Research Project: GitCover.IdP
Official Recognition
The research project "Git-native Compliance for SMEs: Experimental development of an integrated identity and compliance management system with taxonomic identifier system (V7GUID)" was recognized by DLR Projektträger / BSFZ as an eligible R&D project under § 6 FZulG (German Research Tax Allowance Act).
| Attribute | Value |
|---|---|
| Case number | 288-335-338/2026-1 |
| Decision (positive) | 22.04.2026 |
| Research type | Self-operated experimental development |
| Duration | 01.01.2025 - 31.12.2026 |
| Certification body | BSFZ at DLR Projektträger, Bonn |
Research Content
Investigation of whether Gitea-based Git repositories can serve as a Single Source of Truth (SSOT) for an integrated identity and compliance management system for SMEs.
Core Hypotheses
| ID | Hypothesis | Research Risk |
|---|---|---|
| H1 | Git organizations can represent OAuth2 clients and permissions and manage >1,000 users | No reference implementation for >100k objects known |
| H2 | End-to-end policy compliance chain: OPA-Rego -> OSCAL Assessment Results -> Evidence | Hierarchical permissions on flat Rego policies are complex |
| H3 | Deterministic Git structures as containment vessel for autonomous AI agents | No reliable standards for deterministic AI containment exist |
Novelty Characteristics
| ID | Characteristic | Reference |
|---|---|---|
| N1 | V7GUID Dual-Identifier: Integration of taxonomy into UUIDv7 | Patent application 10 2025 003 091.6 |
| N2 | AI Guard-Rails: Cryptographic Git chains to tame autonomous agents | GCUCB; VPRM concept |
| N3 | GCEP Protocol: Advisory Locks and bundle signing for multi-repo sync | Patent application 10 2025 003 359.1 |
BSFZ Research Questions F1-F5 (Frascati-qualified)
| # | Research Question | Uncertainty |
|---|---|---|
| F1 | V7GUID lookup O(1) with >10^6 objects in shared memory? | Hash collisions, NUMA effects |
| F2 | NativeAOT code from OSCAL/OPA YAML binary-compatible with memory layout? | Schema evolution |
| F3 | Typed bus improves LLM result quality vs. JSON text? | No benchmark exists |
| F4 | Cross-container shared memory compatible with GoBD? | Mutability vs. audit |
| F5 | Energy consumption per compliance workflow measurably reducible? | Metrics missing |
Relevance for GCBoK
The BSFZ-recognized research project demonstrates the scientific-technical novelty of the GitCover core architecture from an official perspective and strengthens the normative authority of GCBoK:
- V7GUID as a patented identification system (N1)
- GCEP/GCAL as a lock coordination mechanism (N3)
- AI Guard-Rails via deterministic Git structures (N2)
This threefold confirmation (patent application + BSFZ certificate + utility model) is the empirical foundation for GCBoK's claim to positioning as a normative terminology authority.
Phase 2 — AI Safety Research (from 2026)
From 2026, the focus expands to the security of AI infrastructures. The research thesis: Deterministic Git structures (V7GUID) can serve as physical guard rails for autonomous AI agents — instead of stochastic, injectable prompts.
Delineation from the Previous Project
| Dimension | Previous Project (2025–2026) | New Project (from 2026/2027) |
|---|---|---|
| Subject | Identity management, IdP, PGP/V7GUID, GoBD/GDPR | AI governance, guardrails instead of prompts, MCP agent |
| Technical uncertainty | Context manipulation of identity | Determinism vs. stochastic AI control; prompt injection resistance via Git gate |
| Building block in GCBoK | Identity & compliance core | Tooling & guardrail layer (GCUCB) |
| Reference framework | ISO/IEC 24773 | OWASP LLM Top 10, Harness Engineering, MCP |
The content shift from the topic "Gitea as IdP" to "Git repos as AI guardrails instead of prompts" via GCUCB requires a separate BSFZ certification as a new research project. The delineation at the work package level ensures that both projects remain cleanly separated and the research tax allowance recognition is not jeopardized.
Research Questions (Phase 2)
- Determinism vs. Stochastics: To what extent can the control of an AI agent be deterministically captured through a versioned Git guardrail repo, so that deviations from policy are detected at build time (not just at prompt time)?
- Prompt Injection Resistance: How far does a Git-native guardrail layer (signed, immutable, enforced via Git hook/OPA) protect against the "lethal trifecta" (private data + untrusted content + exfiltration)?
- MCP as Policy Enforcement Point: Can an MCP server with UI resource and iframe UI act as a policy gate that only exposes approved (signed) tools/contexts?
- Auditability: Is the Git commit history of the guardrail repo usable as a GoBD-compliant audit trail for AI decisions?
Note: The research and prototype developments by AFJD (Axel Franz Johann Druschel) are conducted on a self-operated basis outside of GCC working hours. The IP remains with AFJD; GCC is granted usage rights on a contractual basis. This separation secures the non-profit status of GCC (no concealed commercial business operation).
See also: FZulG applications and financial planning on the GCC landing page.