Phon 4.0.0-beta.13
Pre-release
Pre-release
Changes since Phon 4.0.0-beta.12
Session Editor
- Validators can now enable validation mode and run auto-validation of blind transcriptions from a new transcript toolbar button; the button appears only when editing as the validator on a session that has blind transcribers (#520).
- Removed a non-working validation-mode menu from the transcript gutter's blind-transcription icon; the icon and its hover tooltip are unchanged (#520).
- Fixed tier labels intermittently disappearing in the transcript view, leaving a bare ":" until the next full repaint (most visible on Windows).
- Fixed a blank transcript and cascading dialog-rendering failures that could occur after enabling validation mode or any full transcript rebuild (#517).
- Fixed internal errors and stray text deletion when moving the cursor onto a blind transcriber line with validation mode enabled (#517).
- Fixed a silent failure that left a blank or partial transcript when opening a session in single-record mode; such failures are now recorded in the log (#517).
AI Assistant (Schwa)
- Schwa now maintains a live task checklist for multi-step requests, shown above the message input and kept with the finished reply.
- Schwa asks before acting when a request leaves a meaningful choice open (for example, IPA Target versus IPA Actual), offering clickable options with a recommendation instead of silently guessing (#508).
- Capable models re-check their own edits with read-only tools before giving a final answer, correcting anything that does not match the request.
- Long conversations are automatically condensed as they approach a model's context limit, so answers no longer degrade over time.
- Reasoning effort and context window are now per-model settings, editable in Preferences and from the chat header; reasoning is also requested from Ollama and Groq models.
- Failed tool calls are reported back to the model so it can correct and retry, instead of ending the reply with an error.
- Documentation lookups are faster: the search index stays open between queries, warms in the background, and accepts several queries in one call.
Reports & HTML Viewing
- Analysis reports and other in-app web pages now follow the active light or dark theme instead of always rendering light, and no longer flash white while loading (#503).
- Report toolbar and menu icons are now inline SVG that tint with the theme, replacing the mismatched jQuery UI sprite images and shrinking every generated report file (#503).
- Web view scroll bars and docked report-panel titles now match the dark theme (#503).
Project Window
- Selecting your workspace, home, or a common personal folder (Documents, Desktop, Downloads, and similar) in the open-project dialog is now blocked with an explanation, instead of misreading its subfolders as corpora and writing a stray project file (#515).
- Such folders are also dropped from the recent-projects and workspace histories and skipped when reopening projects at startup, so the block cannot recur (#515).
TextGrid Export
- The Export TextGrid wizard can now export any record tier — including Orthography, IPA, morphology, and user tiers — not just tiers with their own interval timing; intervals are derived from word alignment, or the full record segment when no alignment exists (#513).
- Tier selection moved to its own wizard step, with tiers grouped by timing source and short inline hints matching the import wizard (#513).
CHAT Export
- Intra-word pauses now receive a time interval on the phone intervals tier (%xphoint), so their segment boundaries can be seen and adjusted; phone alignment (%xphoaln) is unchanged.
Performance
- Closed session editor windows are now released from memory; heap usage previously grew with every editor opened and never came back down (#510).