Fix KeyError in cmd_compress: stats dict key mismatch

[helhindi]

Summary

• mempalace compress crashes with KeyError: 'compressed_chars' because cmd_compress in cli.py references stale dictionary keys that no longer match what Dialect.compression_stats() returns
• Updated 6 key references in cli.py and 2 test mock dictionaries in test_cli.py to use the correct keys: summary_chars, original_tokens_est, summary_tokens_est, size_ratio

Key mapping

cli.py used (old)	dialect.py returns (correct)
compressed_chars	summary_chars
original_tokens	original_tokens_est
compressed_tokens	summary_tokens_est
ratio	size_ratio

Repro

$ mempalace compress
KeyError: 'compressed_chars'

Test plan

• All 6 compress-related tests pass (pytest tests/test_cli.py -k compress)
• Verified mempalace compress --dry-run works against a real palace with 21K+ drawers

🤖 Generated with Claude Code

[web3guru888]

Fix is correct. I verified against dialect.py — compression_stats() has returned summary_chars, original_tokens_est, summary_tokens_est, and size_ratio since the 39c14be refactor. The old key names in cli.py were never updated, so this was a guaranteed crash on any mempalace compress invocation.

One non-blocking observation: line 374 in the patch stores to comp_meta["original_tokens"] (the ChromaDB metadata field name) while reading from stats["original_tokens_est"] (the dialect key). The asymmetry is intentional — the stored metadata key predates the _est suffix — but it's a potential source of confusion for future contributors. A brief comment (# stored key name predates _est suffix) would help.

The test mock updates are exactly right. No new tests needed since this is a key-name-only fix with coverage already in place.

LGTM — ship it.

[web3guru888]

Clean fix — the key mismatch is mechanical (cli.py references stale keys that no longer match what compression_stats() returns) and the PR addresses all 4 mismatches consistently across both cli.py and the test mocks.

The key mapping table in the PR description is exactly what this kind of rename fix needs — makes the diff reviewable in 30 seconds.

One note: #562 includes the same compress key fix (along with 14 other items). If #562 lands first, this PR would conflict. Either way one of them fixes it — just flagging for maintainer coordination.

LGTM.

[web3guru888]

Correct fix. The key mismatch is easy to miss because compression_stats() returns a dict that looks structurally similar to what cmd_compress expects — same rough shape, slightly different key names, no type error until the wrong key is accessed.

The test mock updates are exactly right: mock dicts should mirror what the real method returns, not what the caller expects.

One small observation: ratio → size_ratio is the only key rename that changed semantics slightly (ratio implies compression ratio; size_ratio is more precise). The others are just naming convention (_chars, _est suffix). Worth a brief comment in the mock explaining what size_ratio means (chars-before / chars-after) so future test authors don't confuse it with token ratio or compression speed.

Otherwise clean — the fix is minimal and targeted.

[web3guru888]

Clean fix. The key mapping table in the issue description is exactly what a reviewer needs — makes it unambiguous that the rename is one-to-one and the semantics are preserved.

Two things worth noting for context:

1. PR feat: batch writes, concurrent mining, MCP tools, hooks, export, search improvements #562 contains a very similar fix in the same commit (3dd85b66 — "cherry-pick upstream bug fixes — cosine distance, WAL rotation, compress keys"). If feat: batch writes, concurrent mining, MCP tools, hooks, export, search improvements #562 merges first, this PR will conflict on cli.py. Worth coordinating or rebasing against feat: batch writes, concurrent mining, MCP tools, hooks, export, search improvements #562.

2. The test mock updates (changing expected keys in test_cli.py) are the most important part — without those, the bug could regress silently if someone refactors compression_stats() again. The test now anchors the interface contract.

LGTM otherwise. The cmd_compress KeyError is a real crash that makes the compress feature completely unusable in current builds.

---

[MemPalace-AGI integration — 540+ discoveries, 5,251 KG triples]

[jphein]

Already fixed in #562 (commit 3dd85b6) — same key alignment: summary_chars, original_tokens_est, summary_tokens_est, size_ratio. Our fix also updates the test mocks to match.

[bensig]

Closing in favor of #569, which has the identical cli.py fix plus stronger test coverage (adds the "note" key to mocks and a new integration test verifying key parity with Dialect.compression_stats()). Thanks for the contribution!