Publications — what we have shown, and under which assumptions.
Our papers are self-archived preprints on Zenodo and not peer-reviewed. The series was revised mathematically in September 2026. An independent review is pending; we say so because it matters for how the work should be read. Each entry links the concept DOI — it always resolves to the latest version, including after the next upload. Earlier versions remain available under their version DOIs.
The papers are written in German. English editions exist for FM-3 only, and that edition has not yet been brought to the current version.
The series at a glance
FM-1 — Foundations for the scientific evaluation of clinical information
We place patients, encounters, diagnoses, procedures and measurements on a common time axis, keeping apart which event is meant, when it occurred and what it belongs to. For you this means: length of stay, occupancy figures and cohorts can be defined unambiguously and recomputed. We keep descriptive measures separate from causal or clinically validated claims.
FM-2 — Extended mathematical model of clinical information
We extend the base model by event types that inherit properties, by process relations and by a language for temporal connections. Existing data and queries remain valid under stated assumptions. For you this means: a data model can grow without old analyses silently changing their meaning. The assumptions are stated in the paper.
FM-3 — AION: Algebraic Interval Ontology for Clinical Networks
AION adds uncertain time specifications, temporal patterns, FHIR mappings, federated evaluation and explainable components. For each extension we state the conditions under which it holds. A small executable FHIR experiment is included. For you this means: you can see what the model can do and where its assumptions end — without any claim of clinical efficacy.
FM-4 — SILD: Signal-Loss Inspection at Data-boundaries
We describe when a data transfer loses distinguishable information, and assign such losses to four patterns: type, time, additional attributes and references. Under explicitly stated assumptions, every lossy transfer falls under at least one of them. For you this means: an inspection can ask specifically what may be lost. Which assumptions your route satisfies is something we clarify case by case.
FM-4.1 — Chain theorems, observation boundary and loss decomposition
We examine chains of several systems. For every lost item there is exactly one first stage at which it is missing. Determining that stage requires knowledge about the processing. A downstream inspector can establish that a loss occurred without recovering the original value. For you this means: where you place measuring points determines how precisely a cause can be narrowed down.
FM-4.2 — Meaning preservation and branched transfers
We separate three questions that are often conflated in practice: is an item reconstructible in principle? Does a concrete program reproduce it correctly? Is its provenance evidenced? Branching and merging are added to this. For you this means: a populated field is not yet evidence. A finer classification of error kinds is explicitly marked as open research.
Determination of limits — Where FHIR profiles no longer carry the semantic load
We show what profile validation can achieve and where its limit lies: an inspection that only accepts or rejects target data can rule out an ambiguity, but cannot recover a lost distinction. For you this means: validation and boundary inspection complement each other. We do not claim a general impossibility of later inspection.
What we have retracted
- We no longer say that information loss is detectable only at the transfer boundary. A later inspector can establish a loss given a known model — recovering the original value is not possible.
- We no longer say that four loss patterns capture every error completely. That holds only under explicitly stated assumptions.
- We no longer cite earlier performance and stage figures as evidence. They rested on stored inspection markers, not on semantic validation.
Versions and DOIs
As of 17 September 2026. The versions carry the date 13 September 2026; they have been on Zenodo since 16 September 2026.
| Paper | Version | Date | Concept DOI |
|---|---|---|---|
| FM-1 | 1.3 | 13 Sep 2026 | 10.5281/zenodo.19205556 |
| FM-2 | 1.3 | 13 Sep 2026 | 10.5281/zenodo.19329921 |
| FM-3 (German) | 1.5 | 13 Sep 2026 | 10.5281/zenodo.20646841 |
| FM-3 (English)not yet updated to version 1.5 | 1.3 | 30 Jul 2026 | 10.5281/zenodo.19548856 |
| FM-4 | 3.2 | 13 Sep 2026 | 10.5281/zenodo.20375434 |
| FM-4.1shared record with FM-4.2 | 0.10 | 13 Sep 2026 | 10.5281/zenodo.22788842 |
| FM-4.2shared record with FM-4.1 | 0.2 | 13 Sep 2026 | 10.5281/zenodo.22788842 |
| Determination of limits | 7 | 13 Sep 2026 | 10.5281/zenodo.21318995 |
| Scaling layernot part of the revision | 1.4 | 28 Aug 2026 | 10.5281/zenodo.20095469 |
| CAIRNsoftware record — not a paper | 1.0.5 | 29 Aug 2026 | 10.5281/zenodo.19483181 |
The unit of counting is the paper: FM-3 counts once although it has two records (German and English); FM-4.1 and FM-4.2 count as two although they share one record. That yields eight papers. CAIRN is a software record and is not counted.