feat: Medical stress test (safety+security coupled) + Missing Convergence report (Phase Ω #3)

Medical before Payment: the harder scientific test (safety AND security coupled, full lifecycle,
deep risk/evidence demands). ISO 13485 runs through the SAME engine as ISO 27001 -> CRA, only new
data, 0 runtime. The key result: IEC 81001-5-1 (health-software security) pulls in the SAME security
MCAPs as the CRA, so Medical REUSES cyber capabilities (the safety/security coupling appears as
capability reuse) while adding 7 genuinely new medical caps (clinical evaluation, software safety
classification, ISO 14971 risk file, benefit-risk). rejected_assumptions intact.

Effect on the convergence core: secure_signed_update_distribution 18 -> 24 and
technical_vulnerability_management 17 -> 23, now spanning 3 domains (cyber + industrial + medical) —
the core visibly GROWS, exactly the convergence signal.

New 5th report: MISSING CONVERGENCE — deterministic (no ML) token-cluster detector for potential
structural duplications: a name token shared by >=3 MCAPs across >=2 distinct sources is flagged for
EXPERT REVIEW (never auto-merged). Surfaces e.g. the `risk` cluster (6 risk MCAPs across 6 sources)
and `security`/`software`; single-source decompositions are filtered out. Complements Suspicious by
looking at cross-source duplication, not single MCAPs.

Also records the durable modelling rule extracted from the frequency fix: evidence is attributed to
its ORIGIN; its value against a target is computed later (relevance(evidence,target)). Ledger now 8
sources, Architecture Stability 8/8 = 100%. Non-runtime -> no deploy. 29 tests pass, check-loc 0.

backend-compliance/knowledge/architecture_stability/integration_ledger.yaml

@@ -86,6 +86,16 @@ sources:
     integration_kind: data_only
     family: cyber               # convergence test: same capability fed by many sources, model stayed stable
     exercised_by: "automotive_convergence_stress_test"
+  - source: "ISO 13485 -> Medical device (MDR / IEC 62304 / ISO 14971 / IEC 81001-5-1)"
+    domain: medical
+    target_type: regulation
+    integrated_as: transition_pattern_data
+    new_runtime_classes: 0
+    new_pipeline: false
+    new_capability_types: 7     # of 11 delta caps, 4 REUSE cyber MCAPs (IEC 81001-5-1 = CRA security coupling)
+    integration_kind: data_only
+    family: non_cyber           # safety/clinical domain WITH a security coupling — the harder joint test
+    exercised_by: "medical_stress_test"
 
 # --- One-time, domain-AGNOSTIC pipeline functions (built once, now FROZEN per Phase Ω). ---
 # Listed for honesty so the stability KPI cannot be gamed: these are NOT per-domain costs. The last

backend-compliance/knowledge/transition_patterns/transition_pattern_iso13485_to_medical_v1.yaml

