SpyderBot Documentation

Observatory — Changes

The Changes view answers what materially changed, with comparability and evidence checks before treating a fluctuation as a Change Event.

What the Changes view answers

The Changes view answers:

What materially changed?

Prompt Observatory is designed to distinguish meaningful change from normal AI variability.

A typical change flow is:

Observation
↓
Candidate Change
↓
Comparability check
↓
Materiality / Evidence check
↓
Change Event
↓
Change Investigation
↓
Change Drivers
↓
Evidence

Change event families

User-facing change families may include:

Recommendation
Citation
Entity
Competitive
Ecosystem
Benchmark
Operational

Not every small fluctuation becomes a material Change Event.


Severity vs Evidence Strength

Keep them separate.

Severity
= how important/material the detected change is

Evidence Strength
= how strongly the available evidence supports the finding

A High-severity event can still require careful interpretation if evidence is limited.


Change Drivers

Change Drivers answer:

Which evidence-supported factors moved with the observed change?

They do not automatically answer:

Which factor caused the external AI system to change?

Preferred driver interpretation:

Supported
Indicative
Unresolved

Next step

Validate with Evidence or inspect source movement in Sources.


Common interpretation rules

Use these rules across Prompt Intelligence:

0% can be a valid observed value.
Missing is not 0%.
Unknown is not Neutral.
Not Applicable is not poor performance.
Not Observed is not Brand absence.
No AI Answer is not Brand absence.
Not Comparable is not No Change.
A citation is not proof of causality.
A higher Prompt Priority does not increase an observed outcome.
More raw executions do not automatically mean stronger evidence.

Getting help from the Evidence

When a finding matters to a business decision:

  1. Confirm the Measurement Scope.
  2. Check Applicability.
  3. Check Evidence Strength.
  4. Check AI-system coverage.
  5. Check Comparability for temporal claims.
  6. Open Evidence.
  7. Review exact responses/citations/observations.
  8. Then decide whether action is justified.

This sequence is especially important for competitor recommendation changes, sentiment/portrayal changes, source ecosystem changes, Benchmark gap movement, and cross-system disagreement.

SpyderBot observes AI outputs and supporting evidence. It does not have access to the private internal reasoning of third-party AI systems. Prompt Intelligence distinguishes observed outcome, observed association, measured contributor, evidence-supported driver, and causal explanation. A citation appearing alongside a recommendation does not by itself prove that the citation caused the recommendation.

Measurement Scope describes what a result represents. Applicability answers whether a metric makes sense for this measurement scope. Comparability answers whether observation periods can be compared normally. Evidence Strength indicates how strongly eligible observations support a finding. Keep these meanings separate from Prompt Priority and Monitoring Health.