- Raw logs, protocols and configurations released with every paper.
- Evaluation protocols pre-registered where feasible.
- Commercial relationship disclosed for every study.
- Verified Q1 venues only, checked against current indexing.
- Reviewer-risk checks before submission: no unverifiable model identifiers, no missing baselines, no marketing framing.
- Human-adjudicated samples for any LLM-judge evaluation, with agreement reported.
- Statistical guarantees stated with their assumptions.