@@ -0,0 +1,82 @@
+# Transition KNOWLEDGE Pattern (TKP) — ISO 13485 (Medical QMS)  ->  Medical device compliance
+# The HARDER scientific test (Phase Ω, test #3). Medical brings, at once, properties not yet jointly
+# tested: safety and security TIGHTLY COUPLED, a full product lifecycle, very strong risk-management +
+# evidence demands, and high regulatory depth. Sources: MDR, IEC 62304 (software lifecycle),
+# ISO 14971 (risk management), IEC 81001-5-1 (health-software security).
+#
+# KEY for the convergence analysis: IEC 81001-5-1 (health-software security) REQUIRES the same security
+# capabilities the CRA does (secure update, vulnerability management, access control, SBOM). So Medical
+# REUSES the cyber MCAPs — the safety/security coupling shows up as capability reuse, growing the core
+# into a 3rd domain. Alongside that, Medical adds genuinely new caps (clinical evaluation, software
+# safety classification, the ISO 14971 risk file, benefit-risk). Capabilities are VERBS. Expert L1 draft.
+
+id: TP-ISO13485-MEDICAL-v1
+status: draft
+version: 1
+
+transition_goal:
+  from:
+    standard: "ISO 13485"
+    edition: "2016"
+    nature: organizational_medical_qms
+  to:
+    domain: "Medical device compliance (software-containing device)"
+    nature: product_safety_and_security
+    sources: ["MDR", "IEC_62304", "ISO_14971", "IEC_81001_5_1"]
+  one_line: "Move a manufacturer whose management system is ISO 13485 toward placing a software-containing medical device on the EU market."
+
+provenance:
+  author: "Claude (Reasoning session) — AI first draft (L1)"
+  basis: "ISO 13485:2016 (4.2 documentation, 7.3 design, 7.1/risk, 8.2.1 feedback/PMS, 8.5 CAPA) vs MDR + IEC 62304 + ISO 14971 + IEC 81001-5-1."
+  reviewed_by: null
+  validated_by: null
+
+disclaimer: >
+  Curated expert knowledge, NOT a normative proof. KEY INSIGHT: ISO 13485 is a medical QUALITY
+  management system — it gives QMS + design-control + risk-process + PMS discipline, but not the
+  concrete clinical, software-lifecycle, risk-file or product-security EVIDENCE. Safety and security
+  are coupled: IEC 81001-5-1 pulls in the SAME security capabilities the CRA requires. Welt-1.
+
+source_state_variants:
+  certified:       "ISO 13485 certified -> QMS/design/risk-process assumptions hold; concrete evidence still missing."
+  qms_introduced:  "QMS implemented but not certified -> downgrade 'partially_supports' to needs_confirmation."
+
+# ── A) LIKELY COVERED — medical QMS management discipline (partially_supports), NOT evidence. ──
+likely_covered:
+  - {capability: manage_document_control, iso13485_basis: "4.2", relationship: partially_supports, confidence_source: relationship, verification: required, expected_evidence: [document_control_procedure], rationale: "ISO 13485 document control maintains records but creates no clinical/safety content.", reviewable_claim: "Document control maintains, not creates, medical evidence."}
+  - {capability: operate_capa_process, iso13485_basis: "8.5", relationship: partially_supports, confidence_source: relationship, verification: required, expected_evidence: [capa_procedure], rationale: "CAPA gives corrective/preventive discipline, not product safety evidence.", reviewable_claim: "CAPA does not establish device safety."}
+  - {capability: conduct_design_controls, iso13485_basis: "7.3", relationship: partially_supports, confidence_source: relationship, verification: required, expected_evidence: [design_history_file], rationale: "Design controls structure development but do not by themselves produce IEC 62304 software-lifecycle records.", reviewable_claim: "Design controls do not produce the IEC 62304 lifecycle records."}
+  - {capability: run_risk_management_process, iso13485_basis: "7.1", relationship: partially_supports, confidence_source: relationship, verification: required, expected_evidence: [risk_management_procedure], rationale: "ISO 13485 requires a risk-management PROCESS, but the ISO 14971 risk FILE with concrete analyses is separate.", reviewable_claim: "A risk-management process is not a completed ISO 14971 risk file."}
+  - {capability: control_suppliers_medical, iso13485_basis: "7.4", relationship: partially_supports, confidence_source: relationship, verification: required, expected_evidence: [supplier_controls], rationale: "Supplier controls give purchasing discipline, not component security.", reviewable_claim: "Medical supplier control does not establish component security."}
+  - {capability: operate_post_market_surveillance, iso13485_basis: "8.2.1", relationship: partially_supports, confidence_source: relationship, verification: required, expected_evidence: [pms_procedure], rationale: "Feedback/PMS process exists, but MDR PMS + vigilance reporting evidence is separate.", reviewable_claim: "A feedback process is not MDR post-market surveillance evidence."}
+
+# ── B) DELTA — concrete medical evidence + (coupled) product security. ──
+delta_requirements:
+  # B1) safety/security COUPLING — IEC 81001-5-1 reuses the SAME cyber MCAPs as the CRA:
+  - {capability: secure_signed_update_distribution, missing_because: "QMS has no secure update.", why_asked: "IEC 81001-5-1 requires authenticity/integrity-protected health-software updates (= CRA).", dropped_if: ["Updates are signed and verified."], needed_information: verify_existence, covers_targets: [IEC_81001_5_1, MDR], expected_evidence: [config_export], priority: high, reviewable_claim: "ISO 13485 does not establish signed update distribution."}
+  - {capability: technical_vulnerability_management, missing_because: "QMS has no vulnerability management.", why_asked: "IEC 81001-5-1 requires handling health-software vulnerabilities (= CRA).", dropped_if: ["A vulnerability management process exists."], needed_information: verify_existence, covers_targets: [IEC_81001_5_1, MDR], expected_evidence: [vulnerability_management_procedure], priority: high, reviewable_claim: "ISO 13485 does not establish vulnerability management."}
+  - {capability: access_control_and_authentication, missing_because: "QMS has no product access control.", why_asked: "IEC 81001-5-1 requires authentication/access control for health software (= CRA).", dropped_if: ["Authentication is documented + tested."], needed_information: verify_existence, covers_targets: [IEC_81001_5_1], expected_evidence: [access_control_policy], priority: medium, reviewable_claim: "ISO 13485 does not establish product authentication."}
+  - {capability: sbom_creation, missing_because: "QMS produces no SBOM.", why_asked: "IEC 81001-5-1 expects a software bill of materials (= CRA).", dropped_if: ["A machine-readable SBOM is produced per release."], needed_information: determine_sbom_maturity, covers_targets: [IEC_81001_5_1], expected_evidence: [sbom], priority: high, reviewable_claim: "ISO 13485 does not produce an SBOM."}
+  # B2) genuinely NEW medical-specific evidence:
+  - {capability: conduct_clinical_evaluation, missing_because: "QMS produces no clinical evidence.", why_asked: "The MDR requires a clinical evaluation / clinical evidence.", dropped_if: ["A clinical evaluation report exists."], needed_information: request_evidence, covers_targets: [MDR], expected_evidence: [clinical_evaluation_report], priority: high, reviewable_claim: "ISO 13485 does not produce clinical evidence."}
+  - {capability: classify_software_safety_iec62304, missing_because: "QMS does not classify software safety.", why_asked: "IEC 62304 requires a software safety classification (Class A/B/C) driving the lifecycle.", dropped_if: ["A software safety classification exists."], needed_information: request_evidence, covers_targets: [IEC_62304], expected_evidence: [software_safety_classification], priority: high, reviewable_claim: "ISO 13485 does not classify software safety per IEC 62304."}
+  - {capability: maintain_risk_management_file_iso14971, missing_because: "QMS has a process, not the file.", why_asked: "ISO 14971 requires a maintained risk management file with concrete analyses + controls.", dropped_if: ["An ISO 14971 risk management file exists."], needed_information: request_evidence, covers_targets: [ISO_14971, MDR], expected_evidence: [risk_management_file], priority: high, reviewable_claim: "An ISO 13485 risk process is not a completed ISO 14971 risk file."}
+  - {capability: perform_benefit_risk_analysis, missing_because: "QMS does not weigh benefit vs risk.", why_asked: "The MDR + ISO 14971 require a documented benefit-risk determination.", dropped_if: ["A benefit-risk analysis exists."], needed_information: request_evidence, covers_targets: [MDR, ISO_14971], expected_evidence: [benefit_risk_analysis], priority: medium, reviewable_claim: "ISO 13485 does not document a benefit-risk analysis."}
+  - {capability: implement_software_lifecycle_iec62304, missing_because: "QMS has design controls, not the IEC 62304 lifecycle.", why_asked: "IEC 62304 requires a defined software development + maintenance lifecycle.", dropped_if: ["IEC 62304 lifecycle records exist."], needed_information: request_evidence, covers_targets: [IEC_62304], expected_evidence: [software_lifecycle_records], priority: high, reviewable_claim: "ISO 13485 does not produce IEC 62304 lifecycle records."}
+  - {capability: assign_unique_device_identification, missing_because: "QMS has no UDI.", why_asked: "The MDR requires UDI assignment + registration.", dropped_if: ["UDI is assigned and registered."], needed_information: verify_existence, covers_targets: [MDR], expected_evidence: [udi_record], priority: medium, reviewable_claim: "ISO 13485 does not assign UDI."}
+  - {capability: compile_medical_technical_documentation, missing_because: "QMS has no MDR technical documentation.", why_asked: "The MDR (Annex II/III) requires complete technical documentation + DoC.", dropped_if: ["MDR technical documentation exists."], needed_information: request_evidence, covers_targets: [MDR], expected_evidence: [technical_documentation, declaration_of_conformity], priority: high, reviewable_claim: "ISO 13485 does not satisfy MDR technical documentation."}
+
+rejected_assumptions:
+  - "ISO 13485 does NOT produce clinical evidence or a clinical evaluation."
+  - "ISO 13485 does NOT produce the ISO 14971 risk management FILE (only the process)."
+  - "ISO 13485 does NOT produce IEC 62304 software-lifecycle records or a software safety classification."
+  - "ISO 13485 does NOT establish health-software security (IEC 81001-5-1 = the same security caps as the CRA)."
+
+determinism_goal: >
+  Two independent auditors should agree that an ISO-13485-only manufacturer has medical QMS discipline
+  but is missing the clinical, software-lifecycle, risk-file and product-security evidence — and that the
+  security part is the SAME capability set as the CRA (safety/security coupling).
+
+review_checklist:
+  - "Confirm delta + rejected_assumptions with a medical regulatory expert."
+  - "Replace capability ids with Capability Registry MCAP ids once assigned."

