Free Data Analytics Learning Path for Beginners

75%; keep the denominator in your explanation.

Only after checking whether they represent the same event and documenting the rule.

No. Audience, timing and selection can differ.

WIKIFREE / THE EVIDENCE DESK
Original synthetic training case. Version 1.0 — 12 October 2026.
No real customer, company or transaction data. Currency: illustrative USD.

QUESTION
Which channel deserves a small follow-up experiment based on completed revenue
in this sample? You do not have acquisition costs or exposure counts.

RAW CSV — save the following as northstar_orders.csv. Do not edit the raw copy.
order_id,week,channel,quantity,unit_price,status
D001,Week 1,Web,2,20,Completed
D002,Week 1,Email,1,35,Completed
D003,Week 1,Partner,3,20,Completed
D004,Week 1,web ,1,50,Completed
D005,Week 1,Email,2,35,Completed
D006,Week 1,Partner,2,50,Completed
D007,Week 1,Web,1,20,Cancelled
D008,Week 1,,1,50,Completed
D009,Week 1,Partner,4,20,Completed
D010,Week 2,Web,3,20,Completed
D011,Week 2,Email,1,50,Completed
D012,Week 2,Partner,1,35,Cancelled
D013,Week 2,Web,2,35,Completed
D014,Week 2,Email,2,20,Completed
D015,Week 2,Partner,3,35,Completed
D016,Week 2,Web,1,35,Completed
D017,Week 2,Email,3,20,Completed
D018,Week 2,Partner,2,35,Completed
D004,Week 1,web ,1,50,Completed
D019,Week 2,Web,,20,Completed
D020,Week 2,Email,-1,35,Completed

RULEBOOK
One row represents one order. The CSV has six source fields; your clean table
adds a seventh field, revenue. Keep one copy of the exact duplicate D004.
Quarantine D019 (quantity missing) and D020 (quantity negative); do not guess.
Normalize channel labels after trimming spaces and ignoring letter case.
A missing/unrecognized channel becomes Unknown and remains in overall totals.
Completed revenue excludes cancelled orders. Do not treat cancelled orders as
zero-value completed orders. Retain exclusion reasons and reconcile row counts.

DATA DICTIONARY TEMPLATE
Field | Meaning | Type | Allowed values | Missing-value treatment

QUALITY LOG TEMPLATE
Raw line | Order ID | Issue | Action | Reason | Remaining limitation

ANALYSIS TEMPLATE
Decision owner:
Question and sample:
Primary metric and exact denominator:
Cleaning steps:
Channel | Eligible orders | Completed revenue | Average order value
Chart title, units and source note:
Recommendation:
Alternative explanation:
Next test and success measure:

REFERENCE CHECK — open after your attempt
{
  "raw_rows": 21,
  "exact_duplicates": 1,
  "invalid_rows": 2,
  "cancelled_rows": 2,
  "eligible_rows": 16,
  "revenue": 975,
  "channels": {
    "Web": {
      "orders": 5,
      "revenue": 255
    },
    "Email": {
      "orders": 5,
      "revenue": 255
    },
    "Partner": {
      "orders": 5,
      "revenue": 415
    },
    "Unknown": {
      "orders": 1,
      "revenue": 50
    }
  },
  "excluded": [
    [
      8,
      "D007",
      "Cancelled: outside metric"
    ],
    [
      13,
      "D012",
      "Cancelled: outside metric"
    ],
    [
      20,
      "D004",
      "Exact duplicate"
    ],
    [
      21,
      "D019",
      "Invalid or missing quantity"
    ],
    [
      22,
      "D020",
      "Invalid or missing quantity"
    ]
  ]
}
Reconciliation: 21 raw rows = 1 duplicate + 2 invalid + 2 cancelled + 16 eligible.
The grand total includes Unknown. A channel ranking cannot prove causation or
profitability. The dataset does not support a conversion-rate calculation.
Suggested next test: collect comparable channel exposure, cost and completed
orders over a stated period before increasing spending. This is an example of
a bounded recommendation, not the only acceptable answer.

No. This route deliberately selects the spreadsheet and analysis foundations. SQL and R are useful later, but they do not replace the quality checks in this project.

Yes, if you are allowed to use it and can document its meaning and exclusions. Remove personal information before sharing your portfolio.