Last updated April 21, 2026
Sources
Architecture diagrams rot fast when they're disconnected from reality. Sources are how ContextDX stays grounded — they connect your boards to the actual systems, repos, documents, and infrastructure that your architecture describes.
Prerequisites
- Workspace created with at least one board
- Integration credentials configured at the org level (for OAuth sources)
- Access to the external system you want to connect (repo permissions, Confluence space access, etc.)
How sources work
Sources live in a workspace-level catalog. Each source is registered once, then bound to one or more boards. The same Confluence space can feed multiple boards without being imported twice.
Rendering diagram...
Source types
These require org-level integration credentials configured in advance.
| Source | Description |
|---|---|
| Confluence | Atlassian Confluence pages and spaces |
| Google Docs | Google Docs documents |
| GitHub | GitHub organization or account level |
| GitHub Repo | Specific GitHub repo |
| GitHub File | Individual file in a GitHub repo |
| Jira | Jira projects and issues |
| Slack | Slack channels and conversations |
| Notion | Notion pages and databases |
| Intercom | Intercom conversations and articles |
Source roles
When you bind a source to a board, you assign it a role that determines how it participates:
| Role | Behavior |
|---|---|
| Governing | Contributes to graph structure. Nodes and edges are extracted and synced to the board. Participates in conflict detection. |
| Reference | Attached as context only. Chat and insights can read it, but it doesn't change the board's structure. |
A board can only have governing sources of one classification type — you can't mix push-based (smart) and extraction-based (dumb) governing sources on the same board.
Sync lifecycle
Each source binding moves through a strict sequence as it syncs:
| State | What it means |
|---|---|
| Unsynced | Newly bound, not yet processed |
| Evidence collected | Raw evidence extracted from the source |
| Embeddings generated | Semantic embeddings created for matching |
| Conflicts checked | Cross-source conflicts identified |
| Synced | Fully processed and applied to the board |
States progress in order. If a sync fails partway through, the binding stays at the last completed state and can be retried from there.
Health tracking
Sources are monitored for connectivity:
- Healthy — Accessible and syncing normally
- Degraded — Accessible but experiencing issues
- Failing — Not accessible or sync is broken
Trust scores
ContextDX computes a trust score (0–10) for each source based on several weighted factors:
How the trust score is computed
| Factor | Weight | What it measures |
|---|---|---|
| Accuracy | High | How well the information matches known facts and other corroborating sources |
| Recency | Medium | How recently the source was updated — stale docs score lower |
| Conflicts | High (negative) | Count of conflicting evidence from other sources — more conflicts = lower trust |
| Verifications | Medium | Count of corroborating sources that agree — more agreement = higher trust |
A source with a trust score below 4 gets a visual indicator in the UI. Below 2, the binding card turns red. This doesn't block anything — it's a signal to the architect that the source might need attention.
Governing sources are your "single source of truth" for a board. Reference sources give chat extra context without changing your diagram. When in doubt, start with reference and promote to governing once you trust the data.