backend-compliance/reference_scenarios/architecture_stability_kpi.md

@@ -13,11 +13,12 @@ _Der Fokus hat sich verschoben: nicht mehr „kann die Architektur das?", sonder
 | OEM Specification (Lastenheft) | cyber | 0 | 0 | 4 | ✅ |
 | ISO 14001 -> Environmental/Material (REACH/RoHS/Batterie/Wasser/Energie/Abfall) | non_cyber | 0 | 0 | 16 | ✅ |
 | Automotive ECU for OEM X (CRA / UNECE R155+R156 / IATF 16949 / TISAX / ASPICE / OEM spec) | cyber | 0 | 0 | 14 | ✅ |
+| ISO 13485 -> Medical device (MDR / IEC 62304 / ISO 14971 / IEC 81001-5-1) | non_cyber | 0 | 0 | 7 | ✅ |
 
-- **Architecture Stability: 7/7 = 100%** der Quellen ohne neue Runtime-Klasse und ohne neue Pipeline.
+- **Architecture Stability: 8/8 = 100%** der Quellen ohne neue Runtime-Klasse und ohne neue Pipeline.
-- **Knowledge Velocity: 7/7 = 100%** der Quellen **data-only** integriert (kein Entwickler nötig).
+- **Knowledge Velocity: 8/8 = 100%** der Quellen **data-only** integriert (kein Entwickler nötig).
-- **Generalität über Cyber hinaus: 1/7 Quellen NICHT-Cyber** (Umwelt) — trugen die Pipeline ebenfalls 0/0. Das ist der eigentliche Test (ein anderes Denkmodell, nicht noch ein Cyber-Regelwerk).
+- **Generalität über Cyber hinaus: 2/8 Quellen NICHT-Cyber** (Umwelt) — trugen die Pipeline ebenfalls 0/0. Das ist der eigentliche Test (ein anderes Denkmodell, nicht noch ein Cyber-Regelwerk).
-- **Capability-Modell-Frühindikator: 59 neue Typen gesamt, Maximum 16** (Umwelt, erste Nicht-Cyber-Domäne) — in Range, KEIN Granularitätsalarm (Alarm ≈ eine Domäne braucht plötzlich ~80 neue Typen bei 0 Runtime-Change → Modell zu grob/fein).
+- **Capability-Modell-Frühindikator: 66 neue Typen gesamt, Maximum 16** (Umwelt, erste Nicht-Cyber-Domäne) — in Range, KEIN Granularitätsalarm (Alarm ≈ eine Domäne braucht plötzlich ~80 neue Typen bei 0 Runtime-Change → Modell zu grob/fein).
 
 ## Ehrlichkeit: die Pipeline-Funktionen sind EINMALIG (jetzt eingefroren)
 - 6 domänen-AGNOSTISCHE Funktionen, einmal gebaut, nicht je Domäne: `transition_reasoning (RS-005)`, `optimization`, `journey_matcher (ADR-011)`, `playbook`, `completeness`, `company (2A)`.
@@ -31,8 +32,8 @@ _Der Fokus hat sich verschoben: nicht mehr „kann die Architektur das?", sonder
 | **Produktion** (wie TUN/BEWEISEN) | Playbooks, Verification, Reference Scenarios |
 
 ## Die drei Erfolgsfragen ab jetzt (statt Coverage)
-1. **Musste für eine neue Domäne Runtime-Code geändert werden?** → bisher: **nein** (7/7).
+1. **Musste für eine neue Domäne Runtime-Code geändert werden?** → bisher: **nein** (8/8).
-2. **Knowledge Velocity** — neues Wissen ohne Entwickler aufnehmbar? → bisher: **ja** (7/7 data-only).
+2. **Knowledge Velocity** — neues Wissen ohne Entwickler aufnehmbar? → bisher: **ja** (8/8 data-only).
 3. **Architecture Stability** — bestehende Capability/Journey strukturell ändern oder nur Daten ergänzen? → bisher: **nur Daten**.
 
 > **Befund:** Über fünf Zielarten und sechs Quellen blieb `Reality → Evidence → Capability → Required → Delta → Journey → Roadmap → Playbooks → Verification` unverändert. Das ist der eigentliche Nachweis: keine Compliance-Architektur, sondern eine allgemeine Requirements-Verifikationsarchitektur, die ihre Generalität UNTER realer fachlicher Belastung behält. Der nächste Test ist nicht ein Feature, sondern die nächste echte Domäne (Umwelt-Cluster · Automotive · Medizintechnik · Payment) — jede als neue Ledger-Zeile, bei stabilem KPI.

backend-compliance/reference_scenarios/mcap_convergence_analysis.md

@@ -1,47 +1,59 @@
 # Cross-Domain MCAP Convergence Analysis — wo konvergiert das Wissensmodell?
 
