Build a 30-Line Provenance Log for AI-Generated Code

Short summary

When code is cheap to regenerate, the durable asset is verified intent. Use an append-only JSONL log to record task, constraints, model, input commit, verification command, and human decision for every AI-generated code change. Commit the log alongside architecture decision records and add a second event after review with the output commit, test result, and accept/reject reason. This creates a searchable chain from intent to evidence that lets you swap models, regenerate implementations, or explain constraints months later.

• •Use append-only JSONL logs to record task, constraints, model, commits, and verification for AI-generated code
• •Commit logs beside architecture decision records; add post-review events with output commit and test results
• •The durable asset is verified intent, not generated lines—enabling model swaps and future explainability

Generated with AI, which can make mistakes.