WHYTEvision™ SENTINEL™
AI-Powered
Website Intelligence
for The Learning People
February 2026 · WHYTEBOX AI
The Sprinkler, Not the Clipboard
Traditional website auditing is reactive investigation — a human auditor with a clipboard examining damage after the fact.
SENTINEL™ is the automated sprinkler system. It detects smoke and acts before damage spreads.
Detection over investigation
Automation over manual process
Evidence over opinion
Prevention over remediation
This isn't a metaphor for marketing. It's the architectural principle that governs every product decision.
What SENTINEL™ Does
SENTINEL™ deploys an 8-stage AI pipeline to any public website. In approximately 3 minutes, it:
  • Captures the page — video recording, DOM snapshot, screenshots
  • Classifies evidence — semantic, observational, red flag, user flow
  • Detects dark patterns — deceptive design, compliance violations
  • Applies legal frameworks — DMCCA 2024, UK GDPR, CMA, UCPD, DSA
  • Scores severity — each finding rated /10 by the AUDITOR
  • Delivers verdicts — Guilty or Not Guilty, determined by the JURY
  • Generates actionable assets — Engineer Ticket, Creative Brief, Strategy Protocol
  • Redacts PII — sanitised outputs ready for any audience
  • Validates adversarially — a Governor stage that challenges its own findings
Every finding is grounded in captured evidence. Every output is governed before it reaches you.

STATUS: LIVE at Coming soon…
8-Stage Governed Pipeline
SENTINEL™ deploys a sophisticated pipeline where evidence flows through parallel and sequential stages:
AGENT
Evidence Gathering
LIBRARIAN
Evidence Cataloguing
The Crucible: Parallel Analysis
Following the LIBRARIAN stage, evidence enters The Crucible, where three independent analysis streams execute simultaneously before converging.
3. AUDITOR
Analysis
3a. Brand Auditor
Brand-Specific Review
4. JUDGE
Legal Framework Application
These parallel analyses (all in Gold) then converge, feeding into the next sequential stage.
JURY
Verdict Determination
SYNTHESISER
Actionable Asset Generation
SANITISER
Output Validation
GOVERNOR
Adversarial Validation
The pipeline visualises in real time — you watch each stage activate, see the reasoning as it happens through ThoughtPulse™, and track the evidence chain.
The Sovereign Studio
When SENTINEL™ completes an audit, results appear in a 3-column intelligence workspace:
LEFT (20%): Evidence Vault
Classified evidence fragments from the LIBRARIAN. Four types: semantic (blue), observation (purple), red flag (red), user flow (amber).
CENTRE (60%): Split Stage
Video evidence player + live reasoning canvas. 16:9 aspect ratio. Play, seek, and scrub through the captured session. ThoughtPulse™ shows real-time AI reasoning with distinct choreographies per stage.
RIGHT (20%): Asset Studio
Three deliverables: Engineer Ticket (red), Creative Brief (purple), Strategy Protocol (amber). Copy to clipboard or download as PDF.
Evidence Bridge links thoughts to evidence fragments — hover a finding and the source evidence illuminates.

STATUS: All components LIVE
You Watch It Think
SENTINEL™ doesn't deliver conclusions from a black box. ThoughtPulse™ displays the AI's reasoning in real time as each pipeline stage executes.
AUDITOR reasoning: Soft Pulse
Findings emerge gently, appearing with a subtle fade-and-rise animation.
JUDGE reasoning: Velocity Stream
Legal analysis streams character by character with a typewriter effect and blinking magenta cursor, conveying precision and deliberation.
LIBRARIAN/SYNTHESISER: Narrative
Smooth contextual flow for evidence cataloguing and asset generation.
While reasoning is active, peripheral panels dim (Locus Focus) — directing attention to the active thought without hiding context.
Minimum 5-second display per thought — ensuring reasoning is readable, not just visible.

STATUS: LIVE (ThoughtPulse™, Locus Focus, all choreographies)
Jurisdiction-Aware Legal Analysis
SENTINEL™ automatically detects the regulatory jurisdiction of any audited website and applies the relevant legal frameworks:
UK
Digital Markets, Competition and Consumers Act 2024, UK GDPR, Competition and Markets Authority guidelines
EU
Unfair Commercial Practices Directive, GDPR, Digital Services Act, Digital Markets Act
Australia
Australian Consumer Law, Privacy Act
United States
FTC Act, UDAP statutes, California Consumer Privacy Act
The JUDGE stage maps findings to specific regulatory provisions — not generic "compliance" warnings, but citations to applicable law.
Default jurisdiction: UK (configurable per audit).

