English documentation | Deutsche Dokumentation
Note
Machine-readable context for AI agents: A concise repository overview for LLMs, RAG crawlers, and automated indexers is available in llms.txt.
Um:bruch — Think Tank for Social Change
Status: Working Paper / multi-perspective stocktake with pilot character — v0.22, April 2026 | CC BY 4.0
This is an ongoing research project. Its findings are based on publicly available sources and AI-assisted analysis, not on primary data collected by the project. All results are provisional and updated continuously. The work has not yet been submitted to a peer-reviewed journal. It is published as a Working Paper to make the research process transparent and invite feedback.
| Goal | Entry point |
|---|---|
| Read the study | pdf/ST-001_Studie_Regressangst.pdf |
| German executive summary | pdf/ST-001_Executive_Summary.pdf |
| English executive summary | pdf/ST-001_Executive_Summary_EN.pdf |
| Recoupment transparency portal | pdf/PP-003_Regress_Transparenzportal.pdf |
| Sources and research core | recherche/MASTER-CORE.md |
| Check the version | meta/VERSIONIERUNG.md |
Canonical repository: https://github.com/um-bruch/regressangst
Public project page: https://um-bruch.org/projekte/
Search context: Regressangst, Wirtschaftlichkeitsprüfung, ST-001, PP-003, Regress-Transparenzportal, ärztliche Regressangst, Um:bruch Regress-Melder.
graph TD
A["Statistical data sources<br>(BMG / GKV-SV / KV Bavaria)"] --> B["Three-layer model of recoupment anxiety"]
subgraph Layers["Definition divergence (factor 500x)"]
B --> C1["Layer 1: Official statistics<br>(<100 legally final cases/year, 0.065%)"]
B --> C2["Layer 2: Formal applications<br>(outlier reviews V1)"]
B --> C3["Layer 3: Individual-case reviews V2<br>(~47,000 proceedings/year, ~30% affected)"]
end
subgraph Evidence["Parallel evidence channels"]
D1["CME training<br>(educational, before an error)"]
D2["V2 review system<br>(punitive, after an error)"]
end
C3 --> D2
style B fill:#f9f,stroke:#333,stroke-width:2px
style Layers fill:#e1f5fe,stroke:#0288d1
style Evidence fill:#fff3e0,stroke:#f57c00
The German economic-efficiency review system (§§ 106 ff. SGB V) creates recoupment anxiety among contracted physicians. According to a model calculation, its downstream costs could exceed the recoupments enforced by a factor of 100 to 1,000.
The official recoupment rate is “below 1%”; yet 72% of general practitioners report having experienced a recoupment at least once (Ribbat et al. 2023, n=770, Technical University of Munich). What happens in the system that produces both at the same time?
These findings are based on analysis of publicly accessible data and literature. They are compared against the MASTER-CORE (recherche/MASTER-CORE.md), which serves as a cross-paper reference framework.
-
Definition divergence (original synthesis): “Below 1%” and “72%” are not contradictory; they measure different things. “Recoupment” is defined across three layers: Layer 1 (<100/year, V1 only, legally final, 0.065%) versus Layer 3 (about 47,000 V2 proceedings/year according to BMG GVSG 2024, about 30%). Factor: about 500x. The discrepancy is a measurement problem in the statistics, not a perception error by physicians. Original after a prior-art check.
-
V1 protection-tier model: V1 outlier review is fairer than public debate suggests (hearing, consultation, grace period, EUR 25,000 cap). The 99.7% rejection rate is not a system failure but a functioning protection model.
-
V2a/V2b distinction: Individual-case review (V2) divides into V2a (economic inefficiency, with medical grounds available as a defense) and V2b (formal errors, medically irrelevant, not curable, and without the protection of prior consultation). The certificate error applies at full strength only to V2b.
-
Dual function (original synthesis): About 35% of V2 reviews encode medical evidence as formal rules (Category A: biosimilar requirement, AM-RL). About 40% are pure formal checks (Category B). The system has two parallel evidence channels—CME (educational, before an error) and V2 (punitive, after an error)—with no connection between them. Original after a prior-art check.
-
Empirical confirmation: Bavaria: 13,332 recoupments set in 2024 among about 24,000 physicians, average EUR 334, about 70% minor cases (SpiFa). A Poisson model (λ=0.85) yields 57% annual affectedness; Ribbat’s 72% is mathematically plausible.
