| Below is a proposed rubric to evaluate the quality of health-related content on |
| web pages. The goal is to filter for high-quality data suitable for training |
| language models in the health domain. Points are accumulated based on the |
| satisfaction of each criterion: |
|
|
| - Add 1 point if the extract provides basic health information that is clearly |
| focused on a medical or health topic. It moves beyond a passing mention and |
| offers at least minimal informational value, even if mixed with irrelevant or |
| low-quality elements. |
|
|
| - Add another point if the extract offers usable health information while |
| keeping noise low. Promotional or sensational tone does not dominate, and |
| internal consistency is maintained across terms and claims. This tier reflects |
| baseline rigor without assessing external truth. |
|
|
| - Award a third point for specificity, density, and domain precision. The |
| content moves beyond generalities to present concentrated, explicit |
| information with clear conditions or boundaries. Domain-appropriate medical |
| terminology is used accurately and consistently, the signal-to-noise ratio is |
| high, and vague lifestyle phrasing or generic wellness buzzwords are avoided. |
| Organization may still be simple; this tier emphasizes informational richness |
| and precise language. |
|
|
| - Grant a fourth point if the extract is coherent and well-structured. |
| Information is organized with a clear progression suitable for the purpose, |
| integrating details into cohesive explanations. Headings or implicit structure |
| support navigation; lists (if present) are contextualized; paragraphs connect |
| ideas; terminology remains consistent throughout; mechanical or templated |
| artifacts that disrupt readability are absent or minimal. |
|
|
| - Bestow a fifth point for expert synthesis or actionable guidance. The extract |
| organizes knowledge into transferable structures that support reasoning or |
| decisions, or it achieves exceptional depth or breadth with clear boundaries |
| and rationale. Decision logic, criteria, or stepwise guidance are articulated |
| in a way that can be followed. Case narratives that abstract into |
| generalizable patterns also qualify. This tier rewards knowledge organization |
| and procedural clarity rather than external truth claims. |
|
|
| After examining a given extract: |
|
|
| - Output in the following JSON format only (no extra text). |
| - Briefly justify your total score, up to 100 words. |
| - Conclude with an overall quality score as an integer from 0 to 5. |
|
|
| { |
| "reasoning": "<Your justification, under 100 words>", |
| "score": "<0-5 integer>" |
| } |
|
|
