[docwilde]

lore memory replace without --match answered review_changed and changed nothing. That code means "the reviewed snapshot changed since you looked", so a caller goes looking for a concurrent writer when the real cause is their own match: an empty --match matches every entry, and the substring path refuses anything but exactly one hit.

The direct path now names the match. A missing or empty --match on memory replace|remove|move and filemap replace|remove is invalid_request, as is a positional argument the subcommand does not take, which the parser used to drop silently. A match that hits no entry is the new match_not_found, one that hits several is match_ambiguous, and memory move --to and project move report an unknown project name the same way. review_changed stays wherever it means a stale snapshot or a concurrent change: the exact replay path used by sync and project move, reviewed --expected writes, pending, the gate and the belief stores.

The stray positionals were worse than a misleading code on the old binary: filemap remove src/a.rs deleted the only row, filemap replace src/b.rs second rewrote it, and memory remove --match old stray applied. All of them are refused now.

Text stays variadic for memory add|replace and filemap add|replace, matching the Python CLI's text nargs="+", so replace --match X old new still writes "old new".

Releases 0.62.25: version in the plugin manifests and Cargo files, and a CHANGELOG entry.

Validation: cargo check --locked --workspace, cargo test --locked (all native suites), the native launcher checks, the 41 Python test files and tests/test_native_interop.py pass locally. New unit tests cover a zero and a two-entry match for memory, move and filemap with the file left unchanged, plus an exact-path miss that still reports review_changed. A CLI test drives the binary through eleven refused forms, including the reported one, and through both new codes. With the fix reverted, the new tests fail with review_changed where they expect the new codes. The built binary, run against a scratch LORE_ROOT, returns invalid_request for the reported call, match_not_found and match_ambiguous for the two misses, and applies a correct --match.