Introduction - MATE Docs

Audience#

Role	What MATE gives you
Analyst	A guided path from an uploaded log to charts, models, and dashboards — no code.
Researcher	A runtime for your method: scheduling, progress, caching, isolation, a settings UI, dashboard cards, and citations that travel with the code.
Operator	One compose stack to deploy, back up, and monitor.

How a log becomes a result#

fileXES, CSV, XML, or OCEL 2.0

import jobparse, normalise, write Parquet

Parquet logthe event table plus cached case aggregates

module runsone per (log, module), started when a module is first openedfilters & editscommitting a filter re-runs the modules you have already opened

module resultscached per (log, module)

panels, widgets, and datasetsMATE AIaggregates only, behind the data wall

dashboardscards from any module, on one canvas

One log, from upload to result. An import normalises the file once. A module computes from that cached result the first time you open it, and the interface reads the cache — never the raw upload.

Three properties shape everything else in this manual:

Local-first. Logs, results, and metadata stay on your host. Nothing is sent anywhere unless you configure an external service on purpose.
Modular. The bundled analyses use the same SDK, loader, and gating rules as modules you install later. There are no privileged hooks.
Typed end to end. Pydantic defines the API, one OpenAPI schema generates the web app's types, and module authors program against Protocols.

Reading paths#

Goal	Read
Run MATE	Install and start
See it work once	Your first analysis
Import and explore real logs	Using MATE
Build a module	Your first module
Call MATE from a script or an agent	REST API · MCP server
Run a deployment	Deployment · Troubleshooting
Work on the platform itself	How MATE works · Contributing

Glossary#

Process mining has its own vocabulary; MATE adds a few terms.

Term	Meaning
Event log	A table of events: what happened, when, and in which case.
Case	A process instance (an order, a claim) identified by a case id; its events form a trace.
Activity	The label of what happened — `Registered`, `Approved`.
Variant	The sequence of activities a case followed. Cases that took the same path share one.
Directly-follows graph (DFG)	Activities as nodes, observed successions as edges.
Petri net · process tree · BPMN	Model notations. Discovery produces them; conformance replays a log against one.
Conformance	Fitness (was every trace replayable), precision, and the deviations found.
Concept drift	A change in the process over time: a new branch, a vanishing activity, shifted timing.
OCEL	Object-centric event log: events relate to many objects of different types, so there is no single case notion.
Module	The unit of extension: a folder with a manifest, a handler class, and optionally a frontend.
Manifest	`manifest.yaml` — identity, log requirements, dependencies, capabilities, frontend, citations.
Precompute	The job a module runs for a log to fill its cache. It starts the first time the module is opened (or right after import, for the few eager modules), so every later open is instant.
Capability	A named request/response function one module publishes for another.
Event bus	In-process pub/sub; modules emit and subscribe to declared topics.
Job	A persisted, observable long operation: import, precompute, install, re-import.
Dataset	A typed data output (`table`, `graph`, `kpi`, `tree`, `blob`) the platform renders itself.
Widget	A dashboard card a module ships as a React component.
Guidance	Cached, module-scoped AI explanation, generated on demand.
Data wall	The rule that raw event rows never reach an AI model — MATE AI or MCP.