Changes in version 0.2.0 (2026-02-25)                  

New Features

  - stamp() replaces commit_keyed() for snapshot creation.
    commit_keyed() is deprecated with a lifecycle warning.

  - diff() method for keyed data frames: cell-level comparison using key
    columns to align rows. Reports added, removed, and modified rows
    with per-column change detail.

  - watch() / unwatch(): Mark keyed data frames as "watched" so dplyr
    verbs auto-stamp before executing. Turns drift detection from a
    manual ceremony into an automatic safety net.

  - check_drift() now returns cell-level reports. When both the snapshot
    and current data are keyed with the same key columns, the drift
    report includes a full keyed_diff with per-column change detail.
    Falls back to structural comparison (row count, columns) when keys
    differ or are lost.

  - stamp() gains a .silent parameter for suppressing cli output during
    auto-stamping.

  - list_snapshots() gains a size_mb column showing memory usage per
    snapshot.

  - compare_structure(), compare_keys(): structural comparison helpers.

Breaking Changes

  - Renamed commit_keyed() to stamp(). The old name is soft-deprecated.

Internal Changes

  - Snapshot cache now stores full data frames (not just hashes),
    enabling cell-level drift comparison without re-reading source data.

  - Cache reduced from 100 to 20 entries and adds a 100 MB soft memory
    cap. Eviction remains LRU-based but now considers both count and
    memory.

  - All dplyr methods (filter, mutate, select, arrange, rename,
    summarise, slice, distinct, group_by, ungroup) now propagate
    snapshot references and watched state through transformations.

                        Changes in version 0.1.1                        

Breaking Changes

  - Operations that break key uniqueness now error instead of warning.
    Previously, keyed would warn and silently remove the key. Now it
    stops with an error, requiring explicit unkey() to proceed. This
    prevents silent data corruption.

Improvements

  - Added Codecov integration for test coverage tracking
  - Improved README with problem-solution framing and Row UUIDs section
  - Added CONTRIBUTING.md with contribution guidelines
  - Expanded test suite to 90%+ coverage
  - Fixed find_duplicates() to work with keyed data that has duplicates

                        Changes in version 0.1.0                        

Initial Release

Key Definition

  - key() / unkey(): Define and remove keys from data frames
  - has_key() / get_key_cols() / key_is_valid(): Query key status
  - Keys survive dplyr transformations (filter, mutate, arrange, etc.)

Assumption Checks

  - lock_unique(): Verify column uniqueness
  - lock_no_na(): Check for missing values
  - lock_complete(): Ensure expected values are present
  - lock_coverage(): Validate reference coverage
  - lock_nrow(): Check row count bounds

Join Diagnostics

  - diagnose_join(): Analyze join cardinality before executing
  - Optional integration with joinspy for enhanced diagnostics

Row Identity

  - add_id() / remove_id(): Add/remove stable UUIDs
  - has_id() / get_id(): Query ID status
  - extend_id(): Fill missing IDs after binding
  - make_id(): Create composite IDs from columns
  - bind_id(): Combine data with ID handling
  - check_id() / check_id_disjoint(): Validate ID integrity
  - compare_ids(): Detect lost/gained rows

Drift Detection

  - commit_keyed(): Commit reference snapshot
  - check_drift(): Detect changes from snapshot
  - list_snapshots() / clear_snapshot() / clear_all_snapshots(): Manage
    snapshots

Diagnostics

  - key_status(): Quick status summary
  - summary.keyed_df(): Detailed summary method
  - compare_structure(): Compare schema between data frames
  - compare_keys(): Compare key values between datasets
  - find_duplicates(): Locate duplicate key values