18 / 18 · Reason beyond certainty
A conclusion you can stand behind
Reconstruct, test, and qualify a real-looking argument without overstating certainty.
Builds on Count before you trust the percentage
Go to practice ↓A question to keep in mind
Would you adopt a tool based on a promising but incomplete pilot?
Case dossier, fictional teaching data: 20 volunteers tried a writing assistant. Their mean editing time was 12 minutes per task, compared with 20 minutes for a different team. The volunteers had more prior experience. No common accuracy rubric was used. The author claims: 'The assistant caused a 40% improvement and will help every team; roll it out now.'
Separate four layers: observed facts; the inferential bridge; the population the conclusion ranges over; and the decision criteria. The time difference is 40% relative to the comparison mean, but that arithmetic does not identify a causal improvement in quality or productivity.
A useful review preserves the measured signal while naming what is unsupported. State the comparison, list a rival explanation, propose a test, and set a decision boundary. Avoid both promotional certainty and the claim that an imperfect study tells us nothing.
Work through an example
- Supported: the volunteer group had a lower observed mean time than the other team in this pilot. Unsupported: the assistant alone caused the difference, quality stayed equal, and every team will benefit.
- Next test: a randomized or carefully counterbalanced comparison on shared tasks, with time and blinded accuracy scoring, recorded exclusions, and a predeclared stopping rule.
- Decision: run a small follow-up on shared tasks before a broad rollout, with agreed accuracy targets and a fixed time budget. These are explicit priorities, not conclusions forced by the time means.
Your turn
0 / 3Classify the dossier statements, relative to what was provided.
Read solution · does not award completion
- The volunteer mean was 12 minutes. → Reported observation
- The assistant alone caused the difference. → Unsupported inference
- Every team will benefit. → Unsupported inference
- Adopt only if accuracy remains acceptable. → Decision criterion
An observation in a fictional dossier is given for this exercise, not independently verified. Causality and universal scope add claims beyond it. The adoption condition expresses a priority.
Which is the best-supported summary?
Read solution · does not award completion
The pilot showed lower mean time in one group; group differences and unmeasured accuracy limit interpretation.
A precise limited conclusion is more informative than either an overclaim or a blanket dismissal.
Which follow-up most directly improves the evidence for a rollout decision?
Read solution · does not award completion
Compare matched tasks under a prespecified assignment plan, measuring both time and accuracy.
Finish with a testable next step and explicit decision criteria. Do not promise certainty the design cannot supply.
Bring it back to your own work
Write your own 150-word argument audit: claim, premises, hidden assumptions, strongest counterexample or rival explanation, evidence needed, and a qualified conclusion. Then review it against those six headings. This notebook is not automatically graded.
How do I retain independent judgment when using AI here?
Keep your own answer before consulting an AI candidate. Check sources, calculations, inference and conditions; explain adoption or rejection. Then solve a new-context task independently and apply the checks to this lesson.
Practice the verification workflow and keep a recordMy learning status
Mark statuses manually; visits do not establish mastery. Languages share statuses while lessons keep their own answers. This backup contains statuses only, not answers or notes.