Download the complete part-04 practice pack (ZIP). Extract it and preserve its folders. All organizations, records, and results in this teaching pack are fictional.
Use these templates with permitted business records in an approved environment. The supplied Mesa examples are fictional. Blank approval, publication, and observed-result fields are intentional. The Python verification checks fixture arithmetic and consistency; it does not run an AI model, publish content, inspect a live CRM, or validate an experiment's causal conclusions.
1. Customer evidence brief
Decision to inform:
Research question:
Owner and reviewer:
Collection dates and relevant experience dates:
Recruitment method, participant groups, and missing perspectives:
Permitted uses, access, and retention arrangements:
| Finding | Source/participant IDs | Observation | Interpretation | Alternative explanation | Limit | Next investigation |
|---|---|---|---|---|---|---|
Count distinct participants for participant-level claims. Keep generated hypotheses outside the evidence table. Track permission to publish quotations separately from permission to participate in research.
Practice: Open practice/research-excerpts.csv. Compare a model's proposed themes with the original six excerpts, especially the three different fee-clarity problems. A valid grouping must not turn R04's request for a written estimate into an approved Mesa policy. The expected fee-related count is three of six constructed participants; it is not a market estimate.
2. Content assignment and claim ledger
Reader and decision:
Question to answer:
Useful outcome:
Approved source packet and effective dates:
Format, business voice, and next action:
Accuracy reviewer / editor / publication owner:
Owner approval recorded: No / Pending / Yes, with evidence
Publication status: Draft / Approved, not published / Published, with verified URL and date
| Claim | Exact supporting evidence | Source date | Limitation or condition | Decision | Dependent assets |
|---|---|---|---|---|---|
Questions that must remain unanswered until verified:
Review trigger and responsible owner:
Practice: Use practice/Marketing-Sources.md and practice/content-claims.csv. Produce a main explanation and two short adaptations. Check that all three retain $45, approval within thirty days of inspection, and availability checking before confirmation. Reject the archived $35 fee, invented testimonials, and guaranteed appointments.
3. Question-to-page and discovery review
| Reader question | Evidence that this question matters | Primary existing/proposed URL | Distinct page purpose | Useful next link | Owner |
|---|---|---|---|---|---|
One-page review:
- Reader task and whether the opening addresses it:
- Missing answer or unsupported claim, with passage:
- Title and description proposal:
- Current URL and preferred canonical URL:
- Link and navigation checks, with observed evidence:
- Technical access/indexing checks still pending:
- Current source for any search-feature claim:
- Outcome metric, denominator, and observation window:
- Change date and concurrent changes:
Practice: practice/search-review-cases.csv contains five review scenarios. The task is to reject unsupported conclusions while explaining what evidence would resolve them. It does not contain a new crawl or Search Console export. The proposed routes in practice/question-page-map.csv are planning records and are not assertions that pages are live.
4. CRM note and follow-up review
Case/contact ID:
Source interaction and date:
Customer's stated request:
Verified facts and sources:
Unknowns / conflicts / identity issue:
Permitted purpose and channel, with evidence:
Latest opt-out or suppression information:
Proposed next action and owner:
Status justified by the evidence:
Draft response:
Pre-action recheck: Latest reply, identity, permissions, suppression, existing task, current facts
Reviewer decision and evidence:
Action actually completed, by whom, when, and evidence:
Practice: Review practice/crm-cases.csv and practice/crm-301-expected.json. Keep availability unresolved, budget unknown, newsletter permission false, and message-sent false. For an actual connected implementation, test identity, suppression propagation, permitted writes, duplicate prevention, and stale-record rechecks in an authorized environment. A plausible written answer is not proof those controls work.
5. Campaign experiment brief
Decision, hypothesis, and business owner:
Approved offer and proposed variants:
Change being isolated:
Eligible population, assignment unit, and assignment method:
Returning visitors / accounts / cross-device limitations:
Primary metric, numerator, denominator, and observation window:
Qualification rule and reviewer:
Guardrails and operational stop conditions:
Smallest effect worth detecting, sample plan, duration, analysis method, and stopping rule:
Tracking checks and planned exclusions:
Cost definitions, proposed budget, and spending approval evidence:
Downstream metric and maturity window:
Launch approval / actual start date:
Practice: practice/campaign-summary.csv contains constructed totals, and practice/experiment-plan.json records an unapproved design. The data is sufficient for the stated arithmetic only. It does not establish randomized assignment, tracking validity, statistical significance, or a winning campaign.
6. Campaign decision memo
What was planned:
What actually ran, deviations, and missing information:
Counts, rates, absolute difference, and relative difference:
Uncertainty, method, and limits of interpretation:
Guardrail outcomes:
Completed business outcomes and itemized costs:
Decision, rationale, owner, and date:
What to investigate or maintain next:
For the constructed example, the expected rate difference is two percentage points and the relative increase is 40%. Contribution after the specifically listed costs is $460 for A and $340 for B. These are not net profit or a causal result. Do not fill observed-result fields with these expected answers.
Run the practice-file verification
From the extracted part-04/ directory:
python3 practice/verify_examples.py
The script uses Python's standard library. It checks source references, participant counts, fictional policy values, CRM expected states, and campaign calculations. Example-Verification.json records its execution. It leaves observed AI outputs and actual business actions blank. Use practice/observations.csv to record a later authorized evaluation, including the actual system, input version, output, reviewer, and result.