STATUS: LIVE — jurisdiction detection in production
Five Ways to Feed SENTINEL™ Evidence
GOLD — Sovereign Capture [LIVE]
Full agentic pipeline. Deploy SENTINEL™ to any public URL. Video, DOM, screenshots, evidence chain. ~3 minutes. SENTINEL™ controls the entire capture environment.
SILVER — Delegated Capture [STAGED]
Browser extension captures authenticated environments. Test behind logins, in staging, in gated experiences.
SILVER — External Capture [STAGED]
Analyse YouTube content — product demos, user testing recordings, onboarding walkthroughs.
BRONZE — Static Uplink [STAGED]
Upload screenshots, wireframes, PDF mockups. Interface designed, backend validated. SENTINEL™ integration in progress.
BRONZE — Kinetic Uplink [STAGED]
Upload screen recordings, QA walkthroughs, user journey captures.
The hierarchy reflects governance level — how much SENTINEL™ controls the capture environment. Gold = full sovereignty. Silver = delegated. Bronze = supplied evidence.
Three Artefacts. Four Export Classifications.
Every audit produces three deliverables, each shaped for a different audience:
Engineer Ticket (red)
Technical implementation brief. Prioritised by severity. The February 2026 audit of learningpeople.com produced a P0 Critical ticket targeting hidden navigation (ACC-001) and pre-selected cookie consent (DP-001).
Creative Brief (purple)
Strategic direction for design and content teams. The same audit generated "Ethical Urgency and Transparent Consent" — reframing compliance violations as brand opportunities.
Strategy Protocol (amber)
Executive-level moves. Three-phase structure: Contain (2 hours) → Correct (2 days) → Monitor (1 week). Business impact analysis with revenue-at-risk assessment.
Each artefact can be copied to clipboard or downloaded as PDF.
AUDIT TRAIL
The full 8-stage evidence chain is available in four classification levels:
  • Internal — full detail + raw JSON
  • Client — professional format
  • Legal — PII redacted
  • Technical — engineering review
Export format: YAML frontmatter + Markdown.

Full artefact examples are available on the WAIN Control Plane. STATUS: All artefacts and export classifications LIVE
The Governor: Adversarial Quality Gate
Before any finding reaches you, it passes through the GOVERNOR — an adversarial AI stage designed to challenge SENTINEL™'s own work.
A confidence score for the overall audit
A status: Approved or Blocked
A written rationale explaining its decision
This isn't a rubber stamp. The Governor is tuned for scepticism. It tests for consistency, logical coherence, and fabrication signals. If a finding doesn't survive adversarial challenge, it doesn't reach you.

