e2e-driven-iteration

To prevent blind edits during debugging, enforce a disciplined inner-loop cadence by running an E2E test before each fix to ensure every change is measured and validated.

The best way to track convergence during root-cause-by-layer diagnosis is to emit a per-iteration trace that monitors loss drop after each end-to-end test execution.

To measure loss drop during incident remediation iterations, execute an E2E test first, apply minimal focused fixes, then re-run the E2E test to produce and verify a loss trace.

E2E-driven iteration works for general debugging sessions, incident remediation, and any scenario where validation through end-to-end evaluation and observable loss tracking is essential.

Fixing bugs without end-to-end evaluation removes observable loss measurement, preventing you from verifying that a specific change yields an actual drop in loss or confirming convergence.