PRODUCTS & SYSTEMS · RELEASED · PUBLIC BETA · 2026
Uibang Seongdo (醫方星圖) went live as a public beta on 13 August 2026. It lays out 38,120 unique source IDs for formulas, medicinal materials, and diseases or symptoms from Korea’s Traditional Knowledge Portal as a field of stars.
I handled the concept, public-data model, quality and display rules, interface implementation, curation, and release documentation. The site and its preservation record are public. Its user base, educational value, and institutional adoption have not been established.
First verified external feature · checked 4 September 2026
Uibang Seongdo appeared in the HOT section of KMCRIC’s official homepage under “[Korean Medicine AI] 醫方星圖: Prescriptions–Herbs–Conditions.” The listing links directly to the site.
View the KMCRIC homepage ↗ · The page does not show a posting date or describe the listing as selection, certification, or endorsement.
38,120
unique formula, herb, and disease source IDs
231,319
deduplicated undirected name relations
13 Aug 2026
public-beta release date
The rule came before the star map
The stubborn question was when two records that looked alike should remain separate. Searching for cheukbaeksan (측백산), for example, returns three records with the same displayed Korean and Chinese name. Their source IDs differ, so Uibang Seongdo keeps three stars. The screen is busier, but a reader can still trace each result back to its source.
Formula records follow the same rule. Two entries for Hyeongbangjihwang-tang, P0004443 and P0018333, retain separate permanent pages even though their labels match. Merging on the name alone could erase differences in provenance or composition. Each page therefore keeps the source ID, source label, and link to the provider.
A line needed a narrow definition
The public build contains 231,319 deduplicated, undirected source-label relations. Of these, 88,880 are declared by both records and 142,439 appear on one side only. The interface and exported records distinguish those two states.
A line means that a name appears in the relevant public-source field. It says nothing about clinical efficacy, disease causation, or similarity between medicines. Nor has every line been checked by a person against a volume, chapter, page, and line in a primary text. Those boundaries sit beside the graph because they change how the graph can be used.
One comparison made the idea concrete
Selecting one Donguibogam record for Ssanghwatang and one for Sipjeondaebotang produces six shared ingredient names. The selected Ssanghwatang record has one additional name, while the Sipjeondaebotang record has four. The comparison stays at the level of names written in those two records; it does not mix in doses or infer equivalence.
The useful result is modest: a reader can see exactly where two public records overlap and where they differ, with the record numbers and provenance still in view. It is not a judgement that one prescription is better, safer, or interchangeable with the other.
Missing values stayed missing
The source displayed 20,157 formula rows, including 37 repeated source IDs. Counting unique IDs produces 20,120 formulas. Adding 5,500 medicinal-material IDs and 12,500 disease or symptom IDs gives the published total of 38,120. The number describes source records, not distinct medical concepts.
When a source field is empty, the interface does not silently complete it. The Chinese-name field for Yangno-tang record P0016491, for instance, remains an unresolved quality item. Name collisions and uncertain matches are exposed for review instead of being folded into cleaner-looking data.
The release included its paper trail
Alongside the permanent record pages, I published a data dictionary, relation rules, a coverage register, rights notes, a correction policy, a change log, and a SHA-256 integrity manifest. A casual visitor can use the map, while a reviewer can follow a source ID into the machine-readable documentation.
Zenodo preserves a documentation package for the 13 August release. DOI 10.5281/zenodo.21916379 identifies a 75,673-byte ZIP of rights-scoped documentation. It does not identify the full website source, third-party source datasets, or the bulk search and relation assets.
What is established now
The public beta is live, its maker, version, method, and rights boundaries are documented, and its limited preservation package has a persistent identifier. On 4 September 2026, it was also visible in the HOT section of KMCRIC’s official homepage with a direct link to Uibang Seongdo. That external feature is verifiable; it is not evidence of selection, certification, or institutional adoption.
Sustained use, learning or research benefit, institutional adoption, and independent peer review remain unverified. Those signals should come before another round of feature expansion.
Project record
Period: August 2026
Role: concept, public-data modelling, quality and display rules, interface implementation, data curation, and release documentation
Outcome: public-beta web app, permanent record pages, evidence documents, and a limited preservation DOI
Externally visible: the live web app, public Zenodo record, and KMCRIC homepage HOT feature
Not established: adoption, educational or research impact, institutional endorsement, or peer review