Skip to content

decisions, derivation rules, richer FCO-IM, a smarter chat, and more

CaseTalk 15 gave everyone a free Community Edition and opened models to AI assistants. Version 15.2 builds on that.

We have resolved 234 tickets since 15.1, and 128 of them are new features. Here is what is in it.

Business decisions, modeled as facts (DMN)

Derivable fact types and business rules get a decision table.

  • You start with examples. Each row is a real scenario drawn from the population, marked as allowed or forbidden.
  • Then you write the rules. You generalise the examples into rule rows, with comparisons between columns and formulas such as = Base Rate * 1.21.
  • Your examples become the test set. Every rule is checked against them.

The decision table is not a separate store bolted onto the model. It is another view on the same fact types, so the rules verbalise in plain sentences like everything else. When you need DMN, CaseTalk exports it: DMN 1.3 with the decision diagram, or DMN 1.5 with your examples as test cases, and formulas in FEEL.

Derivation rules you can build, read and run

  • Write the rule where the fact type lives. A derivable fact type gets a Decision Table tab in its own editor, with filters and functions on the path endpoints.
  • Read it back in natural language, and choose what a derived fact type reads from its own editor.
  • Run it. CaseTalk generates the derived population and reports the violations.
  • Rules about change. A dynamic rule compares the old and the new value of a fact in one transaction: "an employee's salary may not decrease".
  • Rules travel with the model through merges, the server repository and the JSON export, and AI assistants can read and write them.

More of FCO-IM, implemented

  • Split and overlapping expression parts. One sentence can reuse a value it has already stated, or spread an object type over several places: "In Station track TrackNr lies next to platform PlatformNr". The grouped readings are stored, drawn and populated.
  • Set expressions. A role can hold a set, a bag or a sequence. Enter example data as {a, b} or (a, b), and rules can aggregate over it.
  • Expression unification. Switch an object type expression to a fact type expression and back, and read expressions at a mix of type and label level.
  • Decompose a fact type into elementary fact types, and compose them back into one sentence.
  • Subtype indications. The constraint assistant asks for subtypes, and the transformation adds a discriminator for the subtypes it groups into one table.
  • Richer constraints. A constraint can be left out of a named scope or flagged instead of enforced, and it can be obligatory rather than necessary. A subset constraint can state that role combinations exclude each other, and every constraint dialog verbalises the rule while you edit it.
  • Templates pass their settings (Annotatable, Temporal, Localize, ...) on to all their members.

NORMA import and export, round trip

CaseTalk now imports the ORM constraints that NORMA joins through the model: uniqueness, ring, subset and equality constraints whose roles span several fact types, and derivation rules that used to come in only partly. The ORM export writes these joined constraints back to NORMA.

A chat that works like an assistant, not a chatbot

The AI chat in the Modeler has a new engine.

  • It plans before it answers. It works through a to-do list that shows up in your project.
  • It reads the open model directly.
  • It streams its answers as they are written, and you can stop it halfway.
  • It keeps the conversation when you switch projects.
  • You can make it your own. Write a standing opening prompt (general or per LLM), choose a coaching persona, and test each AI service with one click.
  • It works with the AI service you already use: Claude, ChatGPT, Azure OpenAI, Gemini, Mistral, Groq, OpenRouter, GitHub Models, your GitHub Copilot subscription, or a local model through Ollama.
  • It can query every model on the repository server that your account may read.
  • Ask the coach from any dialog. The help button explains the dialog, asks the coach or sends feedback.
  • The chat can carry out the work: it runs the model transformation and generates CaseTalk's outputs for you.
  • External AI agents can connect in every commercial edition. They are asked for access once, when they connect.

Sketch terms in TermBoard, right inside your project

Draw terms and their relations on termboard.com without leaving CaseTalk. The board is saved in the project, layout included.

From law text to model

  • Annotate parts of a law paragraph and turn them into expressions.
  • Keep the official JuriConnect references on what you import, and export them again.
  • Re-import a changed law. The paragraphs update, removed articles are retired, and your annotations stay in step.
  • Organise and classify paragraphs. Reorder them by drag and drop, classify them as ArchiMate motivation elements, and configure the Wetsanalyse schemas (JAS, JRM, ...) through custom attributes.
  • A glossary of the words in the law texts, with their senses.
  • Legal sources in the MIM 2.0 export, linked to the concepts.
  • Import laws or policies from more sources. Besides Dutch laws from wetten.overheid.nl, CaseTalk now reads German federal laws from gesetze-im-internet.de, Dutch regulations and policies in STOP (the Omgevingswet publications), and Akoma Ntoso, the international standard used by the EU, Switzerland, Italy and the UK.

One law sentence with each part annotated as source definition, legal object, parameter value, operator, condition and legal fact, each linked to a concept

Every part of a law sentence, annotated and linked to the model.

Annotated paragraphs of a Dutch tax law next to the concept editor for the term aansluiting

Each paragraph keeps its annotations, and each term becomes a concept with its aliases and Wetsanalyse (JAS) classification.

An annotated article in the MIM report, with its JuriConnect reference and a link to wetten.overheid.nl

In the report: the annotated article, its JuriConnect reference and the official source.

Diagrams and reports

  • Lines route around boxes instead of running through them.
  • Diagrams keep their identity when you rename or move them, so references to them keep working.
  • Reference data looks the part. A static population is drawn as a closed box.
  • Diagram verbalizations get their own expression profile and look like expression files. ArchiMate colours are available in the diagram styles.
  • Zoom into pictures, and find generated pictures and reports in a doc folder next to the project.
  • The project report shows lineage: where every logical table and column comes from. Generated artifacts come with their own documentation, and you choose which custom attributes the report shows.

Import, integration and data

  • Reverse engineering reads more of the database: primary, unique and foreign keys, with per-DBMS catalog queries where a driver falls short, and column types mapped to domains.
  • Data mapping joins rows from different source tables, with filters and transforms.
  • Conflicting facts from different source systems are shown side by side.
  • Build information models from large OWL ontologies such as FIBO, and export the conceptual model to Common Logic (ISO 24707).
  • Import Apache OSSIE semantic models, and import a model from its own JSON export.
  • Describe deliveries for data catalogues (DCAT), with the mapped source systems and their meaning.
  • The Data Vault transformation now builds versioning, hash keys and load metadata.

And further

  • A smarter totality constraint wizard. It decides from your population and asks its questions in sentences. It also proposes XOR (exclusive-or) constraints.
  • FreeStyle workshops start from the model you have open and merge back into it.
  • A new Power BI generator writes TMDL scripts.
  • A richer Turtle import, with lineage in the RDF export. The import now brings in role readings, inverse pairs and provenance, and the RDF export records where each model came from (PROV-O). Imported ontologies can be drawn in VOWL notation.
  • Git versioning for projects, as an alternative to archives. Upload messages, commit comments and archive comments are prefilled from what was done and why.
  • Show and switch what you are working on from the toolbar. Sessions survive several CaseTalk windows.
  • The population editor edits all facts about an instance together.
  • Bring all linked models up to date with the server in one go, and clean up unused model versions in the repository.
  • Coach messages can arrive through the Windows notification center.
  • Screen-reader support for CaseTalk's own controls.
  • 22 crash fixes, plus a redesign of model locking that ends the freezes some of you saw when an AI assistant worked on a model.