Transcript export only includes first line

[hyper-charlie]

Steps to reproduce

1. Record a meeting (36+ minutes)
2. Verify the full transcript appears in the app
3. Export the transcript as PDF or text file
4. Review the exported file

Expected behavior

The exported file should contain the complete transcript matching what appears in the app.

Actual behavior

Only the first line of the transcript is exported. The issue occurs consistently across both PDF and text file exports.

Environment

• Platform: macOS
• OS Version: macOS 15.7.4 Sequoia
• App Version: 1.0.14
• Build: 57054b4

Additional notes

• Full transcript displays correctly in the app
• Meeting length: 36 minutes
• No special formatting or characters in transcript
• Affects both PDF and text file export formats

[devin-ai-integration]

Investigation: Transcript export only includes first line

I explored the codebase around the export functionality. Here's what I found:

Relevant Files

Frontend export UI (v1.0.14 uses ExportModal):

• apps/desktop/src/session/components/outer-header/overflow/export-modal.tsx — The main export dialog with format selection (PDF/TXT/MD/Org) and checkboxes for Summary + Transcript
• apps/desktop/src/session/components/outer-header/overflow/export-pdf.tsx — PDF-only export (older flow, still present)
• apps/desktop/src/session/components/outer-header/overflow/export-transcript.tsx — VTT transcript export

Backend PDF generation:

• crates/export-core/src/typst/content.rs — build_typst_content() builds Typst markup from ExportInput, iterates over all transcript.items in a loop (lines 158-166)
• crates/export-core/src/export.rs — export_pdf() orchestrates Typst compilation
• plugins/export/src/ext.rs — Tauri plugin wrapper

Transcript data retrieval:

• apps/desktop/src/stt/utils.ts — parseTranscriptWords() reads the words JSON cell from TinyBase for each transcript row
• apps/desktop/src/stt/segment/index.ts — buildSegments() pipeline processes words into speaker-labeled segments

How the Export Data Flow Works

1. useSliceRowIds(transcriptBySession, sessionId) fetches all transcript row IDs for the session
2. For each transcript ID, parseTranscriptWords(store, transcriptId) parses the JSON words cell
3. All words are collected, sorted by start_ms, then fed into buildSegments()
4. Segments are mapped to TranscriptItem[] (speaker + concatenated text per segment)
5. For PDF: items are sent to the Rust backend via exportCommands.export(), which builds Typst markup and compiles to PDF
6. For TXT/MD/Org: getTranscriptText() joins items with \n\n and writes via fs2Commands.writeTextFile()

Likely Root Cause

The buildSegments() pipeline groups words into segments by speaker turns. If speaker segmentation produces only a single segment (e.g., all words assigned to the same speaker with no turn boundaries), the entire transcript would appear as one TranscriptItem — but would still contain all text.

However, the issue states "only the first line" is exported. This points to a potential problem in how the segment text is rendered or truncated:

1. Typst compilation issue (PDF): In content.rs:162-165, each TranscriptItem is rendered as a #block[...] in Typst. If the text content contains unescaped Typst special characters, the Typst compiler could fail silently or truncate. The escape_typst_string() function handles common characters but may miss edge cases in long transcripts.

2. Text truncation in TinyBase cell: The words cell stores the entire transcript as a JSON string. For a 36-minute meeting, this could be a very large string. If TinyBase or the persistence layer truncates this value, parseTranscriptWords() would fail to parse the JSON (caught by try/catch, returns []), but the first transcript chunk might still parse correctly — resulting in only partial data.

3. Multiple transcript rows: A 36-minute session likely has multiple transcript rows (one per recording segment or STT chunk). If useSliceRowIds returns IDs but some transcript rows have empty/corrupted words cells, only the first transcript's words would be collected.

4. useMemo reactivity: The transcriptItems memo in export-modal.tsx (line 157) depends on [store, transcriptIds]. If the store hasn't fully loaded all transcript data when the export modal opens, transcriptItems could be computed with incomplete data.

Why It Displays Correctly In-App

The in-app transcript renderer (note-input/transcript/) uses useTranscriptData() which reads each transcript's words cell reactively via main.UI.useCell(). This means each transcript row is read independently and reactively — if data loads incrementally, the UI updates. The export modal computes everything in a single useMemo pass, which may capture an incomplete snapshot.

Suggested Investigation Steps

1. Add logging in export-modal.tsx around transcriptIds and collectedWords to confirm how many transcript rows exist and how many words are collected at export time
2. Check TinyBase persistence — verify that all transcript rows have their words cell fully populated (not truncated or empty)
3. Test with a shorter meeting to see if the issue is length/size-dependent
4. Check if the Typst compiler silently fails on certain transcript content by logging the generated Typst markup before compilation

Recent Changes

• The export UI was recently refactored from separate PDF/Transcript buttons to a unified ExportModal with format selection and content checkboxes (Summary + Transcript)
• plugins/pdf was refactored into crates/export-core + plugins/export
• The TinyBase store was optimized to inline previously nested words (commit b43035b13) — this changed how transcript words are stored and retrieved, which could affect data availability during export