-_Nicht „welche MCAPs kommen am häufigsten vor?" (Häufigkeit täuscht), sondern „welche MCAPs TRAGEN den größten Teil des Systems?". Deterministischer **Impact-Score** (kein ML), internes Engineering-Werkzeug, reine Aggregation über vorhandene Daten (5 Transition Patterns + 7 Automotive-Quellen). Non-runtime, keine echten Namen._
+_Nicht „welche MCAPs kommen am häufigsten vor?" (Häufigkeit täuscht), sondern „welche MCAPs TRAGEN den größten Teil des Systems?". Deterministischer **Impact-Score** (kein ML), internes Engineering-Werkzeug, reine Aggregation über vorhandene Daten (6 Transition Patterns inkl. Medical + 7 Automotive-Quellen). Non-runtime, keine echten Namen._
 
 ## Impact-Score (deterministisch)
 > `Impact = distinct Sources + distinct Target-Types + distinct Domains + distinct Journeys + Regulatory Leverage + Business Leverage`
-- 62 distinct Capabilities (MCAP-Kandidaten) über alle Quellen aggregiert.
+- 75 distinct Capabilities (MCAP-Kandidaten) über alle Quellen aggregiert.
 
 ## 1. Core MCAPs — höchster Impact (die tragenden Knoten)
 | Capability | Impact | Sources | Types | Domains | Journeys |
 |---|---:|---:|---:|---:|---:|
-| `secure_signed_update_distribution` | **18** | 5 | 2 | 2 | 4 |
+| `secure_signed_update_distribution` | **24** | 7 | 2 | 3 | 5 |
-| `technical_vulnerability_management` | **17** | 5 | 3 | 2 | 4 |
+| `technical_vulnerability_management` | **23** | 7 | 3 | 3 | 5 |
-| `access_control_and_authentication` | **15** | 4 | 2 | 2 | 5 |
+| `access_control_and_authentication` | **19** | 5 | 2 | 3 | 6 |
 | `incident_management` | **14** | 4 | 2 | 2 | 4 |
 | `product_cyber_risk_assessment` | **13** | 3 | 1 | 2 | 4 |
+| `sbom_creation` | **13** | 2 | 1 | 3 | 5 |
 | `secure_development_lifecycle` | **11** | 2 | 2 | 2 | 4 |
 | `supplier_security` | **11** | 3 | 2 | 2 | 3 |
 | `ce_conformity_assessment_and_technical_documentation` | **9** | 2 | 1 | 1 | 3 |
 | `coordinated_vulnerability_disclosure` | **9** | 1 | 1 | 2 | 4 |
-| `sbom_creation` | **9** | 1 | 1 | 2 | 4 |
 
 → Hoher Impact = ein Knoten verbindet viele Quellen ÜBER Typen/Domänen/Journeys hinweg — nicht „in 40 Dokumenten einer Normenfamilie".
 
 ## 2. Emerging MCAPs — verbinden ≥2 Domänen (Brücken zwischen Anforderungswelten)
-- `secure_signed_update_distribution` — 2 Domänen (automotive, industrial_automation), 2 Typen.
+- `secure_signed_update_distribution` — 3 Domänen (automotive, industrial_automation, medical), 2 Typen.
-- `technical_vulnerability_management` — 2 Domänen (automotive, industrial_automation), 3 Typen.
+- `technical_vulnerability_management` — 3 Domänen (automotive, industrial_automation, medical), 3 Typen.
-- `access_control_and_authentication` — 2 Domänen (automotive, industrial_automation), 2 Typen.
+- `access_control_and_authentication` — 3 Domänen (automotive, industrial_automation, medical), 2 Typen.
 - `incident_management` — 2 Domänen (automotive, industrial_automation), 2 Typen.
 - `product_cyber_risk_assessment` — 2 Domänen (automotive, industrial_automation), 1 Typen.
+- `sbom_creation` — 3 Domänen (automotive, industrial_automation, medical), 1 Typen.
 - `secure_development_lifecycle` — 2 Domänen (automotive, industrial_automation), 2 Typen.
 - `supplier_security` — 2 Domänen (automotive, industrial_automation), 2 Typen.
-- `coordinated_vulnerability_disclosure` — 2 Domänen (automotive, industrial_automation), 1 Typen.
 - _(Echtes „Wachstum über Zeit" braucht historische Snapshots — hier Proxy = Domänen-Spannweite jetzt.)_
 
 ## 3. Isolated MCAPs — nur 1 Quelle/Journey (Review: spezialisiert ODER Konvergenz übersehen?)
-- 36 Stück, u. a.: `account_energy_consumption`, `cybersecurity_management_system`, `document_update_campaigns`, `document_waste_streams`, `issue_battery_passport`, `machine_safety_risk_assessment`, `measure_air_emissions`, `mechanical_safety_and_guards`.
+- 47 Stück, u. a.: `account_energy_consumption`, `assign_unique_device_identification`, `classify_software_safety_iec62304`, `compile_medical_technical_documentation`, `conduct_clinical_evaluation`, `cybersecurity_management_system`, `document_update_campaigns`, `document_waste_streams`.
 
 ## 4. Suspicious MCAPs — Abstraktionsgrad-Verdacht (Experten-Review)
 - **Evtl. zu grob** (generisches Verb, breit aber nur 1 Typ): `document_and_change_control`, `manage_chemical_substances`.
-- **Evtl. zu fein** (isoliert + sehr spezifischer Name): `operating_instructions_and_safety_information`, `provide_dedicated_security_contact`, `provide_functional_safety_evidence`, `restrict_hazardous_substances_rohs`, `secure_by_default_no_default_credentials`, `threat_analysis_and_risk_assessment`.
+- **Evtl. zu fein** (isoliert + sehr spezifischer Name): `assign_unique_device_identification`, `compile_medical_technical_documentation`, `implement_software_lifecycle_iec62304`, `operating_instructions_and_safety_information`, `provide_dedicated_security_contact`, `provide_functional_safety_evidence`.
 - Die Analyse sagt damit nicht nur WELCHE MCAPs wichtig sind, sondern auch, ob sie auf dem **richtigen Abstraktionsniveau** definiert sind.
 
