50 User Requirements Specifications (URS) + 50 paired Functional Specifications (FS)
- 50 paired Design Specifications (DS) for GxP-regulated computer systems — open, free, and built from public regulatory sources.
A public, openly-licensed reference corpus covering the regulated pharma / biotech / medical-device system landscape — from compendial instruments to AI/ML platforms, enterprise quality systems, clinical-trial systems, post-market and safety systems, and shared IT infrastructure. Three paired specification tiers (URS → FS → DS) for the same 50 systems, anchored in primary regulatory sources (FDA, EMA, ICH, EU GMP Annex 11, EU AI Act, DACH authorities).
Hub page: https://neuralarchitects.ae/gxp-corpus
Every file in this corpus carries do_not_use_as: regulated_record and
labelling.source_risk: ai_authored_disclosed in its YAML frontmatter. Companies, dates, and
identifiers are marked (synthetic). No FDA, EMA, BfArM, Swissmedic, or other competent
authority has reviewed or signed off on this material.
If you derive an operational artefact from these templates, your Quality Unit must review, edit, sign, and approve it under your own change-control before it has any regulated status. Use this corpus as a starting point for house templates, RAG grounding, fine-tuning, or training contrast sets — never as a drop-in deliverable.
gamp5-corpus/
├── README.md (this file)
├── LICENSE Creative Commons Attribution-ShareAlike 4.0
├── CITATION.cff Citation metadata (v1.4)
├── CHANGELOG.md v1.0 → v1.1 → v1.1.1 → v1.2 → v1.3 → v1.4
├── KNOWN_LIMITATIONS.md Honest list of what the corpus does well / less well
├── METHODOLOGY.md Canonical rule-set for URS + FS (§ 2A, the law)
├── EVAL_RUBRIC.md Grading rubric for derived URS + FS
├── URS/ 50× *_URS_v1.3.md
├── FS/ 50× *_FS_v1.3.md
└── DS/
├── METHODOLOGY.md Canonical rule-set for DS authoring
├── EVAL_RUBRIC.md Grading rubric for derived DS
├── CHANGELOG.md DS-specific version history (v1.0 → v1.1)
└── 50× *_DS_v1.1.md
DS ships with its own methodology / rubric / changelog inside the
DS/folder because the DS authoring rules (Cat 4 Configuration Specification vs. Cat 5 Software Design Specification vs. Cat 1 Infrastructure Design Specification vs. Cat 3 vendor-design-reliance statement) are distinct enough from the URS / FS rules that they warranted a separate canonical document rather than a new section in the top-level METHODOLOGY.
Depth is sized to real-world system complexity.
| Tier | Class | Target reqs | Examples |
|---|---|---|---|
| T1 | Small / standalone instruments + utilities | 30–50 | Karl Fischer, particle counter, compendial calculator |
| T2 | Configured lab instruments + mid-complexity apps | 50–80 | UV-Vis, NIR, TOC, dissolution, SPC, stability, APR, LMS, autoclave, cold chain, EMS, BMS |
| T3 | Large enterprise + production systems | 100–150 | LIMS, ELN, EDMS, eQMS, RIM, RTRT, SCADA, AD, audit-trail workbench, CV inspection |
| T4 | Mission-critical / multi-module | 150–250 | EDC, CTMS, eTMF, Vault Submissions, AI/ML Model Server, GenAI LLM Service, MES PAS-X, PV DB |
- Lab instrumentation & analytics — Chromatography, LC-MS, UV-Vis, NIR, TOC, dissolution, stability, SPC, APR, particle counter, Karl Fischer
- Production & process control — MES PAS-X, SCADA water systems, bioreactor / continuous fermentation, tablet coater, autoclave, lyophilizer, cold chain
- Enterprise quality — LIMS, ELN, EDMS, eQMS, ValGenesis VLM, Watson LIMS, RIM, RTRT, Annual Product Review
- Clinical trials — EDC, CTMS, eTMF, Vault Submissions, ePRO portal, LMS, randomisation, study build
- Post-market & safety — Pharmacovigilance DB (Argus), EU MDR PMS, Adverse Event DB, Submissions drafting
- AI / ML / Custom Cat-5 — AI/ML Model Server, GenAI LLM Service, CV Inspection (sterile products), audit-trail workbench
- Shared IT infrastructure — Active Directory identity service, Backup tier (T1/T2/T3, RPO/RTO targets)
URS — single canonical structure (METHODOLOGY § 2A.12):
- § 1 Purpose · § 2 Scope · § 3 System description + GAMP category · § 4 User roles · § 5 Requirements (per-domain subsections, including 21 CFR Part 11, ALCOA+, EU AI Act where relevant) · § 6 Acceptance criteria · § 7 Constraints · § 8 Assumptions · § 9 References · END marker.
- Each requirement carries ID + Priority (H/M/L) + Risk class (R1/R2/R3) + verifiable
shall-clause. - From v1.3 every URS declares its Project Mode in the Document Control block (config-on-COTS, custom-build, infrastructure-qualification, or instrument-configuration).
- The URS is FS-namespace-independent — no Appendix A, no embedded FS-IDs.
FS — paired one-to-one with the URS:
- § 1 Purpose + seed_corpus_basis · § 2 Architecture · § 3 Component decomposition · § 4 Functional specifications (per-FS-ID rows citing originating URS-IDs) · § 5 Interface specs · § 6 Data + persistence · § 7 Non-functional behaviour · § 8 URS → FS Traceability Matrix (every URS-ID gets its own row, no range compression) · § 9 Implementation Risk Register (new in v1.3) · Revision history.
