Optical-character recognition of degraded images (rotation, low-contrast)

[kilodesodiq-arch]

Problem Statement. OCR tests run on fixture images. Real field evidence arrives
rotated 90°, low-contrast, watermarked, low-res.

Technical Context.
app/ai-service/services/ocr.py and services/preprocessing.py.

Expected Outcome. A new regression dataset named regression_harness/dataset/degraded/
containing deliberately rotated/blurred samples. Test asserts ≥ 0.6 accuracy per the
preprocessing pipeline.

Acceptance Criteria.

• New workflow ocr-regression.yml runs the degraded dataset.
• 90% pass threshold.

Files or modules likely to be affected.
new app/ai-service/regression_harness/dataset/degraded/,
.github/workflows/ocr-regression.yml.

Difficulty. Medium
Estimated effort. M

---

---

Backlog item #72 from `docs/maintainer-issue-backlog.md.

[Killerjunior]

Hi, I would love to work on this issue. Kindly assign.

[AgilityB]

Hi Lead, kindly oblige me the opportunity to resolve this issue. I've got experience with similar tasks and I'm confident I can investigate, implement a clean solution, and thoroughly test it before submitting a PR.

[AgilityB]

Hi Lead, kindly oblige me the opportunity to resolve this issue. I've got experience with similar tasks and I'm confident I can investigate, implement a clean solution, and thoroughly test it before submitting a PR.

[grantfox-oss]

🦊 GrantFox — @AgilityB has been assigned to this issue as part of the Official Campaign | FWC26 campaign!

Next steps:

1. Open a Pull Request referencing this issue (e.g., Closes #280)
2. Your PR will be reviewed by the ChainForgee maintainers

Good luck! Track your progress on GrantFox.