SHIRO & Co.

Visible World Audit

Technology · EXPERIMENT · ACTIVE

Neutral vs AI Framing Audit

Observe what a system makes visible when the same subject is presented under different framings.

EXPERIMENT — OBSERVATIONS RECORDED · OBSERVED · DIFFERENCE OBSERVED

Target platform
Visible World Lab
Target base URL
https://visible.shiroand.io/lab/recommendation
Comparison mode
Framing comparison
Purpose
Compare visible options under Neutral, AI, and Security framing without reusing prior runs.
What is observed
What becomes visible under each framing condition.
What is not concluded
Does not claim that framing causes bias. Existing Lab Neutral / AI / Security runs are not attached automatically.
Condition setup
Configured conditions: Neutral framing, AI framing, and Security framing of the same product-marketing role discovery task. Only framing language and the matching observation sequence change.
Personas
3
Observation runs
3 · OBSERVED
Evidence claims
38

Observation chain

  1. SIGNAL→
  2. INFERENCE→
  3. CANDIDATE SET→
  4. RANKING→
  5. EXPOSURE→
  6. USER ACTION→
  7. NEXT EXPOSURE

Personas

Manage
  • Neutral framingEnabled · framing-default
  • AI framingEnabled · framing-ai
  • Security framingEnabled · framing-security

Runner observations

Individual run history is retained. Statuses are not overwritten in place.

Condition comparison

INFERENCE

DERIVED COMPARISON · DIFFERENCE OBSERVED

Derived from runner-observed option sets and positions. This is a comparison, not a judgment of cause or intent. Ranking differences are observations of exposure, not conclusions about intent or fairness. This table is not an overall score.

Rank Exposure Distance describes how many shared options changed position relative to the control, plus Top 5 / Top 10 entry and exit, and the largest absolute rank change. It is a descriptive comparison, not a quality score.

ConditionVisible countSharedOnly hereRank movementTop 5 +/-Top 10 +/-Largest moveRun status
Neutral framingControl2424—0 / 240 / 00 / 00COMPLETEDEvidence
AI framing2424—20 / 24+2 / -2+3 / -311COMPLETEDEvidence
Security framing2424—23 / 24+3 / -3+3 / -311COMPLETEDEvidence
  • Rank Exposure Distance · AI framingChanged positions: 20 / 24 · Top 5: 2 entered, 2 exited · Top 10: 3 entered, 3 exited · Largest movement: +11
  • Rank Exposure Distance · Security framingChanged positions: 23 / 24 · Top 5: 3 entered, 3 exited · Top 10: 3 entered, 3 exited · Largest movement: +11

Ranking differences are observations of exposure, not conclusions about intent or fairness.

OBSERVEDSTATEDINFERENCEUNKNOWNNOT ESTABLISHED

Provenance describes the source of a statement. Causal status is separate: NOT ESTABLISHED means a sequence may have been observed without demonstrated cause.