+## 5. Missing Convergence — mögliche strukturelle Doppelungen (Experten-Review, KEIN Auto-Merge)
+> Nicht „welche MCAPs existieren?", sondern „welche hätte ich aufgrund der Daten ERWARTET, existieren aber als GETRENNTE MCAPs?". Deterministische Heuristik (kein ML): geteiltes Namens-Token über ≥3 MCAPs UND ≥2 verschiedene Quellen = Konvergenz-Kandidat (eine Einzelquelle-Zerlegung zählt NICHT). Nur anzeigen, nie automatisch zusammenführen.
+
+- **Token `secure` → 3 MCAPs / 8 Quellen:** `secure_by_default_no_default_credentials`, `secure_development_lifecycle`, `secure_signed_update_distribution` — _Review: eine Fähigkeit in mehreren Facetten?_
+- **Token `update` → 4 MCAPs / 7 Quellen:** `document_update_campaigns`, `secure_signed_update_distribution`, `security_update_support_period`, `software_update_management_system` — _Review: eine Fähigkeit in mehreren Facetten?_
+- **Token `risk` → 6 MCAPs / 6 Quellen:** `machine_safety_risk_assessment`, `maintain_risk_management_file_iso14971`, `perform_benefit_risk_analysis`, `product_cyber_risk_assessment`, `run_risk_management_process`, `threat_analysis_and_risk_assessment` — _Review: eine Fähigkeit in mehreren Facetten?_
+- **Token `document` → 5 MCAPs / 6 Quellen:** `document_and_change_control`, `document_update_campaigns`, `document_waste_streams`, `manage_document_control`, `treat_and_document_wastewater` — _Review: eine Fähigkeit in mehreren Facetten?_
+- **Token `security` → 8 MCAPs / 4 Quellen:** `information_security_management`, `physical_security`, `provide_dedicated_security_contact`, `public_security_advisories`, `security_awareness_training`, `security_logging_and_monitoring`, `security_update_support_period`, `supplier_security` — _Review: eine Fähigkeit in mehreren Facetten?_
+- **Token `safety` → 6 MCAPs / 4 Quellen:** `classify_software_safety_iec62304`, `machine_safety_risk_assessment`, `mechanical_safety_and_guards`, `operating_instructions_and_safety_information`, `protection_against_corruption_of_safety_functions`, `provide_functional_safety_evidence` — _Review: eine Fähigkeit in mehreren Facetten?_
+
+→ Ergänzt `Suspicious`: dieser Report schaut nicht auf einzelne MCAPs, sondern auf **strukturelle Doppelungen** über Quellen hinweg — der häufigste systematische Modellierungsfehler.
+
 ## Befund
 
-> **Ein Kern beginnt sich zu zeigen:** 11 von 62 Capabilities erreichen Impact ≥ 8 (tragende Knoten), 14 verbinden ≥2 Domänen. Bislang ist das Wissensmodell noch jung (5 Patterns + 1 Automotive-Profil), aber die Methode steht: sobald Medical/Payment/weitere Domänen als DATEN hinzukommen, zeigt dieselbe Aggregation, ob sich der erwartete stabile Kern von 30–50 hochkonvergenten MCAPs bildet — der gemeinsame Strukturkern hinter sehr unterschiedlichen Anforderungswelten. Das ist ein tieferer Wertnachweis als „eine weitere Norm unterstützt". Reine Aggregation, 0 Runtime, 0 neue Architektur.
+> **Ein Kern beginnt sich zu zeigen:** 11 von 75 Capabilities erreichen Impact ≥ 8 (tragende Knoten), 14 verbinden ≥2 Domänen. Mit Medical (6 Patterns inkl. ISO 13485 + Automotive) zeigt sich erstmals die **Safety/Security-Kopplung als Capability-REUSE**: IEC 81001-5-1 zieht dieselben Security-MCAPs wie die CRA herein → diese Knoten spannen jetzt Cyber + Maschinenbau + Automotive + Medizin. Die Methode steht; sobald Payment/weitere Domänen als DATEN kommen, zeigt dieselbe Aggregation (+ der neue Missing-Convergence-Report), ob sich der erwartete stabile Kern von 30–50 hochkonvergenten MCAPs bildet — der gemeinsame Strukturkern hinter sehr unterschiedlichen Anforderungswelten. Tieferer Wertnachweis als „eine weitere Norm unterstützt". Reine Aggregation, 0 Runtime, 0 neue Architektur.
 

backend-compliance/reference_scenarios/mcap_convergence_analysis.py

@@ -42,6 +42,8 @@ PATTERN_META = {
     "transition_pattern_isms_to_tisax_v1.yaml":               ("automotive", "certification", ["TISAX"]),
     "transition_pattern_iso14001_to_environmental_v1.yaml":   ("environmental", "regulation",
         ["REACH", "RoHS", "Batterieverordnung", "Wasserrecht", "Abwasservorschriften", "Energiemanagement", "Kreislaufwirtschaft", "Emissionsschutz"]),
+    "transition_pattern_iso13485_to_medical_v1.yaml":         ("medical", "regulation",
+        ["MDR", "IEC_62304", "ISO_14971", "IEC_81001_5_1"]),
 }
 
 # capability -> dict of sets we aggregate
@@ -95,7 +97,7 @@ GENERIC = ("document_", "manage_", "control_", "conduct_", "operate_", "run_", "
 
 w("# Cross-Domain MCAP Convergence Analysis — wo konvergiert das Wissensmodell?")
 w("")
-w('_Nicht „welche MCAPs kommen am häufigsten vor?" (Häufigkeit täuscht), sondern „welche MCAPs TRAGEN den größten Teil des Systems?". Deterministischer **Impact-Score** (kein ML), internes Engineering-Werkzeug, reine Aggregation über vorhandene Daten (5 Transition Patterns + 7 Automotive-Quellen). Non-runtime, keine echten Namen._')
+w('_Nicht „welche MCAPs kommen am häufigsten vor?" (Häufigkeit täuscht), sondern „welche MCAPs TRAGEN den größten Teil des Systems?". Deterministischer **Impact-Score** (kein ML), internes Engineering-Werkzeug, reine Aggregation über vorhandene Daten (6 Transition Patterns inkl. Medical + 7 Automotive-Quellen). Non-runtime, keine echten Namen._')
 w("")
 w("## Impact-Score (deterministisch)")
 w("> `Impact = distinct Sources + distinct Target-Types + distinct Domains + distinct Journeys + Regulatory Leverage + Business Leverage`")
@@ -135,12 +137,42 @@ w("- **Evtl. zu fein** (isoliert + sehr spezifischer Name): %s." % (", ".join("`
 w("- Die Analyse sagt damit nicht nur WELCHE MCAPs wichtig sind, sondern auch, ob sie auf dem **richtigen Abstraktionsniveau** definiert sind.")
 w("")
 
