Path Schema Guide

Path schemas allow finding out how two entities (events or objects) are connected in an OCEL, even if they share no direct relationship and are only linked through chains of other entities. For example, you might be interested in how long it takes from place order to package delivered, and through which entities the two events are connected.

A path schema is a path on the type level, such as Order -[places]-> Item -[part_of]-> Delivery. OCPQ enumerates all schemas between a selected source and target type, and ranks them using metrics such as coverage and selectivity. As such, meaningful connections can be distinguished from noise. Each schema can then also be opened as a regular OCPQ query for further analysis.

The approach is described in the paper Connecting the Dots: Path-Based Entity Linking in Object-Centric Event Data (BPM Forum 2026, PDF).

After loading an OCEL, the path schema view can be opened using Paths in the sidebar.

Path Schemas view, ranking schemas from place order to package delivered

Selecting Source and Target Types

The toolbar on top contains a Source and a Target picker, which list all event and object types together with their number of instances. Alternatively, you can click on nodes in the type graph below, where the first click sets the source and the second click sets the target. In the screenshot above, the source is the place order event type and the target is the package delivered event type.

As soon as both are set, OCPQ enumerates all path schemas between them (up to the selected length). This step is fast and runs automatically, but does not yet compute any metrics.

Enumeration Settings (Optional)

Max k is the maximum path length, i.e., the number of steps a schema can have.
The scope (the type filter on the graph) limits which intermediate types a path can pass through, while the source and target are always allowed. By default, the scope is chosen automatically (all types for small logs, and the connected top types for large ones), but it can be widened or narrowed using the controls. A smaller scope results in fewer, more relevant schemas and faster computation.

Computing Metrics

Next, click on Compute metrics, which traverses the actual instances and fills the table. Three settings determine how connections are counted:

Temporal can be None (time is ignored), Forward (the target event has to occur after the source), or Bounded (the target has to occur within a configurable time window).
Selection determines which target events are kept if a source reaches several of them, i.e., All, First, Last, or Closest (closest in time).
Pruning (advanced) sets a selectivity threshold. Schemas that are guaranteed to stay below it are dropped early and marked as pruned. Setting the threshold to 0 disables pruning.

Reading the Results

Each row corresponds to one schema, with the following (sortable) metrics:

Support is the number of distinct (source, target) pairs that are connected.
Coverage is the fraction of source instances that reach at least one target.
Selectivity is the inverse average fan-out (1 / average number of targets per connected source). A high selectivity indicates that each source is mapped to few targets, i.e., that the schema is discriminating.
Reach is the fraction of target instances that are reached.
Exclusivity is the inverse average fan-in. A high exclusivity indicates that each target is reached by few sources.
Throughput is the mean time between source and target (for event-to-event schemas).
Eq is the connection-equivalence class. Schemas producing exactly the same set of connections share a class.

Schemas without any connections are marked as dead, and can be hidden using Hide dead schemas.

Inspecting and Reusing a Schema

Clicking on a row opens the schema in the panel on the right:

Radar compares the selected schemas across all metrics at once.
Details shows the path diagram and base metrics, and allows recomputing them with different Temporal and Selection settings. Additionally, it plots the throughput histogram as well as the distribution of connections per source and per target. Download connections exports the connections as a CSV file, and Open as OCPQ query turns the schema into an editable query in the query editor.