DATACAREERHUB — PRODUCT ANALYTICS
Eight original lessons, fictional data, Python 3.10+ standard library only.
Extract the entire ZIP. Open a terminal in product-analytics-kit.
Run: python3 analysis.py
Run: python3 -m unittest discover -s tests -v
Windows: use py or your chosen Python executable in place of python3.
No pip, account, network, website DB, real telemetry or payment required.
Open lesson-guide.html for all lessons offline; worked-report.html for the visual reference.

FILES
users.csv:13 signup records including1 internal user;12 eligible fictional users.
events.csv:45 event rows, including1 exact delivery retry,1 internal action and 2
at/after the exclusive2026-01-26T00:00:00Z watermark;41 accepted event rows.
assignments.csv:200 fictional users; reproducible balanced shuffled100/100 assignment.
outcomes.csv:one fully observed seven-day outcome per assignment. Outcomes were
CONSTRUCTED to yield20/28 completions, not observed from a real randomized experiment.
The product and experiment datasets are separate populations; never join their IDs.
data-dictionary.txt, metric-contract-template.txt and metric-contract-worked.txt define the rules.
exercises.txt, worked-solutions.txt, experiment-plan-template.txt, experiment-plan-worked.txt,
decision-memo-template.txt, decision-memo-worked.txt, capstone-rubric.txt support practice.
expected-controls.json contains independent hand-established controls, not generated expectations.
analysis.py and tests/test_analysis.py contain the executable reference.

KEY CONTROLS
Funnel12->8->5; overall5/12=41.67%, conditional8/12=66.67%,5/8=62.50%.
Day 1 cohort returns3/4,2/4,1/4. Day 7:2/4,1/4,N/A; pooled mature3/8=37.50%.
Synthetic experiment20/100 vs28/100; difference+8 percentage points, relative+40%.
Approximate95% unpooled normal interval[-3.7859,+19.7859] points: inconclusive.
Planning example:baseline 20%,absoluteMDE8 points,two-sidedalpha.05,power.80 ->447/arm.
Planning is approximate, not a guarantee. Sample-ratio diagnostic assumes planned50:50.

LIMITS
The watermark and observed outcome flags are explicit fictional source assumptions,
not independent evidence about real event delivery. No vendor UI, live randomization,
identity stitching, timezone/DST policy, cluster inference or production tracker is implemented.
Normal intervals are refused for sparse counts; no exact/score/sequential inference is claimed.
No actual product effect, legal approval, instructor grade or employment guarantee.
Use only invented data; keep your own changes and record unexecuted environments honestly.
Payments/public registration remain disabled. This kit does not alter course access/progress.
