Changelog¶

All notable changes to TracePipe will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.4.2] - 2026-02-04¶

CheckResult change tracking: Added n_changes and changes_by_op properties in debug mode to track value changes across pipeline steps
TraceResult status fields: Added status, dropped_by, and dropped_at_step properties for clearer dropped row analysis
DiffResult completeness: Added cells_changed, changes_by_column, rows_unchanged, and changed_rows for detailed snapshot comparison
Ghost value API: Implemented dbg.get_ghost_values(row_id) for retrieving last known values of dropped rows
Merge provenance: trace.origin and trace.merge_origin now properly populated for merged rows
Documentation alignment: All documented APIs now match actual implementation with comprehensive test coverage

tp.trace() API enhancement: Added row_id= parameter for explicit internal row ID tracking
row= now strictly refers to DataFrame positional index
row_id= refers to TracePipe's internal row identifier (stable across operations)
Supports tracing dropped rows by ID: tp.trace(df, row_id=42)
tp.why() API enhancement: Added row_id= parameter matching tp.trace() signature

Comprehensive test suite (test_doc_api_alignment.py) with 27 tests validating documented API features
Better error messages for out-of-bounds row access

Fully implemented CheckResult convenience properties (.passed, .retention, .n_dropped, .n_steps, .drops_by_op)
Added comprehensive tests for CheckResult API to ensure properties work correctly
Properties now properly access underlying .facts dictionary for all metrics

Full row provenance for pd.concat(axis=0): Row IDs are now preserved through concatenation
Each result row maintains its original RID from the source DataFrame
ConcatMapping tracks which source DataFrame each row came from
Concat steps are now marked FULL completeness
Duplicate drop provenance in debug mode: drop_duplicates now tracks which row "won"
DuplicateDropMapping maps dropped rows to their kept representative
Supports keep='first', keep='last', and keep=False
Uses hash_pandas_object for fast, NaN-safe key comparison
Clean TraceResult API for provenance:
trace.origin — Unified origin: {"type": "concat", "source_df": 1} or {"type": "merge", ...}
trace.representative — For dedup drops: {"kept_rid": 42, "subset": ["key"], "keep": "first"}
No need to access internal .store methods
Clean CheckResult API:
result.passed — Alias for .ok
result.retention — Row retention rate (0.0-1.0)
result.n_dropped, result.n_steps, result.drops_by_op
All properties discoverable via autocomplete
Comprehensive test suite: 38 new tests covering concat, dedup, and TraceResult API

DataFrame.fillna double-logging: df.fillna({"col": 0}) now logs exactly 1 event
Added wrap_pandas_transform_method with _in_transform_op flag

Event deduplication: Identical events from parallel pipelines are now deduplicated

Double-logging bug: df['col'] = df['col'].fillna() now logs exactly one event
Merge warning scoping: tp.check(df) now only shows warnings for merges in df's lineage

Merge duplicate key warnings now correctly identify which table (left/right) has duplicates

Cell history now correctly chains through merge operations via lineage traversal
tp.why() and tp.trace() show pre-merge changes for post-merge rows
enable() resets accumulated state when called multiple times