Skip to content
MINATON

Some tools opened from English are still Japanese. English article bodies ship after physician review.

About this site

Article quality

A sentence you read while caring for your child can shape a decision at home. We want to build this site with that responsibility in mind.

Where did the information come from? When did a physician check it? How are changes recorded? We explain the safeguards in place and the work still to do, so you can see what has been checked.

This page does not guarantee that every article is accurate or up to date. The published archive does not yet have physician-review records for every article.

What you can check when reading.

References show where information came from
We present the references and links recorded in an article. A bibliography alone cannot guarantee that the content is correct or that a physician has checked it.
Physician review requires a named reviewer and date
We show physician-review information only where an explicit record exists for that article. The operator being a physician, or an article having a publication date, does not establish that an individual article was reviewed.
Publication and review dates mean different things
The publication date records when an article was published. Revisions and physician reviews are shown where their records exist. A recent date alone does not mean that every part of an article has been checked.

Who decides

AI narrows things down. A physician decides.

This is the target workflow for every future revision. The full current archive does not yet have records proving that each article passed every step.

  1. 01Watch for changeAutomated
  2. 02Find the affected articlesAI
  3. 03Draft the revision and its reasoningAI
  4. Held by a person
    04Decide whether it may be publishedPediatrician
  5. 05PublishAutomated
This is the target revision loop. Monitoring, matching and drafting may be automated; a pediatrician makes the publication decision. Per-article records for this gate are still being implemented.
  1. 01Watch for changeAutomatedSociety guidelines, government notices and drug package inserts are checked on a schedule
  2. 02Find the affected articlesAIThe change is matched against the existing body of articles
  3. 03Draft the revision and its reasoningAIWhat would change, and why, presented side by side
  4. 04Decide whether it may be publishedPediatricianApprove, reject or amend. This step stays with a person
  5. 05PublishAutomatedOnly approved revisions go live, together with the revision record

That line does not move. AI is used to narrow down what the physician needs to look at, so the judgement happens sooner. It is not used to make the judgement.

Full automation is not the goal. Publishing medical guidance cannot be undone, because a parent reads it and acts on it. Approval stays with a person.

Checks before publishing

Preventing known problems from returning.

These are examples of checks implemented today. In the formal release process, a failure stops publication.

  • Previously identified contradictions have not returned. Tests check for known problems in areas such as medication and first aid; they cannot identify every medical error
  • Articles registered as awaiting checks, and articles marked as drafts, stay out of publication. An article placed on hold remains excluded until the required checks are complete
  • Internal links point to pages that exist
  • The disclaimer and the emergency numbers (#8000 and 119) are always present
  • Only references already written into an article are structured. We do not let AI invent citations, and we do not treat a bibliography as proof of physician review

We also have tools for the following checks of external sources. They do not run to completion automatically on every publication, and they do not guarantee that every source is up to date.

  • Whether source URLs still resolve
  • Whether a domain has gone stale because an institution was reorganised
  • Whether DOI and PMID values are well formed

Not letting AI invent citations is the line we hold most firmly. Sources are structured only from references already written in the article. Plausible-looking citations are never added afterwards. A bibliography alone is never treated as proof of physician review.

What we claim

We show what the records establish.

Published

  • The name of the site operator
  • The references an article actually cites
  • The date the article was published
  • Any revision records that exist
  • A reviewer and review date only where an explicit record exists

Deliberately withheld

  • A last-reviewed date we have not verified. The date appears only on articles a physician has actually read again
  • An automatic grading of evidence strength per article. That is not something a machine should assign

Defects we found

A safeguard only matters if it catches something.

Building these checks surfaced problems we had not seen. What we fixed is recorded here.

2026-08-17

We treated a bibliography as proof of physician review

The service that shares article information with other tools marked an article as reviewed whenever it could extract references from the body. References are not a review record.

Sources and documented physician reviews are now kept separate. At the check on August 17, 2026, the published articles had zero explicit per-article physician-review records. This count describes that historical check.

2026-08-07

We were showing a review date we had not verified

In the check on August 7, 2026, for 524 articles, the date presented as “reviewed by a physician” was in fact the date the articles were imported in bulk. It was not a date anyone had read them again.

The date now appears only where a real review happened, and a test was added so it cannot come back.

2026-08-07

Thirty source links did not open

Japanese annotations written next to a reference were being joined onto the end of the link, leaving it broken. Thirty of 234 links were affected.

The 30 affected links found at that time were repaired. A test was added to check for the same defect recurring.

Ahead of us

What is not in place yet.

  • A per-article physician-review record for the full archive currently online
  • Per-article review deadlines in day-to-day operation (designed, not yet running)
  • Approval by more than one physician, with a published record of that approval
  • Rechecking and repairing source URLs that no longer resolve. This page previously listed 16 outstanding links; that number has not been reconfirmed as the current total
  • Detecting contradictions we have not seen before (today we only prevent known ones from recurring)

Please tell us when something is wrong.

If you notice an unclear explanation, a broken link, or an error, please contact us. The article URL and the passage concerned help us check the issue. You do not need to include your child's name or other identifying information.

Our policy is to check reported issues and revise articles where corrections are needed. Documented revisions appear in the update history. This contact channel does not provide individual diagnoses or urgent medical advice.