Source work
Compare answers against a stated rubric
Judge specific requirements against a reference without inventing an overall quality score.
Template, example and checks ↗SMALL TASKS. VISIBLE CHECKS.
12 revised recipes with sample inputs, target results and explicit checks. Start with an example, adapt the template and inspect the answer.
12 recipes shown
Source work
Judge specific requirements against a reference without inventing an overall quality score.
Template, example and checks ↗Source work
Return a small JSON record with evidence for each value and null for information the source does not provide.
Template, example and checks ↗Code work
Generate small counterexamples from a behavioural contract instead of copying the implementation into tests.
Template, example and checks ↗Code work
Use the actual database engine, schema and query plan to propose a testable optimisation without inventing a speed-up.
Template, example and checks ↗Assistant design
Calculate the supplied text-token charges and separate measured usage from unpriced optimisation ideas.
Template, example and checks ↗Code work
Request a small rewrite with explicit behaviour-preservation checks, including awkward inputs already allowed by the code.
Template, example and checks ↗Code work
Ask for a reproducible defect tied to the supplied contract, not a generic checklist or invented repository context.
Template, example and checks ↗Code work
Identify what supplied markup shows and what still needs keyboard, visual or assistive-technology checks.
Template, example and checks ↗Code work
Turn an error report into a small discriminating check without presenting a guess as a confirmed cause.
Template, example and checks ↗Assistant design
Draft role and tool instructions that distinguish the user’s task from instructions found inside retrieved material.
Template, example and checks ↗Assistant design
Compress repeated wording while preserving explicit constraints and showing how to check the revision.
Template, example and checks ↗Source work
Produce a short summary that preserves denominators, uncertainty and the limits of the supplied evidence.
Template, example and checks ↗These are starting points, not guarantees of model behaviour. A worked example is a target to inspect, not a claim that a model produced it. Specific code or arithmetic checks are labelled where they were performed.