Skip to content

PACK-VIEWS.md

Latest commit

 

History

History
153 lines (114 loc) · 3.41 KB

PACK-VIEWS.md

File metadata and controls

153 lines (114 loc) · 3.41 KB
Raw

Pack Manifests and Canvas Views

Causari packs contain a canonical causal graph and may optionally contain presentation metadata.

The separation is deliberate:

canonical graph data
+ audience/view projection
= timeline, canvas, lesson, recap, or decision workflow

Canonical files

Every pack continues to use:

events.json
links.json
insights.json

These files must remain portable across the MCP server, Canvas, static imports, runtime fetches, and future consumers.

Optional files

A pack may add:

manifest.json
views.json

manifest.json

The manifest describes the pack rather than changing graph semantics.

Recommended fields:

{
  "id": "pack-id",
  "version": "0.1.0",
  "title": "Pack title",
  "description": "Purpose and scope",
  "language": "en",
  "defaultLocale": "vi",
  "jurisdiction": "VN",
  "temporalMode": "effective-date-timeline",
  "graphMode": "instrument-to-control-to-evidence",
  "updateCadence": "event-driven",
  "asOf": "2026-07-30",
  "files": {
    "events": "events.json",
    "links": "links.json",
    "insights": "insights.json",
    "views": "views.json"
  },
  "canvas": {
    "defaultView": "executive-impact",
    "recommendedLayout": "timeline-lanes",
    "laneField": "lane",
    "nodeTypeField": "nodeType"
  }
}

Consumers must treat unknown manifest fields as optional.

views.json

A view is a projection over the graph. It may define:

  • audience
  • guiding question
  • included or focused event ids
  • lane definitions
  • highlighted causal paths
  • story steps
  • teaching notes

A view must not duplicate canonical events or links.

Example:

[
  {
    "id": "teaching-causal-chain",
    "title": "Teach the causal chain",
    "audiences": ["teacher", "student"],
    "projection": {
      "trigger": ["instrument", "effective-date"],
      "response": ["action"],
      "adaptation": ["control"],
      "observableOutcome": ["evidence"]
    }
  }
]

Why views are generic

A compliance pack may use:

law → implementation → control → evidence

A World War I lesson pack may use:

structural cause → trigger → escalation → outcome → source evidence

A World Cup pack may use:

match result → table implication → affected team → next watchpoint

The domains differ, but all are narrative projections over event-and-link data.

Design rule: data is canonical, views are disposable

Do not encode a lecturer's slide order, an executive dashboard filter, or one UI layout into the causal graph itself.

Instead:

  1. Curate events and links once.
  2. Add provenance and honest confidence.
  3. Create multiple views for different jobs.
  4. Allow consumers to ignore view metadata entirely.

This avoids three common failures:

  • duplicating nodes for each audience
  • coupling data packs to one frontend
  • weakening causal semantics to satisfy a presentation layout

Suggested Canvas behavior

A Canvas consumer may:

  1. Load manifest.json when present.
  2. Load the canonical graph.
  3. Offer the view list from views.json.
  4. Filter or emphasize nodes according to the selected view.
  5. Render lanes from event metadata such as lane.
  6. Enter story mode using highlightPaths or storySteps.
  7. Always preserve access to sources and confidence.

Views should degrade gracefully. A consumer that only understands events.json, links.json, and insights.json must still load the pack successfully.