# Missing Value Handler — corpus

Acceptance fixtures for `src/core/missing.py`. Each `.csv` under
`test_data/` is paired with assertions in `tests/test_missing_corpus.py`.
Add a new case by dropping a CSV here and adding a parametrize entry to
the runner.

## Use cases (target client profiles)

| File | Buyer profile | Strategy under test |
|------|---------------|---------------------|
| `uc01_shopify_export.csv`     | SMB / Shopify operator    | `detect-only`       |
| `uc02_marketing_audience.csv` | Marketing / RevOps analyst| `safe-fill`         |
| `uc03_consultant_intake.csv`  | Analyst / consultant      | `drop-incomplete` + threshold |

## Edge cases

| File | What it stresses |
|------|------------------|
| `ec01_all_nan_column.csv`         | column 100 % missing — fill must skip, drop_col must catch at threshold |
| `ec02_no_missing.csv`             | clean file — must be a no-op |
| `ec03_zero_is_not_missing.csv`    | numeric `0`, boolean `false`, `"0"` must NOT be treated as missing |
| `ec04_excel_errors.csv`           | `#N/A`, `#NULL!`, `#VALUE!` Excel error sentinels |
| `ec05_unicode_whitespace.csv`     | NBSP, tab-only, ideographic-space cells treated as whitespace |
| `ec06_mixed_dtypes.csv`           | mixed numeric/string in same column — graceful degrade to mode |
| `ec07_real_data_with_padding.csv` | leading/trailing whitespace around real data must NOT be dropped |
| `ec08_single_row.csv`             | one-row file — every operation must still work |
| `ec09_single_column.csv`          | one-column file with header-only line + sentinels |
| `ec10_all_sentinel_variants.csv`  | every `DEFAULT_SENTINELS` entry exercised in one file |
| `ec11_constant_per_column.csv`    | `column_fill_values` differs per column |
| `ec12_drop_threshold_boundary.csv`| boundary values for `row_drop_threshold` (0.5, 0.99, 1.0) |
| `ec13_ffill_leading_nan.csv`      | leading-NaN run survives ffill (no fabrication) |
| `ec14_interpolate_fallback.csv`   | numeric-only strategy on string column triggers fallback |
| `ec15_headers_only.csv`           | empty body — must not crash |
| `ec16_idempotent_apply.csv`       | running `handle_missing` twice yields the same DataFrame |