EDA Report: [Dataset Name]
Analyst: [Name]
Date: [YYYY-MM-DD]
Dataset: [file path or table reference]
Business context: [What does one row represent? What decision does this data inform?]
1. Dataset Overview
| Property | Value |
|---|---|
| Row count | |
| Column count | |
| Memory (MB) | |
| Date range | [earliest] → [latest] |
| Grain | [e.g. one row per user per day] |
| Primary key | [column name] — unique? [Yes / No — N duplicates] |
Schema Summary
| Column | Dtype | Non-null % | Notes |
|---|---|---|---|
2. Null / Completeness Profile
(Copy output from null_profiler.py)
| Column | Null % | Status | Action |
|---|---|---|---|
| FAIL / WARN / OK |
Key findings:
- [Column X is 45% null — expected? Needs business explanation.]
- [Column Y has 0 nulls — pipeline enforcing NOT NULL.]
3. Duplicate Analysis
| Check | Result |
|---|---|
| Full-row duplicates | [N rows, N% of total] |
Key duplicates ([key_col]) |
[N entities appear more than once] |
Notes: [Are duplicates intentional versioning? Data pipeline bug?]
4. Validity Checks
| Column | Check | Result | Notes |
|---|---|---|---|
| [Age ≥ 0] | PASS / FAIL | ||
| [Revenue ≥ 0] | PASS / FAIL | ||
| [Date in range] | PASS / FAIL |
5. Distributions
(Copy descriptive stats from distribution_summary.py)
| Column | Mean | Median | Std | p5 | p95 | Skew | Flag |
|---|---|---|---|---|---|---|---|
Histogram findings:
- [Column X is right-skewed (skew = 2.4) — likely needs log transform before modeling.]
- [Column Y is bimodal — suggests two distinct user populations.]
6. Outliers
(Copy from outlier_detector.py)
| Column | Outlier Count | Outlier % | Classification | Decision |
|---|---|---|---|---|
| Real / Error / Sentinel | Keep / Remove / Investigate |
7. Correlations
(Copy from correlation_explorer.py)
Strong pairs (|r| ≥ 0.8):
| Col A | Col B | r | Explanation |
|---|---|---|---|
8. Business Logic Checks
| Check | Result | Notes |
|---|---|---|
| [Revenue correlates with order count] | PASS / FAIL | |
| [Segment totals reconcile with aggregate] | PASS / FAIL | |
| [Derived columns match source calculation] | PASS / FAIL |
9. EDA Checklist Sign-off
- All 40 checklist items reviewed (
references/eda_checklist.md) - All FAIL items resolved or risk-accepted
- Findings summary written (
assets/findings_summary.md)
EDA status: [Ready to proceed / Needs data fix / Escalated to data engineering]