Build the back-of-photo OCR review UI #21

New issue

Open

opened 2026-07-28 04:55:06 +00:00 by claude-bot · 0 comments

claude-bot commented

2026-07-28 04:55:06 +00:00

Context

The OCR job handler lands in v0.2.0 and produces evidence, but the review workspace has
no dedicated affordance for working with back-of-photo text. Backs carry the richest
handwritten context in most collections, and OCR on handwriting is unreliable enough that
the reviewer needs the raw text and the image together.

Scope

Reviewer-facing OCR presentation and correction.

Implementation notes

Show recognized text alongside the back scan so a reviewer can check the reading
against the image without switching views.
Where the OCR engine provides bounding boxes, highlight the region a date was read
from on hover. This is what makes a wrong reading obvious at a glance.
Let a reviewer correct the transcription and turn the corrected value into manual
evidence. The original OCR row is superseded, never overwritten â€” the raw machine
reading stays in history.
Handle the common failure honestly: when OCR returns nothing useful, say so and offer
direct manual entry rather than showing an empty panel.
Non-date text (names, places, occasions) is valuable context even when it yields no
date. Keep it visible rather than discarding it.
Re-running OCR appends, consistent with the append-only evidence rule.

Done when

Recognized text renders beside the back scan
Reviewers can correct a reading into manual evidence without losing the original
Empty or failed OCR produces a clear state with a manual path
Non-date text remains visible

References

frontend/src/pages/ReviewWorkspace
docs/circa-spec.md Phase 2 (back-of-photo OCR)

Depends on: #4 (OCR job handler).

## Context The OCR job handler lands in v0.2.0 and produces evidence, but the review workspace has no dedicated affordance for working with back-of-photo text. Backs carry the richest handwritten context in most collections, and OCR on handwriting is unreliable enough that the reviewer needs the raw text and the image together. ## Scope Reviewer-facing OCR presentation and correction. ## Implementation notes - Show recognized text alongside the back scan so a reviewer can check the reading against the image without switching views. - Where the OCR engine provides bounding boxes, highlight the region a date was read from on hover. This is what makes a wrong reading obvious at a glance. - Let a reviewer correct the transcription and turn the corrected value into manual evidence. The original OCR row is superseded, never overwritten â€” the raw machine reading stays in history. - Handle the common failure honestly: when OCR returns nothing useful, say so and offer direct manual entry rather than showing an empty panel. - Non-date text (names, places, occasions) is valuable context even when it yields no date. Keep it visible rather than discarding it. - Re-running OCR appends, consistent with the append-only evidence rule. ## Done when - [ ] Recognized text renders beside the back scan - [ ] Reviewers can correct a reading into manual evidence without losing the original - [ ] Empty or failed OCR produces a clear state with a manual path - [ ] Non-date text remains visible ## References - `frontend/src/pages/ReviewWorkspace` - `docs/circa-spec.md` Phase 2 (back-of-photo OCR) Depends on: #4 (OCR job handler).

claude-bot added this to the v0.4.0 milestone

2026-07-28 04:55:06 +00:00

claude-bot added the

frontend

enhancement

labels

2026-07-28 04:55:06 +00:00

claude-bot referenced this issue

2026-07-28 06:03:07 +00:00

No image zoom or pan, and a failed image blanks the canvas permanently #99