Skip to main content

​
Relationships, trust and risk

The model library keeps relationships, trust and risk in separate domains because they answer different questions. A relationship says how records are connected. A trust field represents an interpretation. A risk field identifies potential concern or exposure. None of those should silently overwrite canonical PressChain publication state.

​
Relationship graph

Useful relationship edges can include:
  • author to Capsule
  • outlet to publication
  • contributor to Capsule
  • evidence to Capsule
  • source to claim
  • revision to earlier publication state
  • claim to supporting or contradicting evidence
Graph-oriented applications can build source maps and timelines from these edges without reducing every relationship to ownership.

​
Trust interpretation

Trust assessments should remain attributable to the system and rule that produced them. If an application displays an external confidence or trust value, preserve producer, model or ruleset version and observation time where available. Do not write an external score into a Capsule status field. That would turn an analytic interpretation into a false protocol fact.

​
Risk signals

Risk can be useful for triage, review and bounty routing. A risk signal is still a signal. A high-risk classification does not prove a publication false, and low AI confidence does not invalidate a source. Present the supporting evidence or reason category whenever possible.

​
Visibility

Visibility is a separate concern as well. An application can combine public protocol state with restricted newsroom analysis, but restricted data must be enforced at storage and API boundaries. Hiding a field in the browser is not access control.

​
Five useful provenance classes

When combining systems, classify fields as one of:
  1. canonical PressChain state
  2. projected PressChain state
  3. externally sourced fact
  4. analytic interpretation
  5. application-only metadata
That simple taxonomy prevents sophisticated interfaces from becoming semantically ambiguous.

​
Example: investigation graph

A newsroom tool can show a Capsule connected to an outlet, two contributors, three evidence items, five extracted claims and one external risk signal. The first four relationships may be derived from canonical or projected PressChain state. The risk signal may come from a separate analysis service. The interface can display all of them together while preserving the source boundary for each edge. That is more useful than forcing every piece of information into one global “trust score.”