+# ── 5. Missing Convergence — potential structural duplications (token clusters) ──────────────
+STOP = {"and", "of", "the", "for", "to", "a", "an", "no", "on", "in", "by", "with", "per"}
+GENERIC_TOK = {"management", "process", "control", "distribution", "documentation", "technical",
+               "internal", "continual", "assessment", "evidence"}
+tok2caps = {}
+for cap in idx:
+    for t in cap.split("_"):
+        if len(t) >= 4 and t not in STOP and t not in GENERIC_TOK:
+            tok2caps.setdefault(t, set()).add(cap)
+# cross-source signal only: a token shared by >=3 MCAPs that span >=2 distinct sources (one source's
+# own decomposition is NOT a duplication candidate — that is intended granularity).
+clusters = []
+for t, cs in tok2caps.items():
+    if len(cs) >= 3:
+        srcs = set()
+        for c in cs:
+            srcs |= idx[c]["sources"]
+        if len(srcs) >= 2:
+            clusters.append((t, sorted(cs), len(srcs)))
+clusters.sort(key=lambda x: (-x[2], -len(x[1]), x[0]))
+w("## 5. Missing Convergence — mögliche strukturelle Doppelungen (Experten-Review, KEIN Auto-Merge)")
+w('> Nicht „welche MCAPs existieren?", sondern „welche hätte ich aufgrund der Daten ERWARTET, existieren aber als GETRENNTE MCAPs?". Deterministische Heuristik (kein ML): geteiltes Namens-Token über ≥3 MCAPs UND ≥2 verschiedene Quellen = Konvergenz-Kandidat (eine Einzelquelle-Zerlegung zählt NICHT). Nur anzeigen, nie automatisch zusammenführen.')
+w("")
+for t, cs, ns in clusters[:6]:
+    w("- **Token `%s` → %d MCAPs / %d Quellen:** %s — _Review: eine Fähigkeit in mehreren Facetten?_" % (
+        t, len(cs), ns, ", ".join("`%s`" % c for c in cs)))
+w("")
+w("→ Ergänzt `Suspicious`: dieser Report schaut nicht auf einzelne MCAPs, sondern auf **strukturelle Doppelungen** über Quellen hinweg — der häufigste systematische Modellierungsfehler.")
+w("")
+
 # ── Befund ─────────────────────────────────────────────────────────────────
 core_cut = [c for c, e in scored if impact(e) >= 8]
 cross = [c for c, e in scored if len(e["domains"]) >= 2]
 w("## Befund")
 w("")
-w('> **Ein Kern beginnt sich zu zeigen:** %d von %d Capabilities erreichen Impact ≥ 8 (tragende Knoten), %d verbinden ≥2 Domänen. Bislang ist das Wissensmodell noch jung (5 Patterns + 1 Automotive-Profil), aber die Methode steht: sobald Medical/Payment/weitere Domänen als DATEN hinzukommen, zeigt dieselbe Aggregation, ob sich der erwartete stabile Kern von 30–50 hochkonvergenten MCAPs bildet — der gemeinsame Strukturkern hinter sehr unterschiedlichen Anforderungswelten. Das ist ein tieferer Wertnachweis als „eine weitere Norm unterstützt". Reine Aggregation, 0 Runtime, 0 neue Architektur.' % (len(core_cut), len(idx), len(cross)))
+w('> **Ein Kern beginnt sich zu zeigen:** %d von %d Capabilities erreichen Impact ≥ 8 (tragende Knoten), %d verbinden ≥2 Domänen. Mit Medical (6 Patterns inkl. ISO 13485 + Automotive) zeigt sich erstmals die **Safety/Security-Kopplung als Capability-REUSE**: IEC 81001-5-1 zieht dieselben Security-MCAPs wie die CRA herein → diese Knoten spannen jetzt Cyber + Maschinenbau + Automotive + Medizin. Die Methode steht; sobald Payment/weitere Domänen als DATEN kommen, zeigt dieselbe Aggregation (+ der neue Missing-Convergence-Report), ob sich der erwartete stabile Kern von 30–50 hochkonvergenten MCAPs bildet — der gemeinsame Strukturkern hinter sehr unterschiedlichen Anforderungswelten. Tieferer Wertnachweis als „eine weitere Norm unterstützt". Reine Aggregation, 0 Runtime, 0 neue Architektur.' % (len(core_cut), len(idx), len(cross)))
 w("")
 
 print("\n".join(OUT))

