datatools-dev/docs/REQUIREMENTS.md

# Requirements

Numbered support matrix. Updated with every shipped capability.

## 1. File handling
1.1 Size: ≤ 1.5 GB target (larger works, slower).
1.2 Read: CSV, TSV, XLSX, XLS.
1.3 Write: CSV, TSV.
1.4 Excel: multi-sheet picker.
1.5 Empty file: blocked with `empty_input` error finding.

## 2. Input encodings (auto-detected)
2.1 Unicode: UTF-8, UTF-8-BOM, UTF-16 LE/BE BOM, UTF-16 LE no-BOM.
2.2 Western: cp1252, ISO-8859-1, ISO-8859-15, Mac Roman.
2.3 Eastern European: cp1250, ISO-8859-2.
2.4 Cyrillic: cp1251, KOI8-R.
2.5 CJK: Shift_JIS / cp932, GB18030, Big5, EUC-KR / cp949.
2.6 ASCII → detected as UTF-8.
2.7 User override: any Python codec name.
2.8 BOM: stripped on read, never written.
2.9 Decode failure → `encoding_decode_failed` (error).
2.10 U+FFFD in output → `encoding_uncertain` (error).

## 3. Output encodings
3.1 UTF-8 (default), UTF-8-BOM (Excel-friendly).
3.2 cp1252, ISO-8859-1/15, cp1250, ISO-8859-2, cp1251.
3.3 Shift_JIS, GB18030, Big5, EUC-KR, UTF-16 LE.
3.4 Lossy fallback: `?` + warning when codec can't represent a char.

## 4. Delimiters
4.1 Input auto-detect: `,`, `\t`, `;`, `|`.
4.2 Output: `,` (default), `\t`, `;`, `|`.
4.3 Extension: `.tsv` for tab, `.csv` otherwise.

## 5. Line endings
5.1 Read: LF / CRLF / bare CR — all normalized to LF.
5.2 Embedded in quoted cells: also normalized to LF.
5.3 Write: LF (default), CRLF, CR.
5.4 Mixed → `mixed_line_endings` finding.

## 6. Analyzer detectors

**File-level** (read-time fixes, audit-logged):
- `csv_bom_stripped`, `csv_nul_stripped`, `csv_smart_quotes_folded`, `csv_line_endings_normalized`, `csv_transcoded_to_utf8`, `csv_unquoted_delimiters_repaired`, `csv_unrepairable_rows`.

**Cell-level**:
- `smart_punctuation_in_data`, `nbsp_or_unicode_whitespace`, `zero_width_or_invisible`, `dirty_column_headers`, `whitespace_padding`, `null_like_sentinels`, `suspected_mojibake`, `mixed_case_email_column`, `inconsistent_date_format`, `near_duplicate_rows`, `leading_zero_ids`.

**Encoding integrity**: `encoding_uncertain`, `encoding_decode_failed`, `encoding_lying_bom`, `empty_input`.

Sample size: 1,000 rows (configurable).

## 7. Finding fields
`id`, `severity` (info/warn/error), `confidence` (high/medium/low), `fix_action`, `pre_applied`, `tool`, `count`, `description`, `column`, `samples` (≤5).

## 8. Confidence tiers
- **high** — round-trip safe, one-click auto-fix.
- **medium** — preview before applying.
- **low** — opt-in only, can corrupt if wrong.
- **error** — must resolve or waive before tool pages unlock.

## 9. Decision actions
- `auto` — apply registered fix.
- `skip` — waive (audit-logged).
- `modified` — apply with custom payload.

## 10. Performance (1.5 GB input)
- Initial scan (sample): < 2 s · peak RSS ~110 MB.
- Full-file `repair_bytes`: 30–40 s (UTF-8); non-UTF-8 fold path now
  uses ``str.count`` instead of a Python char-by-char zip walk —
  formerly ~100 s on a 1 GB cp1252 file with smart quotes, now <1 s.
- Full-DataFrame analyze: ~4 min (~25 µs/cell). Near-duplicate detector
  no longer allocates a full-frame copy — peak RSS during the
  near-duplicate pass drops to roughly the size of the string columns
  alone (~50% memory cut on text-heavy 1 GB inputs).
- Full-DataFrame `auto_fix`: ~5 min (~30 µs/cell).
- Output write: ~10 s.
- Recommended RAM: 3–4× input size for the full-Apply path.
- **Format standardizer** (`standardize_dataframe`): ~2.7M rows/sec on
  cache-warm repetition-heavy columns (synthetic 1M-row in-memory
  benchmark, 2 typed columns); the fused single-pass loop replaced a
  3-pass ``.tolist()`` cycle, so per-call overhead is now dominated by
  the underlying parsers (phonenumbers, dateutil) rather than Python
  list materialisation. A 1.5 GB CSV with mixed phone+currency+address
  columns finishes in ~1.5–6 minutes depending on column count.
- **Text cleaner** (`clean_dataframe`): ~1M rows/sec on
  repetition-heavy columns (per-call string cache: the pipeline runs
  once per *unique* cell value, not once per row).
- **Deduplicator**: known O(n²) match step — works to ~50k rows in
  comfortable time. The normalisation pass is now LRU-cached per call
  so repeat values (the common dedup workload) skip re-parsing
  (~2–5× faster on the normalisation step alone). Scale beyond 50k
  needs blocking — flagged in `docs/NEXT-STEPS.md`.

