Method, corpusA research data collection with provenance, time window, rights and exclusion rules. and reproducibilityThe ability for another researcher to check a result using the same safe inputs and method version.

Register the method

A method cardA versioned record of a method's purpose, inputs, steps, version, limits and failure mode. records purpose, inputs, steps, limitations, version and an expected failure mode.

• Write the baseline before automation.
• Do not call a draft method reviewed.
• Separate exploratory discovery from a supported conclusion.

Open Science · View benchmarks

Describe the corpus

Each dataset receives a provenance passport, licence, time slice, limitations and deletion rules.

• A synthetic benchmarkA test set with expected answers that measures a reference implementation, not real-world quality. is not a real-world quality metric.
• Share aggregates and reproducibility bundles only after rights review.
• Do not move personal data into QazLake, Compute or an AI provider without a separate basis.

Publish negative results

Non-reproduction, gaps and abstentionA decision not to make a stronger conclusion when evidence is insufficient or conflicting, with the gap stated explicitly. are part of science, not a presentation defect.

• Keep a method changelog.
• Let an independent reader repeat the safe scenario.
• Give every material correction a new version and date.