backend-compliance/reference_scenarios/medical_stress_test.md

@@ -0,0 +1,26 @@
+# Medical Stress Test — Safety + Security gekoppelt, der härtere Test (Phase Ω #3)
+
+_Medical prüft erstmals gemeinsam: Safety UND Security gekoppelt, voller Produktlebenszyklus, sehr starke Risikomanagement-/Nachweispflichten, hohe regulatorische Tiefe. ISO 13485 als Company Profile durch DIESELBE Engine — nur neue Daten, 0 Runtime. Synthetisch, keine echten Namen._
+
+## 1. ISO 13485 als Profil → Delta über dieselbe Engine
+- ISO 13485 liefert medizinische QMS-Disziplin (Welt-1): `manage_document_control`, `operate_capa_process`, `conduct_design_controls`, `run_risk_management_process`. …
+- Delta (fehlt): **11** Capabilities — über dieselbe `assess_transition`, **0 neue Runtime-Klassen**.
+
+## 2. Safety/Security-KOPPLUNG — Medical REUSED Cyber-MCAPs (IEC 81001-5-1 = CRA-Security)
+- **Wiederverwendete Cyber-Capabilities (4):** `access_control_and_authentication`, `sbom_creation`, `secure_signed_update_distribution`, `technical_vulnerability_management`.
+- → Genau das ist die Kopplung: die Gesundheitssoftware-Security (IEC 81001-5-1) fordert **dieselben** Fähigkeiten wie die CRA. Diese MCAPs wandern damit in eine **dritte Domäne** und werden im Convergence-Core noch zentraler.
+
+## 3. Genuin neue, medizin-spezifische Capabilities
+- **Neu (7):** `assign_unique_device_identification`, `classify_software_safety_iec62304`, `compile_medical_technical_documentation`, `conduct_clinical_evaluation`, `implement_software_lifecycle_iec62304`, `maintain_risk_management_file_iso14971`, `perform_benefit_risk_analysis`.
+- Capabilities als Verben: `conduct_clinical_evaluation`, `classify_software_safety_iec62304`, `maintain_risk_management_file_iso14971`, `perform_benefit_risk_analysis`.
+
+## 4. Was ISO 13485 typischerweise NICHT erzeugt _(rejected_assumptions, Welt-1/Welt-2)_
+- ISO 13485 does NOT produce clinical evidence or a clinical evaluation.
+- ISO 13485 does NOT produce the ISO 14971 risk management FILE (only the process).
+- ISO 13485 does NOT produce IEC 62304 software-lifecycle records or a software safety classification.
+- ISO 13485 does NOT establish health-software security (IEC 81001-5-1 = the same security caps as the CRA).
+
+## Befund
+
+> **Die härteste Domäne bisher (Safety+Security gekoppelt, voller Lebenszyklus, tiefe Risiko-/Nachweispflicht) lief durch die unveränderte Pipeline — 0 neue Runtime-Klassen, nur ein Pattern-YAML.** Das stärkste Einzelsignal: **4 Security-Capabilities werden aus dem Cyber-Bestand WIEDERVERWENDET** (IEC 81001-5-1 = CRA), während **7 medizin-spezifische** neu hinzukommen. Genau so wächst ein Kern: gemeinsame Fähigkeiten verbinden Cyber, Maschinenbau, Automotive UND Medizin; das Domänenspezifische bleibt am Rand. Architecture Stability bleibt stabil; der Engpass ist jetzt die Qualität der Wissensmodellierung, nicht die Architektur.
+

backend-compliance/reference_scenarios/medical_stress_test.py

@@ -0,0 +1,83 @@
+# ruff: noqa
+# mypy: ignore-errors
+"""Medical stress test — the harder scientific test: safety + security coupled (Phase Ω #3).
+
+Medical jointly tests properties not yet tested together: safety and security TIGHTLY COUPLED, a full
+product lifecycle, very strong risk-management/evidence demands, high regulatory depth. ISO 13485 (a
+medical QMS) is run as a Company Profile through the SAME engine as before — only new DATA, 0 runtime.
+
+The interesting result: IEC 81001-5-1 (health-software security) requires the SAME security capabilities
+as the CRA, so Medical REUSES cyber MCAPs (the coupling shows up as capability reuse, growing the core),
+while ALSO adding genuinely new medical caps (clinical evaluation, software safety classification, the
+ISO 14971 risk file, benefit-risk). Synthetic, no real names. Non-runtime -> no deploy.
+Run:  cd backend-compliance && PYTHONPATH=. python3 reference_scenarios/medical_stress_test.py
+"""
+from __future__ import annotations
+
+import os
+import yaml
+
+from compliance.company import (
+    CompanyContext, Certification, CapabilityMappingEntry, build_company_profile,
+)
+from compliance.reasoning.enums import Confidence
+from compliance.transition_reasoning import (
+    TransitionContext, TransitionGoal, TargetRequirement, assess_transition, CoverageStatus,
+)
+
+OUT = []
+
+
+def w(s=""):
+    OUT.append(s)
+
+
+_TP = os.path.join(os.path.dirname(__file__), "..", "knowledge", "transition_patterns")
+MED = yaml.safe_load(open(os.path.join(_TP, "transition_pattern_iso13485_to_medical_v1.yaml"), encoding="utf-8"))
+
+# existing capability universe (everything modelled before Medical) — to detect reuse vs new
+known = set()
+for f in os.listdir(_TP):
+    if f.endswith(".yaml") and "iso13485" not in f:
+        p = yaml.safe_load(open(os.path.join(_TP, f), encoding="utf-8"))
+        known |= {a["capability"] for a in p.get("likely_covered", [])}
+        known |= {d["capability"] for d in p.get("delta_requirements", [])}
+
+mgmt = [a["capability"] for a in MED["likely_covered"]]
+delta_caps = [d["capability"] for d in MED["delta_requirements"]]
+profile = build_company_profile(
+    CompanyContext(company_id="med", certifications=[Certification(certification_id="ISO13485")]),
+    {"ISO13485": CapabilityMappingEntry(capability_ids=mgmt, confidence=Confidence.MEDIUM)})
+assess = assess_transition(TransitionContext(company_id="med", target=TransitionGoal(target_id="Medical")),
+                           [TargetRequirement(capability_id=c) for c in mgmt + delta_caps], profile)
+delta = sorted({c.capability_id for c in assess.coverage if c.status == CoverageStatus.MISSING})
+
+reused = sorted(c for c in delta if c in known)     # safety/security coupling: cyber caps reused
+fresh = sorted(c for c in delta if c not in known)  # genuinely new medical caps
+
+w("# Medical Stress Test — Safety + Security gekoppelt, der härtere Test (Phase Ω #3)")
+w("")
+w('_Medical prüft erstmals gemeinsam: Safety UND Security gekoppelt, voller Produktlebenszyklus, sehr starke Risikomanagement-/Nachweispflichten, hohe regulatorische Tiefe. ISO 13485 als Company Profile durch DIESELBE Engine — nur neue Daten, 0 Runtime. Synthetisch, keine echten Namen._')
+w("")
+w("## 1. ISO 13485 als Profil → Delta über dieselbe Engine")
+w("- ISO 13485 liefert medizinische QMS-Disziplin (Welt-1): %s." % ", ".join("`%s`" % c for c in mgmt[:4]) + " …")
+w("- Delta (fehlt): **%d** Capabilities — über dieselbe `assess_transition`, **0 neue Runtime-Klassen**." % len(delta))
+w("")
+w("## 2. Safety/Security-KOPPLUNG — Medical REUSED Cyber-MCAPs (IEC 81001-5-1 = CRA-Security)")
+w("- **Wiederverwendete Cyber-Capabilities (%d):** %s." % (len(reused), ", ".join("`%s`" % c for c in reused)))
+w("- → Genau das ist die Kopplung: die Gesundheitssoftware-Security (IEC 81001-5-1) fordert **dieselben** Fähigkeiten wie die CRA. Diese MCAPs wandern damit in eine **dritte Domäne** und werden im Convergence-Core noch zentraler.")
+w("")
+w("## 3. Genuin neue, medizin-spezifische Capabilities")
+w("- **Neu (%d):** %s." % (len(fresh), ", ".join("`%s`" % c for c in fresh)))
+w("- Capabilities als Verben: `conduct_clinical_evaluation`, `classify_software_safety_iec62304`, `maintain_risk_management_file_iso14971`, `perform_benefit_risk_analysis`.")
+w("")
+w("## 4. Was ISO 13485 typischerweise NICHT erzeugt _(rejected_assumptions, Welt-1/Welt-2)_")
+for r in MED["rejected_assumptions"]:
+    w("- %s" % r)
+w("")
+w("## Befund")
+w("")
+w('> **Die härteste Domäne bisher (Safety+Security gekoppelt, voller Lebenszyklus, tiefe Risiko-/Nachweispflicht) lief durch die unveränderte Pipeline — 0 neue Runtime-Klassen, nur ein Pattern-YAML.** Das stärkste Einzelsignal: **%d Security-Capabilities werden aus dem Cyber-Bestand WIEDERVERWENDET** (IEC 81001-5-1 = CRA), während **%d medizin-spezifische** neu hinzukommen. Genau so wächst ein Kern: gemeinsame Fähigkeiten verbinden Cyber, Maschinenbau, Automotive UND Medizin; das Domänenspezifische bleibt am Rand. Architecture Stability bleibt stabil; der Engpass ist jetzt die Qualität der Wissensmodellierung, nicht die Architektur.' % (len(reused), len(fresh)))
+w("")
+
+print("\n".join(OUT))