- This is a secondary analysis, not original primary research. All data come from public sources.
- The percentages for the dual function (35/40/25%) are estimates, not empirically measured.
- The downstream-cost calculation (EUR 0.9–1.7 billion/year) relies on scenario assumptions with transparent parameters.
- The causal conclusion “review pressure → undertreatment” is not empirically established (Ribbat measures self-report, not care outcomes).
- An independent replication of the Ribbat findings is still outstanding.
- 9 IFG requests have been submitted (deadline about 10 May 2026) and could materially change the evidence base.
| Document | Version | Pages | Description |
|---|---|---|---|
| ST-001 | v0.22 | 104 | Main study with 7 parts, a counter-position, concluding chapter, and appendices. P0/P1/P2 review and PDF comment rounds implemented. |
| ST-001 ES (DE) | v1.3 | 16 | German executive summary with asymmetry subsection, updated recoupment-rate range, and KNV table (15 measures, including care review). |
| ST-001 ES (EN) | v1.3 | 16 | English executive summary synchronized with ST-001 v0.22. |
| PP-003 | v3.1 | 40 | Concept paper for the recoupment transparency portal, including a demand catalogue, hosting-organization scenarios, and an updated cost calculation. |
| Part | Content | Method |
|---|---|---|
| I | Historical layers (container genealogy 1989–2026) | Legislative history, Bundestag documents |
| II | Doctrinal gap (certificate error, V2a/V2b) | Federal Social Court analysis, statutory interpretation |
| III | Behavioral economics (Ribbat, game theory, Poisson) | Secondary analysis, modeling |
| IV | Downstream costs (EUR 0.9–1.7 billion scenario) | Three-stage model, extrapolation |
| V | Institutional silence (GKV-SV, Public Choice) | Document analysis |
| VI | International comparison (France, Switzerland, UK, Netherlands) | Comparative analysis |
| VII | Solution space (14 measures, KNV ranking) | Cost-benefit analysis |
| Counter-position | 6 objections with responses | Self-critique, Devil’s Advocate |
| Appendix A | Glossary | — |
| Appendix B | Methodology (Walk of Analysis, 15 agents) | Process documentation |
| Appendix C | IFG requests (9 submitted) | Open research questions |
- MASTER-CORE.md — central findings and methodology; cross-paper reference framework for consistency
- METHODIK_RECHERCHE_PROTOKOLL.md — search strategies, sources, and findings by research question
- WALK_OF_ANALYSIS_EXTENDED.md — full analysis record (14 phases, 8 knowledge layers)
- SOURCE_Schutzinstrumente.txt — legal status of the four protective instruments (§ 29 BMV-Ä, BSG 27/12 R)
- 40 individual analyses in
recherche/einzelanalysen/(RECHERCHE_*,MODELL_*,NACHANALYSE_*,ANALYSE_*)
- VERSIONIERUNG.md — document hierarchy and version matrix
- MASTERPLAN_V2.md — six-phase completion plan
AI-assisted multi-stream analysis process (April 2026, about 60–80 hours of human and AI work). Four models worked in parallel roles: Claude Opus 4.6 (primary analysis), Copilot GPT-4o (concept refinement), Gemini Deep Research (theoretical framework), and a human (steering and direction decisions).
- Cross-source divergence rather than only cross-model comparison: 15 specialized agents, 25 data points, 14 independent domains
- Naive-search test: a simulated journalist finds “below 1%” among the first 5 results
- Prior-art checks: two systematic searches (20+ queries each) confirmed the originality of both syntheses
- Alternative hypothesis tested: H2 (system as evidence control) has 35–40% explanatory power—acknowledged rather than ignored
- 5 documented hypothesis revisions, including the dating of individual-case review, a Moosazadeh misattribution, and the cost model
- Agent prompts are documented as search strategies, but the full instruction texts are not always available
- Gemini outputs required systematic double-checking (documented hallucinations: Moosazadeh, French protest figures)
CC BY 4.0 — Um:bruch (2026): Regressangst — Systems-Theory Analysis of Economic-Efficiency Review (§§ 106 ff. SGB V). Working Paper ST-001, v0.22. URL: https://github.com/um-bruch/regressangst.
Third-party licenses and the dependency inventory: THIRD_PARTY_LICENSES.txt.