## 11. Tools
1. Deduplicator — Ready
2. Text Cleaner — Ready
3. Format Standardizer — Ready
4. Missing Value Handler — Ready
5. Column Mapper — Ready
6. Outlier Detector — Coming Soon
7. Multi-File Merger — Coming Soon
8. Validator & Reporter — Coming Soon
9. Pipeline Runner — Ready

### 11.a Recommended pipeline order (soft, not enforced)

The Pipeline Runner ships with a `SOFT_DEPENDENCIES` table; the
following ordering is the default and the basis of the warning
surface. Re-ordering is allowed; the runner emits a warning string
and proceeds.

| # | Tool | Why this slot |
|---|------|---------------|
| 1 | column_map (optional, for header alignment) | Multi-vendor unification — rename early so downstream tools see canonical headers |
| 2 | text_clean | NBSP / smart quotes / zero-width pollution silently breaks downstream parsers |
| 3 | format_standardize | Phones / dates / currencies → canonical form before missing detection and dedup |
| 4 | missing | Sentinel detection, imputation, drop strategies — needs canonical types |
| 5 | column_map (optional, for schema enforcement) | Project to target schema, coerce, drop extras AFTER cleaning |
| 6 | dedup | Fuzzy matching is most accurate on canonicalised, sentinel-laundered data |

## 12. Gate (Review & Normalize)
- Gates every tool page.
- Auto-fix button: applies all `confidence=high` findings in one click.
- Per-finding controls: Auto / Skip / Customize.
- Live before/after preview (≤5 sample rows).
- Audit log per fix (id, decision, cells changed).
- Encoding-override picker (16 codepages + custom).
- Advanced output expander: encoding + delimiter + line terminator.
- Result keyed by upload SHA-256; survives reload, invalidated on re-upload.

## 13. Interfaces
- **GUI**: Streamlit, browser-based, local, no internet. Sidebar language picker (English, Español).
- **CLI**: `python -m src.cli` (dedup) · `src.cli_text_clean` · `src.cli_format` · `src.cli_missing` · `src.cli_column_map` · `src.cli_pipeline` · `src.cli_analyze`. (CLI output is English-only.)
- **Python API**: `from src.core import …` (analyze, repair_bytes, clean_dataframe, deduplicate, standardize_dataframe, …).
- **JSON output**: `--json` on `cli_analyze`.
- **Language packs**: `from src.i18n import t, LANGUAGES`. Add `<code>.json` to `src/i18n/packs/` + entry in `LANGUAGES` to add a language.

## 14. Platforms
- Python ≥ 3.10.
- OS: Linux, macOS, Windows.
- Browser: any modern browser.
- Network: not required at runtime.

## 15. Dependencies
- **Core**: pandas, openpyxl, charset-normalizer, typer, loguru.
- **Dedup**: rapidfuzz, phonenumbers.
- **GUI**: streamlit.
- **Optional**: ftfy (mojibake repair).
- **Dev**: pytest, tox.

## 16. Test coverage
- 1,770 tests passing, 0 skipped, 0 xfailed (incl. perf-shape regression tests).
- Fixture corpora: text-cleaner (21), encodings (31), reference UTF-8 (9), format-cleaner (199 buyer cases + 20-row international stress fixture), missing-handler (3 use cases + 16 edge cases), column-mapper (3 use cases + 5 edge cases).
- Run: `python run_tests.py [--tool …] [--fixtures] [--coverage]`.

## 17. Privacy / data handling
- All processing local; no network calls in the data path.
- No telemetry.
- Original input never modified.
- Audit logs: `logs/` next to each run (timestamped).

## 18. Error handling
- Structured hierarchy: `DataToolsError` → `InputValidationError`, `ConfigError`, `FileFormatError`, `FileAccessError`.
- Subclasses extend stdlib `ValueError` / `OSError` so existing handlers still catch them.
- Every error carries: message, file path, column, operation, suggestion, underlying cause.