For process mining experts

You know what an alignment is. You do not need the concepts explained; you need to know how this tool is built, what it computes exactly versus approximately, and where it stops. This track is written to be read in about half an hour.

The short version

Promenade is an open workbench where every algorithm — including the built-in ones — is a plugin against one interface. Results are artifacts in a provenance DAG rather than files in a folder. The whole thing runs client-side by default: a streaming parser feeds DuckDB-wasm, artifacts persist as Parquet in the browser’s file system, and plugins execute as WebAssembly, as Python under Pyodide, or as SQL against a relational API.

When a log outgrows a browser tab, the same plugin bundle runs on a container-based engine — locally or on your own infrastructure — with no change to the analysis.

What is exact, and what is not

Alignment-based conformance is exact: the cheapest alignment per distinct trace variant, not a sampled or heuristic approximation. Discovery is whatever the chosen algorithm guarantees — the Inductive Miner returns sound models, the Alpha Miner is included as a teaching baseline and should not be your default on a noisy log.

The track

1. Promenade in fifteen minutes — the architecture, the artifact model, and the vocabulary map.

2. Discovery and the artifact DAG — the miners on offer, and how provenance makes parameter sweeps cheap.

3. Exact alignment conformance — what is computed, at what cost, and how the visual replay relates to it.

4. Object-centric techniques — OCEL 2.0 end to end: OCPN, OC-DFG, OCIM, TOTeM, replay.

5. Scaling past the browser — where the ceiling is, and what changes when you cross it.

6. Packaging your algorithm — turning a technique into an installable plugin.

What Promenade does not do

Worth knowing before you invest time:

• No stream processing. Analysis is over a bounded log, not a live event stream.
• No built-in simulation. Discovered models are not executed forward to predict.
• Prediction is not in the core. There is no bundled predictive monitoring; it is a reasonable plugin, and is not written yet.
• One writer per browser origin. Two tabs on the same workspace would corrupt local storage, so the second tab is held read-only by design.