This guide covers the menu, sharing and loading, the standard user journey, editing a crime script, and the LLM wizard. The images use public starter content and synthetic test data.
PAX stores the workspace locally in the browser. On first use, choose the English starter library, an empty workspace, or your own JSON model. Open the menu in the top-left corner to configure the workspace and view.
First select the role that matches your task:
- User: view, search, and export scripts.
- Editor: also edit and delete existing scripts.
- Administrator: additionally create scripts, use the LLM wizard, and edit shared taxonomy such as roles, attributes, and locations.
You can only edit a script after selecting Editor or Administrator. Only an Administrator can add or change roles and other shared taxonomy. The role selector is prototype behavior in the browser, not authentication or access control.
Use Language to select the interface language and the language of a starter library imported later. Existing scripts are not translated automatically.
Use Script mode to switch between:
- Public mode: shows and exports public scripts only.
- Restricted mode: shows a restricted counterpart where one is available. Exporting requires additional confirmation; a permanent link is unavailable while the model contains restricted content.
Script mode is not access control either. Share restricted information only through an appropriately secured channel.
The menu contains actions for the complete workspace:
- Download model as JSON creates an editable export of the collection. In public mode it contains public scripts only.
- Upload model as JSON makes the selected file the current local workspace. Export the existing workspace first if it must be retained.
- Create permanent link copies a URL containing the public model. Its recipient opens a copy in their own browser. Permanent links are disabled for models containing restricted content.
For one script, open More actions on that script:
- Export to JSON creates an editable file containing the script and its required taxonomy.
- Export to Word creates a reading report; it is not an editable PAX model.
- Share this script as a link copies a ZIP-compressed, URL-safe link containing the JSON for this one public script and its required taxonomy. Recipients can review it as a preview and select Merge current script to add it to their workspace without replacing their collection. This action is unavailable for restricted scripts. Large scripts may produce links too long for some email clients.
The recipient selects Upload model as JSON to open a collection or single-script file. This replaces the current local workspace; it does not merge arbitrary files automatically. A shared script link instead opens a preview. Before sharing, always check the classification and filename, and use a secure channel for restricted files.
- Search for or filter a script on the home page.
- Open it by selecting the card or the more action.
- Select a track when the script contains multiple routes.
- Select a modus operandi in a scene to view an alternative process route.
- Use the role pills below activities to highlight activities for one role. Use the link icon beside a taxonomy item to find where that role, attribute, or location is used elsewhere.
- Open a pill under Related crime scripts to navigate directly to a linked process.
- Expand roles, attributes, transports, locations, and sources when those details are needed.
The compact scene navigation shows the number of activities in each scene and, where alternatives exist, the number of modi operandi. The summary above the scenes shows counts rather than every taxonomy item at once.
Select Edit script. The editor contains a scene overview and details for the selected scene.
- Open Script details to change the title, description, language, classification, products, geography, and references.
- Add scenes or change their order in the overview.
- Edit each scene's modus operandi, description, and location.
- Add activities and link roles, attributes, transports, and, where needed, one or more related crime scripts. Only scripts available in the active public or restricted mode can be selected.
- Add indicators, measures, and other available taxonomy.
- Use the track editor to select the required modus operandi for each scene.
- Check the result in the viewer. PAX stores changes locally.
Use + M.O. beside a scene to add an alternative, or the pencil beside the track to rename it. More actions offers JSON and Word exports, barrier models as SVG or PNG, and detaching a starter script. Before substantial changes, save a JSON export as a recovery point.
Open Case file and enter concrete observations, one per line or separated by commas. PAX compares them locally with the available crime scripts. Selected products, locations, roles, attributes, and transports are treated as requirements; leave them empty when they are uncertain.
Each hypothesis shows:
- how many observations match fully or partially;
- which terms were and were not found;
- which scenes contain the matches;
- which entered observations remain unexplained.
A match is a starting point for analysis, not a conclusion about what happened. Where possible, compare two or three hypotheses and record what additional information could distinguish them.
Open Learning mode and choose a public reference script. If you have not selected your own content, PAX uses the starter library. Exercises ask you to complete or order scenes and activities and to select suitable roles, indicators, or barriers. Options may also come from other scripts.
After selecting Compare with reference, PAX shows similarities and differences. A different answer is not treated as automatically wrong: other choices may be defensible and should prompt reflection. For unreviewed starter content, heed the warning that the reference is learning material rather than established truth.
Select Generate with LLM on the home page. PAX uses a manual, provider-neutral hand-off: the application does not contact an LLM, store an API key, open a source URL, or transmit workspace data.
Enter:
- the script language;
- crime type or domain;
- geographical applicability;
- level of detail;
- optional source URLs, one per line;
- optional pasted source text or notes.
URLs are included verbatim in the prompt, but PAX never opens or downloads them. Do not include secret or identifiable information in a prompt sent to an external service.
Select Build prompt and check the generated prompt. Copy it to an LLM of your choice. The prompt requests defensive analysis and excludes actionable offending instructions and evasion tactics.
Copy only the LLM's JSON response back into PAX. The wizard parses and validates it locally. An error identifies the field or missing reference; correct the response outside PAX and try again.
The image uses reproducible synthetic JSON. The selected LLM does not matter for this manual, provider-neutral hand-off.
Before importing, check:
- title, language, and classification;
- scenes, modi operandi, and activities;
- roles, locations, indicators, and measures;
- sources and how they were used;
- the absence of operational offending instructions.
The workspace changes only after explicit import confirmation. An imported script remains AI-generated and Unreviewed until a person has completed a substantive review.
- Export the complete workspace as JSON for a local recovery point before loading another model.
- Use a script JSON file for focused review and exchange, and a collection JSON file to transfer a complete workspace.
- Treat a permanent link as sensitive information: the public model is encoded in its URL.
- Always check the filename and classification before sharing.
- Clear local application data only after verifying a usable recovery file.









