Founders
How to build an evaluation dataset for AI coding tools
AI coding tools: Build an evaluation dataset around the decisions the product must support.
Build an evaluation dataset around the decisions the product must support. Include ordinary cases, difficult cases, and known failure modes with documented provenance. Keep development examples separate from the examples used to judge a change.
Start with the decision
Define the unit being evaluated and the expected result before selecting examples. Record who labeled them, what disagreement means, and how the sample relates to actual use. Revisit coverage when the product or customer workflow changes.
The illustrative scenario
This is an educational workbook. The company or fund and all numerical inputs below are fictional. They are not ADIN results, customer claims, market benchmarks, or a forecast.
The fictional tool proposes small code changes in an internal development environment. Engineers review the patch and run the project's checks before merging. The product's user and the person approving a purchase may be different.
A patch that compiles can still violate the intended behavior. Repository permissions, test coverage, and rollback procedures matter alongside developer enthusiasm.
An evaluation sample with a denominator
Imagine a permissioned evaluation set with 25 cases and 17 that meet a rubric defined before the test. The resulting percentage describes only this invented sample. Seventy percent is not a recommended target, a safety standard, or evidence about any real system. What matters is the rubric, sample coverage, and consequences of the remaining cases.
A patch that compiles can still violate the intended behavior. Repository permissions, test coverage, and rollback procedures matter alongside developer enthusiasm. Break the 8 remaining cases into failure categories before interpreting the aggregate. A recoverable wording issue and an unauthorized action may require different product behavior even when both count as a failed case. Keep reviewer disagreement and missing evidence visible.
The assumed time comparison is 35 minutes before and 23 minutes after, including the stated human step. Validate the same start and end points in a real study, then include retries, repair, and handoffs. Compare time to a reviewed, accepted change with the existing workflow. Count investigation and rework rather than treating generated lines as delivered value. A short average time is not enough if the result cannot be used or trusted.
| Item | Illustration |
|---|---|
| Invented evaluation cases | 25 |
| Cases meeting the fictional rubric | 17 |
| Cases requiring investigation | 8 |
| Observed proportion in this exercise | 68.0% |
| Assumed human minutes before / after | 35 / 23 |
Use this worksheet
Answer the questions with a source, date, owner, and next action. Mark an unknown as unknown. Use public or permissioned information and involve a qualified specialist when a question requires expertise beyond the team’s competence.
- What is the evaluation unit?
- Where did the examples come from?
- Who labeled the expected result?
- Which failures are underrepresented?
- What data must remain held out?
Questions specific to this scenario
- Who owns the repository permission?
- Which tests represent the intended behavior?
- How is a bad change rolled back?
Track accepted and rejected patches, tests run, review comments, and downstream fixes. Use representative repositories that the team is authorized to process.
A concrete next test
Select a small accepted patch and a rejected patch from an authorized repository. Ask the maintainer to explain the intended behavior and the checks used. Record whether the tool shortened review or merely moved investigation into a different step.
Interpret the result
A larger dataset is not automatically more representative. An evaluation should reveal consequential behavior in the intended workflow. If repeated tuning teaches the team the answers, add independent examples before claiming a general improvement.
Compare time to a reviewed, accepted change with the existing workflow. Count investigation and rework rather than treating generated lines as delivered value.
Sources and method
These primary resources provide background for the concepts. The fictional scenario, calculations, and worksheet are ADIN educational examples; the sources do not report or validate those invented results. Reviewed October 11, 2026.