | --- | --- | --- | --- |

Fast Rules

• Evaluate early if setup, users, hardware, distributed systems, long training, or external access is involved.
• Prefer reproducible scripts for graphs and statistics.
• Compare against a baseline where possible, even a simple one.
• Separate implementation testing from research evaluation.
• Do not claim "better", "faster", "robust", or "novel" without evidence.

Default Output

## Claims To Prove

## Evaluation Matrix

## Figures/Tables To Generate

## Limitations To Admit

## Fastest Next Experiment