skills/obi-image-review/references/review-rubric.md

# OBI image review rubric

## Evidence labels

- `[SCRIPTURE]`: explicitly stated by or clearly contradicts the biblical text
- `[STRONG EVIDENCE]`: well supported by archaeology, historical sources, geography, or museum evidence
- `[PLAUSIBILITY]`: one reconstruction is meaningfully more likely, but uncertain
- `[DEBATED]`: responsible scholars differ
- `[SPECULATIVE]`: evidence is insufficient
- `[ARTISTIC]`: presentation rather than historical accuracy

Every scored deduction must be High or Medium confidence. Low-confidence, debated, speculative, or not reliably assessable observations receive no deduction.

## Image type

- **Story:** faithfulness to the selected biblical moment
- **Depiction:** a responsible possible appearance, not the only reconstruction
- **Art:** may be symbolic or culturally localized; do not require photorealism
- **Map:** geographic and historical accuracy
- **Historic Realism:** carefully researched documentary realism appropriate to the ancient setting

Do not assume all ancient settings were beige, dirty, crude, broken, or poorly crafted. Account for period, region, status, plaster, dyes, wealth, craftsmanship, and ceremonial use.

## Scoring

Begin at 100 and deduct only for visible, supported issues:

- `Critical`: minus 20–40 for a biblical contradiction, wrong event, major anachronism, or fundamentally misleading reconstruction
- `Medium`: minus 8–19 for a significant problem requiring meaningful revision
- `Minor`: minus 2–7 for a limited inaccuracy or localized realism problem
- `Encouragement`: no deduction; a strength to preserve

Calibration: 95–100 exceptional; 90–94 very strong and publication-ready; 80–89 plausible and usable; 70–79 revision required; 60–69 multiple significant problems; 40–59 major reconstruction; 0–39 fundamentally inaccurate or misleading. A score of 80 or higher passes unless a Critical issue remains. Scores above 90 may have no Medium or Critical issues.

## Consultant review

Recommend consultant review when publication depends on disputed interpretation, specialist archaeology, ancient language, complex Temple or Tabernacle reconstruction, or another consequential claim that cannot be verified. State the exact question the consultant should answer.

## Output order

1. `[✅ PASS / ❌ FAIL / ❓ PRELIMINARY] Overall Score = ##/100`
2. Review confidence and consultant recommendation
3. `📖 Context`: passage or subject, moment, date/location, assumptions
4. `📉 Score Reasoning`: start at 100, each deduction, final score
5. `⚠️ Priority Feedback`: severity, issue title, evidence label, confidence, visible problem, why it matters, minimum correction, deduction
6. `🟢 Strengths to Preserve`
7. `🔬 Research Notes`: debated, speculative, low-confidence, or unassessable details with no deductions
8. `🟣 Artistic Feedback`: optional and brief
9. `🔵 Overall Assessment`: 50–100 words and no new issues
10. `🟢🛠️ Localized Edit Prompt`: code block using the supplied image as the exact source, changing only justified issues and preserving aspect ratio, composition, crop, camera angle, poses, expressions, placement, lighting, and grading unless targeted

If no image is supplied, request it. If responsible scoring is impossible, return a Preliminary Review and explain what is missing.