Skip to content

Latest commit

 

History

History
61 lines (49 loc) · 3.68 KB

File metadata and controls

61 lines (49 loc) · 3.68 KB

MATISSE — partner metrics (paper Tables M1–M4)

Aggregated metrics over the 24 industrial scenarios. The raw per-partner workflows and generated BPMN models are confidential — see README.md.

Table M1 — Detected issues in partner-provided workflows

97 issues found across 12 scenarios with quantified reviews (avg 8.1 per scenario).

Category Count % Examples
Structural 34 35% Missing actor lanes, unclear start/end events
Data flow 24 25% Undefined data objects, missing associations
Process logic 22 23% Missing decision gateways, absent feedback
Traceability 17 17% Requirement ID mismatches, coverage gaps
Total 97 avg 8.1 per scenario

Table M2 — Partner retention of generated BPMN models

22 LLM-generated corrected BPMN models compared against final Modelio models.

Category Count % Description
Adopted as-is 14 64% Model accepted in Modelio without modification
Minor changes 3 14% Task or lane name adjustments only
Major refactor 5 22% Partners change structure or naming conventions

Table M3 — Structural metrics by partner (final BPMN models)

Partner Scenarios Avg Lanes Avg Tasks Avg Gateways Avg Data Obj. Avg Data Assoc.
P1 2 3.0 13.0 6.0 10.0 17.5
P2 3 4.3 11.0 1.3 4.0 8.7
P3 4 5.0 15.8 3.3 9.0 15.5
P4 4 4.5 11.8 4.0 6.5 11.0
P5 5 5.0 15.8 3.6 6.2 12.4
P6 4 3.75 17.3 3.5 11.8 23.8
P7 2 4.0 14.0 2.5 13.0 26.0
All 24 4.3 14.5 3.3 8.2 14.6

Table M4 — MATISSE vs. PMo dataset comparison

Metric MATISSE (24) PMo (55) Ratio
Avg lanes 4.3 1.1 3.9×
Avg tasks/elements 14.5 22.6 0.6×
Avg gateways 3.3 8.3 0.4×
Avg data objects 8.2 0
Scenarios with data objects 23/24 (96%) 0/55 (0%)
Avg data associations 14.6 0
Avg lines in output 232 (range 116–549) 137 (range 53–328) 1.7×

MATISSE scenarios represent collaborative multi-actor workflows (4.3 lanes avg), while PMo scenarios are predominantly single-lane (1.1 lanes avg) with more complex decision logic (8.3 gateways avg). MATISSE outputs are also significantly richer in data objects and associations, highlighting dimensions that controlled benchmarks may underestimate.