DS — paired one-to-one with the URS and FS (new in v1.4):
- Shape is GAMP-category-driven (see
DS/METHODOLOGY.mdfor the full canonical rule-set):- Cat 5 (custom-build) → full Software Design Specification: module-level decomposition, interface contracts, data models, sequence diagrams, deployment topology, custom-development scope.
- Cat 4 (config-on-COTS) → Configuration Specification: configuration-vs-code split, vendor-feature inventory, environment-scoped configuration items, configuration migration.
- Cat 3 (instrument) → vendor-design-reliance statement + configuration locks.
- Cat 1 (infrastructure) → Infrastructure Design Specification.
- Every DS carries a FS → DS traceability matrix so the URS → FS → DS chain is end-to-end auditable for Cat 4 / Cat 5 systems.
Every file is browsable in the repo tree.
git clone https://github.com/neuralarchitects-de/gamp5-corpus.git- Latest release: https://github.com/neuralarchitects-de/gamp5-corpus/releases/latest
- Mirror on the hub page: https://neuralarchitects.ae/corpus/gxp-corpus-v1.4.zip
The corpus is structured for supervised fine-tuning of regulated-industry specification authoring. Pair each URS with its FS and DS using filename matching (system name prefix), or use the explicit traceability matrices in FS § 8 (URS → FS) and the DS FS → DS table. The METHODOLOGY files ship the authoring rules.
Short form:
Attia, N. GAMP 5 Synthetic URS + FS + DS Corpus, v1.4 (2026). NA IT Consulting (Neural Architects), Sandhausen, Germany. Licensed CC-BY-SA 4.0.
Full BibTeX / RIS / EndNote metadata in CITATION.cff. Right-click → "Cite this
repository" on GitHub also surfaces a one-click citation.
Licensed under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0).
You can use, modify, and redistribute the templates — including for commercial purposes —
provided you attribute the corpus (see CITATION.cff) and license your
derived work under the same terms. ShareAlike is intentional: it keeps the regulatory
ground-truth open as it spreads.
Full license text in LICENSE.
The URS + FS canonical rule-set is METHODOLOGY.md. The DS canonical
rule-set is DS/METHODOLOGY.md. Highlights from the top-level § 2A:
- § 2A.1 Citation currency table — FDA / EU / ICH / DACH / vendor primary sources, single source of truth.
- § 2A.2–2A.5 Sub-section maps for 21 CFR Part 11, ICH M10, EU MDR PSUR cadence, 21 CFR Part 58 GLP roles — locks out common citation defects.
- § 2A.6 DACH authority map (BfArM ≠ GLP authority; BfR + Länder = DE GLP).
- § 2A.7 URS↔FS traceability lives in FS § 4 + § 8 only.
- § 2A.8
shallfor system requirements;mustonly for downstream-consumer constraints or quoted regulator text. - § 2A.9 YAML frontmatter rules + disclosure invariants.
- § 2A.11 EU AI Act provider obligations Arts. 8–21.
- § 2A.12 URS / FS structure invariants (§ 1 → § 9 References → END; FS ends with Implementation Risk Register).
- § 2A.13 Tier system (T1–T4).
- § 2A.14 EU AI Act Annex I (SaMD / safety-component, 2 Aug 2027) vs Annex III (eight listed domains, 2 Aug 2026).
- § 2A.15 Project Mode declaration rule — new in v1.3.
- § 2A.16 FS § 9 Implementation Risk Register rule — new in v1.3.
See CHANGELOG.md for full version history.
- Author: Nabil Attia (NA IT Consulting / Neural Architects), Sandhausen, Germany.
- Authoring approach: AI-augmented with multi-wave human-directed enrichment and external
citation review. Each authoring agent works under METHODOLOGY § 2A (URS / FS) or
DS/METHODOLOGY.md(DS) and produces self-verified output. All AI authorship is disclosed in the YAML frontmatter. - No client data, no NDA material, no regulated records were used. Sources are public regulator documents, ISPE GAMP 5 (2nd ed., 2022), ICH guidance, EU AI Act 2024/1689, and vendor public documentation.
- Defects we have caught (and how we keep them out): see
KNOWN_LIMITATIONS.mdand the CHANGELOG patch history under v1.1.1 and v1.3.
The corpus is structurally stable but not finished — citation accuracy, coverage breadth, and authoring-style diversity all benefit from outside eyes.
- Found a citation defect? Open an issue with the file path, the quoted line, and the authoritative source. Cite-error reports are the highest-value contribution.
- Coverage gap? Open a discussion. We track planned additions in the CHANGELOG TODO list.
- Structural issue (broken traceability, missing END marker, YAML defect, etc.)? Open an issue with the offending file.
- Want to submit a new URS / FS / DS triple? Open a draft PR. New triples must follow
METHODOLOGY § 2A end-to-end for URS + FS and
DS/METHODOLOGY.mdfor the matching DS. The METHODOLOGY files are the contract.
Please read KNOWN_LIMITATIONS.md before opening an issue — many obvious
gaps are already tracked.
- Website — https://neuralarchitects.ae/gxp-corpus (download mirror + 1-page overview)
- Hugging Face dataset — coming soon
- Contact — contact@neuralarchitects.de
- Author profile — https://www.linkedin.com/in/nabil-attia/
Generated and maintained by Neural Architects — NA IT Consulting, Sandhausen, Germany.