Detailed feature reference
Library guide
CADENCE · YOUR MUSIC LIBRARY A HOME FOR YOUR PIECES Open Library in the sidebar, or press Command-2. Existing editable scores and standalone MusicXML studies appear automatically. The collections separate Sheet music, My compositions and the included Examples. Search by title, composer, instrument or tag; filter favorites and sort by title or last opened. PDF imports keep their original source links and page-review controls. New MusicXML imports go into Sheet music, including files opened in Compose. Older editable scores without provenance are treated as compositions unless they exactly match an included example. Use the Collection menu on a piece to correct or change that classification. Moving collection does not move files, change notes or remove an original PDF link. New composition creates a guitar or piano score. Editable examples and imported scores offer Make my own composition copy: a new score with the same music, its own identity and no original-PDF link or inherited AI notebook. The source remains untouched. Native .cadence-score imports retain their source reference when present; library metadata remains local to each library entry. YOUR PIECE NOTEBOOK Each piece has an optional composer/artist credit, tags, a learning goal, notes, a favorite and a stage: To explore, Learning or Comfortable. Stages are your own reflections, not a claim that Cadence assessed your playing. Changes save locally as you make them. The Overview shows the opening notation for editable scores, original-PDF review status and linked practice copies. Open notation returns to Compose; Open study opens a standalone MusicXML score in Studies. Practise creates or reuses a saved practice copy with the selected bars. Copies are listed under their parent piece rather than becoming duplicate library entries. They retain the music and practice progress from when saved. If the parent score is absent, its saved practice copy remains discoverable as its own library entry. Removing a study is still available in Studies. LEARN The Learn tab connects this piece to saved practice pathways and a review schedule. Create a local plan or request AI-tailored passage work. Exercises use the selected notes; passage reports stay separate from checked answers. See Piece pathways guide.txt for the complete workflow and limits. ANALYSIS Choose Analysis, then select a part (for multi-part MusicXML) and up to 16 bars. Bar numbers are positions in the score, which may differ from printed labels. Analyse these bars requests a summary, supported musical observations and small practice steps tailored to this piece's goal and notes. Each observation or step has a validated bar range that opens the passage for editing or practice. Analysis is of the selected passage, not a claim to have analysed the entire piece. Dense selections include only the first 160 note/rest events, clearly identified in the UI and request; choose a smaller passage for complete coverage. Unfinished PDF imports are identified as incomplete and unreviewed drafts. New analysis preserves earlier results; the latest 12 are retained per piece. Each records its scope, date, model and goal at the time. If the score changes, older results stay readable with an earlier-version notice and disabled bar links. Generate a new analysis to get links for the current music. A page's review mark alone does not change the music version. COACH The Coach tab has a saved conversation dedicated to this piece. Ask about technique, harmony, style or your goal. Your selected bars, piece goal and notes provide context. Include my learner profile is optional and initially off. The coach can refer to the latest analysis of the same score and exact passage, while treating it as an unverified interpretation. The last 24 completed question/answer exchanges are retained. Only the four most recent exchanges matching the current score version and exact passage are sent with a follow-up. Older conversations remain readable with their scope and an earlier-version label where appropriate; material from another piece is never silently mixed in. Failed requests leave the question and saved conversation intact. Stopping or navigating away cancels an active request; late results cannot attach themselves to another piece. MODEL, PRIVACY AND SAVING Library analysis and coaching use the model selected in Settings → AI coach and your Keychain-stored OpenRouter key. PDF transcription separately keeps Sol high as its default. Requests occur only when you choose Analyse or Ask coach. They require ZDR and stop after four minutes total, including provider keepalives. There is no automatic paid retry. Already-sent work may still be charged. Requests include symbolic notation, piece metadata/goal/notes, applicable recent conversation and, if enabled, your learner profile. The original PDF and audio are not uploaded by library coaching. It cannot verify transcription accuracy, listen to your performance, or infer an unseen arrangement from a title. Suggestions are interpretations for your review, not assessments of mastery. The notebook is saved in ~/Library/Application Support/Cadence/library.json, alongside existing editor, study and PDF-source files. Back up the entire Cadence data folder; keys remain separately in Keychain. Library notes and AI exchanges are not embedded in MusicXML, MIDI or .cadence-score exports. If a notebook is damaged or from an unsupported future version, Cadence preserves it and disables changes to those notes while your existing scores remain available. A failed local save preserves the previous notebook. If an AI reply arrives but cannot be saved, keep the piece open and use Save received reply again; this retries the local save without making another model call. VERIFICATION The core regression suite covers library grouping, identity-stable music versions, notebook round trips, analysis structure and bar bounds, request context, exclusion of old conversations, ZDR and deadline cancellation. Native checks cover persistence, per-piece response isolation, cancellation, saved practice links, failed-save recovery and preservation of a damaged notebook. New coaching requests are tested with injected responses. Preview analysis and conversation examples are authored local fixtures, not live model evaluations.