Start with a question you can test. Each article offers a method, evidence you can inspect, and a clear account of its limits.
20 original articles. Read the method, inspect the evidence, and take a checklist into your own work.
New here? Follow the five-session practice plan →
Experiment · 8 min
We tested two CSV readers on eight original fixtures. Inspect the failures, reproduce the results, and use the import checklist before your next AI analysis.
Read article →Experiment · 8 min
A local test of ten fictional inputs shows why removing emails is only one preparation step. Includes complete results and a reusable pre-prompt checklist.
Read article →Experiment · 8 min
Eight synthetic claims expose the gap between a real quotation and a supported conclusion. Reproduce our checker and reuse the claim-to-source review ledger.
Read article →Experiment · 8 min
Three chunking strategies, four questions, and one missing guest restriction. Inspect the retrieved fragments and learn to locate document-answer failures.
Read article →Workflow · 9 min
Build a task-specific rubric, separate critical failures from style, and compare prompts with held-out cases. Includes a reusable evaluation worksheet.
Read article →Workflow · 9 min
Follow our source-inspection process from a public portfolio to actual code. Reuse the research ledger and distinguish observations, inferences, and missing evidence.
Read article →Case Study · 8 min
Inspect how the operator’s expense assistant separates AI drafts from approved records. Reuse the approval worksheet and understand what source code can prove.
Read article →Case Study · 9 min
A source-level case study of MaybeTomorrow’s queued intent, cancellation, and stale acknowledgements, with a worksheet for designing reliable AI tool actions.
Read article →Case Study · 8 min
We ran the operator’s actual source-grouping function on three original fixtures. Inspect heading paths, page furniture, and the boundaries of document-to-quiz quality.
Read article →Case Study · 8 min
Inspect the operator’s quiz submissions, progress, grading, and chat snapshots. Learn why current state cannot automatically support a story about learning history.
Read article →Experiment · 9 min
Seven records parsed, four matched the shape, and one met the task rules. Reproduce the local test and separate syntax, meaning, evidence, and authorization.
Read article →Workflow · 9 min
Trace the behavior that matters, challenge it with a small counterexample, and distinguish tests found from tests run. Includes a practical code-review worksheet.
Read article →Workflow · 9 min
Map the work around an AI answer: preparation, review, exceptions, and delivery. Includes a reusable workflow canvas grounded in our labs and project inspection.
Read article →Visual Guide · 9 min
Build a context pack that separates evidence, state, defaults, and instructions. Learn from inspected project code and reuse the preparation checklist.
Read article →Workflow · 9 min
Package the question, inputs, failures, results, and decisions behind an experiment. Includes a reusable manifest based on our actual downloadable labs.
Read article →Workflow · 9 min
Decide when adaptive tools help and when a fixed procedure is better. Includes an agent suitability worksheet grounded in this publication’s actual research workflow.
Read article →Workflow · 9 min
Account for retries, review, and accepted results. Work through transparent fictional arithmetic and reuse the task-cost ledger for your own measured workflow.
Read article →Workflow · 9 min
Produce a task contract, context pack, evaluation record, action map, and reusable field note. Follow a practical learning plan using our original labs and worksheets.
Read article →Case Study · 9 min
Explore Review-in-Korea’s locale, route, and article metadata boundaries. Reuse the release matrix for language availability, factual review, navigation, and corrections.
Read article →Visual Guide · 9 min
Turn recorded results and inspected mechanisms into clear charts and diagrams. Includes a visual provenance record, caption guide, and accessible publication checks.
Read article →