Two-Axis Orthogonal Review (No Cross-Axis Reranking)

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

A requested CSV export is implemented as JSON while following naming and formatting conventions. Review needs separate specification and standards verdicts; clean code cannot compensate for wrong output.

Mechanism

Fix the same target tree, requirements and repository standards. Evaluate each axis with its evidence, severity and unknowns; independent contexts can help when applicable and authorized. Aggregate by axis without cross-axis reranking or an averaged pass. Missing specifications leave that axis unknown; standards come from actual rules.

Bad example

Average everything: wrong format plus clean code yields a high enough overall approval.

Good example

Report separately on the same patch: Spec requests CSV but output is JSON, failing with requirement/artifact evidence; Standards passes only if actual conventions are met, with coverage stated. Do not let standards alter the specification finding’s rank.

Why the change matters

The axes answer what was implemented and how. Separate verdicts keep either failure visible and distinguish requirement misunderstanding from convention problems.

Observable expectation

The teaching report retains Standards pass and Spec fail simultaneously. Changing output to CSV re-evaluates affected items. Unavailable requirements make Spec unknown, not an invented failure. Summaries count findings per axis without compensating totals.

Limits

Standards and requirements may themselves be defective; record conflicts for an owner ruling. Separation does not remove causal relationships, but presentation must preserve verdicts. Frozen instructions explain the mechanism and do not automatically authorize reviewers here.