backend-compliance/tests/test_mcap_convergence_analysis.py

@@ -38,23 +38,36 @@ def test_runs_end_to_end():
 def test_core_is_cross_cutting_not_frequency():
     out = _run()
     core = _section(out, "## 1. Core MCAPs")
-    # the most cross-cutting capability tops the Core report
+    # the most cross-cutting capability tops the Core report; with Medical it now spans 3 domains
-    assert "`secure_signed_update_distribution` | **18** |" in core
+    assert "`secure_signed_update_distribution` | **24** |" in core
     assert "technical_vulnerability_management" in core
     # a high-frequency BUT single-domain management cap must NOT be in Core (frequency != impact)
     assert "conduct_internal_environmental_audits" not in core
 
 
-def test_all_four_reports_present():
+def test_all_five_reports_present():
     out = _run()
-    for header in ["## 1. Core MCAPs", "## 2. Emerging MCAPs", "## 3. Isolated MCAPs", "## 4. Suspicious MCAPs"]:
+    for header in ["## 1. Core MCAPs", "## 2. Emerging MCAPs", "## 3. Isolated MCAPs",
+                   "## 4. Suspicious MCAPs", "## 5. Missing Convergence"]:
         assert header in out
 
 
+def test_missing_convergence_flags_cross_source_duplication():
+    out = _run()
+    mc = _section(out, "## 5. Missing Convergence")
+    # the 'risk' token clusters several distinct risk MCAPs across many sources -> a review candidate
+    assert "Token `risk`" in mc
+    assert "maintain_risk_management_file_iso14971" in mc and "product_cyber_risk_assessment" in mc
+    assert "KEIN Auto-Merge" in out
+    # single-source decompositions (all-environmental) must be filtered out (>=2 sources required)
+    assert "Token `environmental`" not in mc
+
+
 def test_isolated_and_suspicious_are_review_tools():
     out = _run()
     iso = _section(out, "## 3. Isolated MCAPs")
-    assert "issue_battery_passport" in iso or "measure_air_emissions" in iso
+    assert "Review: spezialisiert ODER Konvergenz übersehen" in iso
+    assert "conduct_clinical_evaluation" in iso or "cybersecurity_management_system" in iso
     susp = _section(out, "## 4. Suspicious MCAPs")
     assert "zu grob" in susp and "zu fein" in susp
 

backend-compliance/tests/test_medical_stress_test.py

@@ -0,0 +1,58 @@
+"""Medical stress test — safety + security coupled (Phase Ω #3).
+
+Pins the harder joint test: ISO 13485 runs through the SAME engine (0 runtime, data only), and IEC
+81001-5-1 (health-software security) pulls in the SAME security MCAPs as the CRA — so Medical REUSES
+cyber capabilities (the safety/security coupling shows up as capability reuse) while adding genuinely
+new medical-specific caps (clinical evaluation, software safety classification, ISO 14971 risk file).
+"""
+
+from __future__ import annotations
+
+import os
+import subprocess
+import sys
+
+
+def _run():
+    root = os.path.join(os.path.dirname(__file__), "..")
+    r = subprocess.run(
+        [sys.executable, "reference_scenarios/medical_stress_test.py"],
+        cwd=root, env={**os.environ, "PYTHONPATH": "."}, capture_output=True, text=True,
+    )
+    assert r.returncode == 0, r.stderr
+    return r.stdout
+
+
+def test_runs_end_to_end_zero_runtime():
+    out = _run()
+    assert "Medical Stress Test" in out
+    assert "0 neue Runtime-Klassen" in out
+
+
+def test_safety_security_coupling_reuses_cyber_caps():
+    out = _run()
+    assert "Wiederverwendete Cyber-Capabilities (4)" in out
+    for cap in ["secure_signed_update_distribution", "technical_vulnerability_management",
+                "access_control_and_authentication", "sbom_creation"]:
+        assert cap in out
+    assert "IEC 81001-5-1" in out
+
+
+def test_genuinely_new_medical_capabilities():
+    out = _run()
+    assert "Neu (7)" in out
+    for cap in ["conduct_clinical_evaluation", "classify_software_safety_iec62304",
+                "maintain_risk_management_file_iso14971", "perform_benefit_risk_analysis"]:
+        assert cap in out
+
+
+def test_rejected_assumptions_present():
+    out = _run()
+    assert "ISO 13485 does NOT produce clinical evidence" in out
+    assert "the same security caps as the CRA" in out
+
+
+def test_no_real_company_names():
+    out = _run().lower()
+    for name in ["eto", "owis", "winterhalter", "medtronic", "siemens"]:
+        assert name not in out