Initial validation of the Config+Helpers approach within the MATISSE European project (Nov 2025 – Feb 2026). 24 scenarios · 7 industrial partners · Config+Helpers only.
A 6-stage pipeline was applied to each scenario:
- Partner inputs collection — informal workflow descriptions (draw.io, MS Word, or both).
- LLM-assisted review — systematic issue detection before generation (the 97 issues in Table M1 were found in this stage).
- LLM generation — Config+Helpers solution used for all scenarios.
- Recommendation — corrected BPMN + structured description shared with partners.
- Partner review — accuracy validation, naming alignment, scope decisions.
- Deliverable — final BPMN models consolidated into the project deliverable.
Per-partner review cadence (who performed the stage-2 LLM-assisted
review, who validated at stage 5, typical iteration counts) is
documented in the MATISSE consortium deliverable and is not
disclosed in this public artifact — consistent with the
confidentiality boundary stated in README.md. Aggregate
retention figures and structural metrics are in
partner_metrics.md.
The MATISSE evaluation is qualitative and partner-driven: partners did
not work from a fixed scoring rubric, and "adopted as-is" reflects partner
acceptance rather than an external ground truth. The PMo controlled
benchmark exists precisely to complement this with metric-driven
comparison. See ../PROCEDURE.md for the controlled
benchmark.