How this library is actually built.
What we searched, what we let in, what each label means and where the method runs out. The library currently holds 305 models. The most recent entry was added on 18 August 2026, and the literature was last swept on 7 September 2026.
Where the models come from
Structured searches of three databases, not one.
Candidates are discovered through PubMed, Europe PMC and DOAJ. Europe PMC and DOAJ matter because a great deal of frugal simulation is published in journals that PubMed does not index, and restricting to one database would have quietly excluded them.
Searches run in two shapes. The first is organised by specialty, asking what each discipline has built. The second is organised by procedural skill, asking who has built a model for knot tying, or airway access, or anastomosis, regardless of which specialty published it. The second shape found models the first missed, because a paper about a suturing pad does not necessarily announce a specialty.
Both shapes are combined with frugal wording, including low cost, inexpensive, homemade, improvised, do it yourself and frugal, and with low resource wording, including LMIC, austere and rural. A citation chase follows references from entries already in the library.
This biases the library towards work that describes itself as cheap, and away from models that never used that vocabulary. That is the right bias for the question the library asks, and it is a real limit on treating it as a complete survey of the field.
What gets in
One rule: somebody built it and described how.
That is the whole inclusion test. Not that it was validated, not that it was published in a high impact journal, not that it worked. A model nobody has evaluated yet is still a model somebody built, and filtering on validation would remove exactly the new and experimental work this library exists to surface. A genuinely new model is unvalidated by definition.
What is excluded:
- Reviews, editorials, commentaries and protocols. They are about models, they are not models.
- Evaluations of commercial products where no build method is given.
- Courses, curricula and programmes with no described object.
- Clinical studies in patients, and drug or device trials.
- Statistical, computational and machine learning models. A different sense of the word.
- Anything with no source a reader can follow. Every published entry carries a working link, enforced by the database rather than by good intentions.
How much of each paper we read
Screening and labelling are done from the abstract unless the card says otherwise.
This is the most important limitation on the page, so it is stated plainly rather than buried. Every card that carries a full appraisal says which it was based on, and an abstract-only appraisal records that as a limitation on the card itself. Nothing is presented as a full-text reading when it was not.
The practical consequence: a model whose abstract does not mention a requirement will be described as needing less than it does. Descriptions are our own summary of the source, written to be traceable to it, never copied from it and never embellished beyond it.
The five evaluation states
Every entry carries one. It labels the evidence, it never filters on it.
Somebody built it and wrote down how. No evaluation of any kind is reported.
119Learners rated it, usually for realism or confidence. That is opinion, not evidence that it works.
80Some formal validity evidence is reported, for example whether the model separates novices from experts.
63Tested against something else: another model, a cadaver, or a control group.
37The source does not let us tell. Recorded as unknown rather than guessed at.
6The short name in each row is the term used in write-ups about the library. The chip is what appears on the card. They are the same five states.
These are assigned from the source by the same abstract-level reading described above, and they have not been independently double coded. Read them as the library’s labelling, not as an appraisal anyone else has checked.
Cost bands, and what silence means
A stated figure beats an adjective, an adjective beats nothing, and nothing stays nothing.
Figures are quoted in the currency the source used and are never converted, because an exchange rate on the day of reading is not a fact about the model.
The last band does the real work. When a paper never states a price, the entry says cost not stated in source and is deliberately excluded from the frugal filter. A model is never called cheap because its paper stayed quiet, and the absence of a figure is reported rather than filled in.
A cost band describes what a model was said to cost. It does not describe what you need to have access to in order to build it. Those are different questions and the second is not yet answered on these cards.
How the library has grown
Every addition, by month and by specialty. Read from the catalogue itself, so it cannot flatter.
Library reached 305 models
Dates are when an entry was added to this library, not when its paper was published. The catalogue grows in batches, when a harvest is screened, rather than continuously.
What this method cannot tell you
- It is a curated repository, not a systematic review. There was no protocol-driven search, no dual screening and no risk-of-bias assessment.
- Coverage is uneven. 18 specialties are represented and the largest holds many times what the smallest does. The gaps are visible on purpose rather than smoothed over.
- Labelling is single-coder. No inter-rater agreement statistic has been calculated.
- Country of origin is not recorded, so nothing here can be broken down by where a model was built.
- Nothing here is a recommendation. A model appearing in this library means somebody built it and wrote it down. It does not mean it is good, safe, or right for your trainees. That judgement stays with you.
If you find something wrong, tell us and it gets corrected. Contact, or submit a model.