PR MemPalace#147 renamed compression_stats fields (ratio -> size_ratio,
compressed_chars -> summary_chars) and switched count_tokens to a
word-based heuristic, but the test_dialect tests from PR MemPalace#131 still
assert the old API and fail on main.

Bring TestCompressionStats.test_stats in line with the current dict
keys (size_ratio, summary_chars, summary_tokens_est) and update
test_count_tokens to match the word-based formula, with extra
coverage for the empty and single-word edge cases around max(1, ...).

This unblocks CI on main, which currently fails on these two tests.

- stats["ratio"] → stats["size_ratio"]
- stats["compressed_chars"] → stats["summary_chars"]
- count_tokens now uses word-based estimation (2 not len//3)
- Remove unused tmp_path_factory param from _isolate_home fixture

…nused fixture param

…nused fixture param

…nused fixture param

…nused fixture param

…nused fixture param

…nused fixture param

…nused fixture param

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

PyPI release cut covering 39 merged PRs since v3.0.0 on 2026-04-06.
Highlights: Claude/Codex plugin packaging (#270), security hardening (#387),
honest AAAK stats + benchmark corrections (#147), Windows compatibility fixes,
Knowledge Graph WAL mode + batching, 10K limit safety caps, and much more.

See GitHub release notes for full changelog.

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

PyPI release cut covering 39 merged PRs since v3.0.0 on 2026-04-06.
Highlights: Claude/Codex plugin packaging (#270), security hardening (#387),
honest AAAK stats + benchmark corrections (#147), Windows compatibility fixes,
Knowledge Graph WAL mode + batching, 10K limit safety caps, and much more.

See GitHub release notes for full changelog.

Co-authored-by: milla-jovovich <[email protected]>

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

The new test_cmd_compress_dry_run and test_cmd_compress_stores_results
tests (added upstream after this branch diverged) mock
compression_stats() with the old key names. Update the mocks to use
the post-MemPalace#147 keys (original_tokens_est, summary_tokens_est,
size_ratio, summary_chars) so they match what the fixed cmd_compress
actually reads.

PyPI release cut covering 39 merged PRs since v3.0.0 on 2026-04-06.
Highlights: Claude/Codex plugin packaging (MemPalace#270), security hardening (MemPalace#387),
honest AAAK stats + benchmark corrections (MemPalace#147), Windows compatibility fixes,
Knowledge Graph WAL mode + batching, 10K limit safety caps, and much more.

See GitHub release notes for full changelog.

Co-authored-by: milla-jovovich <[email protected]>

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

The new test_cmd_compress_dry_run and test_cmd_compress_stores_results
tests (added upstream after this branch diverged) mock
compression_stats() with the old key names. Update the mocks to use
the post-MemPalace#147 keys (original_tokens_est, summary_tokens_est,
size_ratio, summary_chars) so they match what the fixed cmd_compress
actually reads.

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

The new test_cmd_compress_dry_run and test_cmd_compress_stores_results
tests (added upstream after this branch diverged) mock
compression_stats() with the old key names. Update the mocks to use
the post-MemPalace#147 keys (original_tokens_est, summary_tokens_est,
size_ratio, summary_chars) so they match what the fixed cmd_compress
actually reads.

The honest-stats rename in PR MemPalace#147 changed the keys returned by
Dialect.compression_stats() (ratio -> size_ratio, compressed_chars ->
summary_chars, original_tokens / compressed_tokens ->
original_tokens_est / summary_tokens_est). cmd_compress still reads
the old names, so mempalace compress throws KeyError on the first
drawer it touches and the feature is effectively dead.

Also fix the summary line at the bottom of cmd_compress. It called
count_tokens("x" * total_original), but count_tokens is word-based
(max(1, int(len(text.split()) * 1.3))), and a string of repeated
xs is a single "word", so both totals were always 1. Accumulate
the per-drawer estimates during the main loop instead, and use a
token-based ratio so the summary line is self-consistent with the
per-drawer dry-run output.

The storage metadata key names on the compressed collection
(compression_ratio, original_tokens) stay the same for compatibility
with anything already reading them. Only the source of the values
is updated.

Fixes MemPalace#159 (points 1 and 2)

The new test_cmd_compress_dry_run and test_cmd_compress_stores_results
tests (added upstream after this branch diverged) mock
compression_stats() with the old key names. Update the mocks to use
the post-MemPalace#147 keys (original_tokens_est, summary_tokens_est,
size_ratio, summary_chars) so they match what the fixed cmd_compress
actually reads.