Limitations

Three things ntropy does not do, each the result of a design choice rather than an oversight. The decision records behind them are linked from each section.

macOS and Linux only

Materialized views are real symlink trees, which is what makes them directories you can cd into and browse with any tool. Windows needs Developer Mode or admin rights to create symlinks, so ntropy targets Unix and leaves Windows out (ADR 0020). There is no Windows build, no symlink fallback, and no Windows testing. Support is deferred, not ruled out. The open questions are the symlink privilege, path and case-folding differences, and the missing SIGPIPE.

Personal scale

Your files are the database. ntropy keeps no index and no daemon. Every list, search, and filter walks all-notes/ and parses frontmatter on demand (ADR 0002). Query cost therefore grows with the vault on every run. That is fine for hundreds to low thousands of notes, which is what ntropy is tuned for, and not for hundred-thousand-note archives. In return everything stays plain, greppable, committable files, and nothing can go stale or corrupt, because there is nothing derived to invalidate. Since the filesystem stays canonical, a cache can be added later without turning it into a database.

Views can drift on out-of-band edits

ntropy refreshes the view trees after its own mutations (create, edit, retitle, delete). Change frontmatter or rename files behind its back, with $EDITOR or a script, and the views will not catch up until the next ntropy reconcile. That one command syncs the view trees, realigns filenames whose slug drifted from the title, and rewrites links that pointed at the old filenames, as described under Commands. The notes themselves are never stale, only the derived views.