Daily routine: prepare experiment → attach output → analyze → record defects → decide → retest. Local browser storage only; no team accounts or cloud sync.
Pass/fail counts are human QA decisions, not successful render counts. AI-assisted review is not connected; selecting the mode only records how a review was performed externally. Every sample keeps its own reason and next action.
Quality gates
Before Pass: attach a generated result, save facial diagnostics and a human comparison/review, and explain the acceptance evidence. Face sharpness alone does not certify quality. Image-only tests need human review even when no comparison reference exists.
Retest after a new render: previous review and QA decision are cleared when media changes. Financial totals are estimates from entered timing/rate, not invoices. No automatic publishing on Pass.