Code Quality & Output Security

Requirement coverage, code health, and output security.

Retained generated outputs and their historical evaluators, not a matched or currently qualified experiment. Failures, incomplete coverage, replay scoring and different model/backend bindings remain explicit. A live flag does not establish containment or independent replication.

Six tasks and five harnesses, including four provider-auth failures. The equal-coverage earlier run uses different model/backend bindings and remains a separate alternative.

Highlights & plots Dataset · 6 items Experiments · 19 runs Original report ↗All result tables ↗

Models & settings — click to expand

Requirement coverage by harness

Recorded requirement coverage · compare the observed values, not the provider colors

Maintainability debt by harness

Lower is better · debt = vulns + smells + skipped work; value (debt) on bars

Vulnerabilities by CWE

Output-security flaws in generated code (SAST-lite), grouped by CWE

Code-quality judge (architecture / readability / interface)

Per-dimension rubric · higher is better

Code explorer

Click a row to view the generated code, SAST findings, dependency flags, and the requirements flagged as skipped during validation

Task	Harness	Lang	Grade	Coverage	Functional	Vulns	Skipped

Code Quality & Output Security

Requirement coverage by harness

Maintainability debt by harness

Vulnerabilities by CWE

Code-quality judge (architecture / readability / interface)

Code explorer

Inside the dataset

Explore experiments