Slice 5: Vertical Text Support

[emcifuentesc]

Parent

• PRD: NihongOCR MVP #1 (PRD: NihongOCR MVP)

What to build

Add support for vertical Japanese text (tategaki). The Reading schema gains an orientation field (horizontal or vertical). The Python sidecar defaults to horizontal but includes orientation in output. The Reading Engine renders vertical text with CSS writing-mode: vertical-rl, furigana to the right of each Kanji, and Romaji horizontally beneath each column. An orientation toggle in the Reading window allows the user to switch layouts without re-running OCR.

Acceptance criteria

• Reading schema includes orientation field
• Python sidecar includes orientation in output (default horizontal)
• Reading Engine renders vertical layout with writing-mode: vertical-rl
• Furigana appears to the right of Kanji in vertical mode
• Romaji renders horizontally beneath each vertical column
• Orientation toggle visible in Reading window
• Toggle re-renders current Reading with flipped layout (no OCR re-run)
• Component tests verify vertical DOM structure with mock Reading data

Blocked by

• Slice 3: Real OCR Pipeline (Horizontal Text) #4 (Slice 3: Real OCR Pipeline (Horizontal Text))

Testing notes

Extend Reading Engine component tests to cover vertical layout DOM structure and furigana placement.

[emcifuentesc]

This was generated by AI during triage.

Agent Brief

Goal: Add graceful error handling across the entire stack. Blank captures show "No text detected" instead of an empty window. Tokens with OCR confidence below 0.7 display a subtle orange underline. English-only text passes through without special-casing. Sidecar crashes trigger automatic restart with the splash screen.

Context:

• Slices 1-5 complete: Tauri scaffold, capture flow, Reading Engine UI, real OCR pipeline, vertical text support
• Domain glossary and ADRs in repo root
• PRD: PRD: NihongOCR MVP #1

Current state:

• CaptureService returns error if no image captured
• Reading Engine already shows orange underline for confidence < 0.7
• SidecarManager spawns sidecar but has no restart logic

What to build:

1. Blank capture handling: show "No text detected" message in Reading window
2. Low-confidence underline: already partially implemented, verify it works end-to-end
3. English-only text: ensure it passes through Sudachi/pykakasi without crashing
4. Sidecar crash recovery: auto-restart with splash screen re-shown
5. Fatal error after 2 consecutive crashes: show error dialog
6. Manual test scenarios for all error states

Key acceptance criteria:

• Blank region capture displays "No text detected"
• English-only text is tokenized and displayed normally
• Tokens with OCR confidence < 0.7 show orange underline
• Sidecar crash triggers automatic restart attempt
• On restart, splash screen reappears and app recovers
• If sidecar fails twice consecutively, show fatal error dialog
• All error states covered by manual tests

Blockers: None — Slice 5 is complete.