Building Conformance-Aware Workflows in Enterprise Systems

Conformance checking is often treated as a post-hoc audit function — you run it on historical logs to understand what happened. Meridian takes a different view: conformance should be part of the operational loop, not a retrospective report.

How conformance checking works

At its core, conformance checking compares an observed process execution (event log) against a reference model (process model). The degree of fit is expressed as a conformance score between 0 and 1. Every case flowing through your system produces a trace — a timestamped sequence of events. The conformance engine aligns that trace against the reference model and counts deviations.

Alignment computation

The alignment algorithm finds the minimum-cost mapping between the observed trace and the reference model. Each move is classified as synchronous (both model and trace agree), log-only (the trace did something the model didn't expect), or model-only (the model expected something that didn't happen). Deviation cost is the count of non-synchronous moves.

// Conformance score computation (simplified)
function computeConformance(trace: Event[], model: ProcessModel): number {
  const alignment = computeAlignment(trace, model)
  const deviationCost = alignment.moves.filter(
    m => m.type !== 'synchronous'
  ).length
  const maxCost = trace.length + model.transitions.length
  return 1 - deviationCost / maxCost
}

The operational loop

Traditional conformance tooling operates offline — you export event logs, run analysis, get a report. The insight we acted on is that this delay is the problem. By the time you see the deviation, the case is closed and the damage is done. Real-time conformance changes the intervention window.

We used to review conformance reports every quarter. Now we catch deviations in the same hour they happen. The difference in outcome quality is not incremental — it's categorical.

— Head of Operations, Fortune 500 logistics firm

Designing conformance-aware workflows

The key insight is to treat your reference model as a first-class operational artifact — version-controlled, reviewed, and maintained alongside your procedures and automation rules.

• Version your process models alongside your code in source control
• Set conformance thresholds per process, not globally — invoice processing tolerates less deviation than employee onboarding
• Use variant analysis to distinguish structural deviations (wrong path) from timing deviations (right path, wrong speed)
• Instrument your automation to emit events that feed back into the conformance engine for closed-loop measurement

---

Further reading

• Process Mining: Data Science in Action — van der Aalst (2016)
• Conformance Checking: Relating Processes and Models — Carmona et al. (2018)
• Meridian Docs: Conformance Engine Reference