Atlas standard
Methodology
A publication standard for AI-assisted medical illustration that keeps anatomy, clinical meaning, provenance, and limitations visible.
- Figures
- 3–5 explanatory plates
- Copy
- 800–1,600 English words
- Evidence
- 5–12 primary references
Scope & boundaries
Illustrate the concept, never fabricate evidence.
Atlas figures are explanatory schematics. They must not imitate a patient scan, claim quantitative accuracy they do not possess, or be used for diagnosis, measurement, navigation, or operative planning.
Label every figure.
Use the standard disclosure beneath each image and record the model, generation date, human edits, and review status in the source log.
Separate schematic from evidence.
An illustration may explain a mechanism. It cannot substitute for clinical imaging, histology, operative findings, or outcome data.
No patient-specific inference.
Do not upload identifiable patient data. Any future clinical image must be rights-cleared, deidentified, and labeled separately from AI-generated artwork.
Editorial limits
A repeatable page, not an endless essay.
Standard limits keep each Atlas entry useful for clinicians, readable for informed patients, and maintainable as evidence changes.
| Element | Standard | Purpose |
|---|---|---|
| Clinical question | One sentence | Defines the exact uncertainty. |
| Page lede | 35–60 English words | Defines the mechanism and the main boundary. |
| Core takeaway | Up to 35 English words | States what the reader should retain. |
| Illustrations | 3–5 figures | Builds one visual chain; Figure 1 may also serve as the social cover. |
| English narrative | 800–1,600 words | Covers mechanism, evidence, relevance, and limits. |
| Evidence summary | 2–4 concise English paragraphs | Synthesizes evidence, uncertainty, and the decision boundary. |
| Figure language | English labels, captions, and alt text | Keeps every visible and machine-readable layer consistent. |
| References | 5–12 sources | Prioritizes guidelines, reviews, and primary evidence. |
| Review record | Version + separate update and review dates | Makes corrections and staleness visible. |
Language & review states
One publication language, one visible review record.
English is the canonical Atlas language. Translation can be produced as a separate derivative later, but a published Atlas page must not mix languages or imply review that has not occurred.
Keep every layer in one language.
Titles, ledes, body copy, image labels, captions, alternative text, metadata, and structured data are English. A translation is a separate page or export, not an inline duplicate title or summary.
Synthesize; do not repeat.
End each entry with two to four concise paragraphs that state what the evidence supports, what remains uncertain, and which clinical inference must be avoided.
Name completed and pending review.
Use Draft — [discipline] review pending while specialist review is incomplete. Use a reviewed label only with an attributable review record or an explicit statement that review is confirmed by the editor but attribution, date, or scope is not published. Confirmation is not independent peer review. Editorial fact-checking, clinical review, and image-anatomy review are separate fields. A disclosure or language update does not reset the scientific-review date.
Image delivery
Keep originals, web files, and versions separate.
Source files remain outside the public deployment. Only versioned, disclosure-labeled web derivatives enter the site, and a published filename is never overwritten.
source-media/atlas/<slug>/
Preserve the highest-resolution original, prompt or generation notes, model and date, editable labels, reference list, and review log here. This directory is never deployed.
assets/atlas/<slug>/
Publish status-labeled WebP display files and PNG only when labels or transparency require it. Use a 1,600-pixel long edge where practical, descriptive alt text, and no embedded patient information.
fig-01-neurovascular-conflict-v1.webp
Lowercase, hyphenated, descriptive, and versioned. Because public
assets are cached for one year, corrections create -v2
rather than replacing -v1.
Publication workflow
Six gates for every release.
A release may carry a clearly disclosed pending-review state, but it cannot omit the anatomy, clinical, source, language, accessibility, or version checks that define its current status.
- 01Draft
Question, outline, sources, and figure plan.
- 02Anatomy status
Check relationships, laterality, scale, and labels; name any pending specialist review.
- 03Clinical status
Check mechanism, relevance, uncertainty, and decision boundaries; never imply undocumented review.
- 04Editorial review
Claims, references, disclosure, language, and accessibility.
- 05Preview
Desktop, mobile, image clarity, links, and metadata.
- 06Publish & recheck
Version the page; review at least annually or when evidence changes.
What a review label establishes
An editorial check addresses claims, sources, language, and disclosure. Clinical review addresses clinical interpretation within an explicitly stated scope. Anatomy review addresses figure relationships, laterality, labels, and procedural state. One does not automatically establish the others.
Reviewer identity, relevant expertise, review date, text/figure scope, version covered, and relevant interests should be recorded before a new review is labeled complete. If information is withheld, the public record states what is not disclosed. No claim of independent peer review is made without evidence of an independent reviewer and the work they reviewed.
Review-pending releases remain visibly labeled at the entry and figure level. An editor’s confirmation date is not the date on which the underlying review occurred. Substantive image edits require a new image version and a fresh assessment of the affected review status.