Visible World Audit Production Candidate Validation

Audit

FROZEN

Template

2-Hour Visible World Diagnostic v1

Client

SHIRO & Co. Internal

Target

Visible World Lab

Personas

3

Channels

WEB · EMAIL

Runs

3 / 3

Comparisons

6 / 6

Evidence

COMPLETE

Report

FROZEN

Delivery

REVOKED

Engagement

DELIVERED

• REPORT FROZENYES
• CLIENT VIEWREADY
• SHARE LINKREVOKED
• PRINTREADY
• EVIDENCE APPENDIXREADY

Client

SHIRO & Co. Internal

Target

Visible World Lab

Observed

3 personas · 2 channels · 3 runs · 6 comparisons

Result

A VWD 0.400 · B VWD 0.415 · Control VWD 0.000

Delivery

Frozen report · Client view · REVOKED

Visible World Divergence is an experimental internal metric that compares observed exposure-set overlap, rank movement, and category distribution. It is not a validated industry standard. It is not a judgment that a system is good, bad, safe, unsafe, biased, or unbiased. This is not a pass/fail judgment on the client’s algorithm.

Client

SHIRO & Co. Internal

Project

Production Validation

Reference

VWA-2026-001

Created by

Phase 7E-5 synthetic production candidate

Template

2-Hour Visible World Diagnostic v1

Duration format

Approximately 2 hours

RANKING DIVERGENCE

Observe whether ranked exposure diverges after different recorded interactions.

• What changed after the recorded interactions?
• Did Persona A and Persona B receive different visible worlds?
• Did the neutral control remain stable?
• Did email exposure precede later web exposure?

Platform

Visible World Lab

Base URL

https://visible.shiroand.io/lab/recommendation

Target type

LAB_SYNTHETIC

Experiment

Recruitment Visible World Demo

Confirmed · 2026-09-20T04:26:11.483Z

I confirm that SHIRO & Co. is authorized to observe this system.

Internal synthetic production validation. Not a real client authorization. Controlled Lab target only.

• Senior Marketing / AI InterestSynthetic experimental persona
• Senior Marketing / Security InterestSynthetic experimental persona
• C — Neutral ControlCONTROL PERSONASynthetic experimental persona

CONTROL. No active interaction sequence is planned for this persona unless explicitly configured.

WEB · EMAIL

• VISIBLE WORLD DIAGNOSTIC
• Persona A: 1 baseline observation, 3 controlled interactions, 1 refreshed observation
• Persona B: 1 baseline observation, 3 controlled interactions, 1 refreshed observation
• Persona C: 1 baseline observation, 0 interactions, 1 refreshed observation
• Comparisons: 5
• Channels: WEB, EMAIL
• Email optional unless selected
• Deliverable: Frozen audit report

• RUN Senior Marketing / AI Interest — observation run
• RUN Senior Marketing / Security Interest — observation run
• RUN C — Neutral Control — CONTROL PERSONA
• COMPARISON A temporal
• COMPARISON B temporal
• COMPARISON C temporal/control
• COMPARISON A vs C
• COMPARISON B vs C
• COMPARISON Email sequence → Web run · optional
• EMAIL_IMPORT Import expected synthetic or authorized email events

• Frozen Visible World Audit Report
• Visible World comparison
• Exposure Chain
• Category shifts
• Evidence appendix
• Known / Inferred / Unknown separation
• Control comparison where configured
• Read-only client report

• A temporalCOMPLETE
• B temporalCOMPLETE
• C temporal/controlCOMPLETE
• A vs CCOMPLETE
• B vs CCOMPLETE
• Email sequence → Web run · optionalCOMPLETE

7 stored · 2 unknown. Classification is optional.

MANUAL

No linked schedules.

• Observations cover the defined audit window only.
• Visible World Audit observes outputs, not proprietary source code.
• Unseen candidates cannot be assumed to have been deliberately suppressed.
• Semantic classifications are inferences from observed text.
• Causal relationships are not established unless separately demonstrated.
• Visible World Divergence is an experimental internal metric that compares observed exposure-set overlap, rank movement, and category distribution. It is not a validated industry standard. It is not a judgment that a system is good, bad, safe, unsafe, biased, or unbiased.
• OBSERVED EMAIL OPEN EVENT. OPENED does not necessarily represent human reading. Some email clients may preload or proxy content. OPENED is not equivalent to engagement.
• This engagement used a controlled synthetic target, not an external production ranking system.

• target · requiredTarget platform and base URL are present.READY
• authorization · requiredAuthorization to observe this system is confirmed.READY
• personas · required3 persona(s) selected.READY
• comparisonPlan · required6 planned comparison(s).READY
• database · requiredPostgreSQL is reachable.READY
• evidenceStorage · requiredEvidence storage is ready.READY
• runner · requiredObservation runner is READY.READY
• openai · optionalClassification is available.READY
• schedule · optionalManual-only execution. Scheduler is not required.OPTIONAL
• email · requiredEmail ingestion is ready. Inbox scraping is not used.READY