D1 · Group A Operational
Versioning & Reproducibility
Reproducible builds, prompt/code/model version-pinning, deterministic re-runs, and artifact lineage.
Wie das Modell aufgebaut ist
Das Modell teilt Belange in operative, Wissens-, Risiko- und Ergebnisgruppen, sodass eindimensionale Ansprüche nicht für das Gesamtbild herhalten können. Unten: ein kurzer Überblick und das technische Detail dahinter auf Anfrage.
Reifegradansprüche mit einer einzigen Zahl laden zur Übertreibung ein. Das Modell teilt die Belange in vier Gruppen, fordert eine Mindestabdeckung in jeder und bleibt querverwiesen mit Standards, die Ihre Auditoren bereits kennen. Der Besucher sieht zuerst die einfache Formulierung; das strukturelle Detail wohnt hinter einem einzigen Schalter.
Was folgt, ist die Version in Alltagssprache. Die technische Schicht ist einen Klick entfernt.
Group A
Day-to-day discipline of running an agentic-software-development practice — the engineering hygiene that keeps the hive accountable, repeatable, and recoverable.
L4 group-floor required to claim L4+ overall
D1 · Group A Operational
Reproducible builds, prompt/code/model version-pinning, deterministic re-runs, and artifact lineage.
D2 · Group A Operational
Run-level traces, agent invocation logs, cost / latency / quality metrics, and live drift signals.
D3 · Group A Operational
Automated test gates, prompt regression suites, eval-on-commit, and merge protections.
D4 · Group A Operational
Progressive rollout, blue/green or canary release, rollback playbooks, and deployment audit trails.
Group B
How the hive captures, validates, and propagates knowledge across humans and agents — KB curation, lessons, retrievability, and onboarding semantics.
L3 group-floor required to claim L4+ overall
D5 · Group B Knowledge
KB section discipline, lessons capture, deprecation hygiene, and read-tracking discipline.
D6 · Group B Knowledge
Retrieval quality, grounding traceability, citation discipline, and hallucination-control evidence.
D7 · Group B Knowledge
Stakeholder onboarding, agent-orientation paths, role-based training, and pedagogy evidence.
Group C
How the hive manages security, regulatory, ethical, and operational risk — drift, red-teaming, jurisdictional obligations, and AI disclosure.
L4 group-floor required to claim L4+ overall
D8 · Group C Risk
Secret hygiene, supply-chain security, agent permission scoping, and hardening drift detection.
D9 · Group C Risk
GDPR, EU AI Act, sector-specific regimes, jurisdiction-obligation registry, and DPIA evidence.
D10 · Group C Risk
Active risk register, scheduled red-team exercises, tabletop simulations, and post-mortem feedback.
D11 · Group C Risk
Public AI-disclosure artifacts, decision-explainability, model cards, and use-case transparency.
Group D
Outcomes the practice produces for stakeholders: deliverable acceptance, customer feedback loops, business-level impact.
L3 group-floor required to claim L4+ overall
D12 · Group D Outcome
Stakeholder-acknowledged acceptance of deliverables, signed sign-offs, and rejection-rate tracking.
D13 · Group D Outcome
Customer-impact metrics, business-value tracking, stakeholder feedback loops, and outcome-based KPIs.
Jede Dimension hat ihre eigene kurze Zusammenfassung, mit dem tieferen Kategorie- und Item-Detail hinter einem Disclosure-Schalter auf der Dimensionsseite.