STATUS: LIVE — Governor validation in production
Intelligence Heritage: PHOENIX™ → SENTINEL™
SENTINEL™ didn't appear from nothing. Its governed intelligence methodology was developed and proven in PHOENIX™ — the R&D laboratory where WHYTEvision™ was born.
1
DECEMBER 2025 — PHOENIX™ AUDIT 1
First governed audit against learningpeople.com. 5-phase pipeline. Judge verification. Governor oversight. 5 findings + 3 positive observations surfaced, including:
  • "Pinball Effect" — homepage visual overload from chatbot widget + toast notifications [CRITICAL]
  • Primary CTA styled as text link, not button [CRITICAL → ✓ Remediated]
  • Three competing CTAs — Analysis Paralysis [HIGH FRICTION]
Two findings remediated by the client within days.
2
JANUARY 2026 — PHOENIX™ EARLY WARNING SIGNALS
Three critical signals from multiple pipeline modes:
  • Grammar error in chatbot: "I'm a individual" [TRUST BREAK]
  • 40% viewport False Floor at first paint [FRICTION POINT]
  • Toxic PBN backlink (prime.itxoft.com) [SAFETY SAVE]
Plus a fourth finding first detected in December, re-confirmed as EWS:
  • Primary CTA — text with underline only [COMMERCIAL → ✓ Remediated]
3
FEBRUARY 2026 — SENTINEL™ PRODUCTION
Full 8-stage pipeline. 189.1 seconds. 6 findings. Severity 9/10. Governor approved at 100% confidence.
The transition from PHOENIX™ to SENTINEL™ was an infrastructure upgrade, not a methodology change. The intelligence was real from day one.
Intelligence Heritage: PHOENIX™ → SENTINEL™
SENTINEL™ operates in a continuous cycle. Findings have statuses. Fixes get re-checked. Partial fixes get caught.
FINDING 9: Primary CTA — Text Link Only
Detected: 23 December 2025 (PHOENIX™ Audit 1)
Re-confirmed: January 2026 (PHOENIX™ EWS)
Client action: Accepted, actioned, logged
Re-detection: No longer detected
Current status: ✓ REMEDIATED
Two independent pipeline runs converged on the same finding. The client actioned it. The system confirmed the fix.
FINDING 6: Grammar Error — "I'm a individual"
Detected: 7 January 2026 (PHOENIX™ EWS)
Client action: Fix deployed
Re-detection: Partial fix — chatbot alternates between "Cora" and "Jasmine" personas. Grammar error persists intermittently in some variants ("a" vs "an").
Verification: Manual verification scripts confirm intermittent nature.
Current status: ⚠ PARTIAL FIX
FINDING 8: Toxic PBN Backlink
Detected: 7 January 2026 (PHOENIX™ EWS)
Remediation: TEP+ Active — R-TOX workstream ongoing
Current status: ⏳ IN PROGRESS
Finding 6 is the proof of concept. The client "fixed" it. The system caught that the fix was incomplete. That's the sprinkler detecting smouldering embers after the first response.
What's Next
MILESTONE 1: User Acceptance Testing (UAT)
You test SENTINEL™ directly on your pages. The Sovereign Studio, the pipeline, the evidence chain — hands-on with the production system. Gamma 2.0 is your preparation for this experience.
MILESTONE 2: Day Zero Intelligence Debriefing
The first live session where SENTINEL™'s findings on The Learning People's properties are presented, discussed, and translated into action.
MILESTONE 3: Operational Onboarding
SENTINEL™ becomes part of your workflow. Recurring audits, data integration, team access, and the capture tier expansion (Silver and Bronze modes).
These milestones are sequential. UAT validates the system. Day Zero proves its value on your content. Onboarding makes it operational.
On the Roadmap
These capabilities are designed and validated but not yet production-active:
BROWSER EXTENSION (Silver Tier)
Capture authenticated environments — staging sites, gated content, post-login experiences. API route exists; integration in progress.
YOUTUBE ANALYSIS (Silver Tier)
Analyse product demos, user testing recordings, onboarding walkthroughs via YouTube URI.
STATIC UPLOAD (Bronze Tier)
Upload screenshots, wireframes, PDF mockups for visual analysis without browser capture. Interface designed, backend validated. WHYTEvision™ integration in progress.
KINETIC UPLOAD (Bronze Tier)
Upload screen recordings and QA walkthroughs for temporal analysis.

STATUS: All STAGED — designed and validated, not production-active. No timeline commitments. Priority informed by your feedback during UAT.
Next Steps
Schedule UAT session
Walk through SENTINEL™ on The Learning People's live pages. See the pipeline, the evidence, the outputs.
Connect your data
Upload GA4 and Search Console exports to unlock data-fused intelligence and Warm Start recommendations.
Share feedback
Your priorities shape what ships next. The capture tier expansion, the analytics depth, the output formats — tell us what matters most.
WHYTEvision™ is live at coming soon….
The WAIN Control Plane is at wain-powerpatch-lite.netlify.app
WHYTEBOX AI
Intelligence that discovers value, not pushes it.
WHYTEvision™ SENTINEL™
Appendix
Appendix: Where SENTINEL™ Came From
PHOENIX™ was the R&D laboratory where WHYTEvision™'s governed intelligence methodology was developed. Built in the WHYTElab to assist with WHYTEBOX AI's own ecosystem development operations. We built this for ourselves first.
Key timeline:
December 2025: First governed audit (5-phase pipeline)
January 2026: Architectural pivot to Gemini 3 series
January 2026: Early Warning Signals (3 critical findings)
February 2026: SENTINEL™ production launch (8-stage pipeline)
The transition path:
PHOENIX™ ran a 5-phase governed pipeline with adversarial oversight. When Gemini 3 throughput made real-time agentic workflows viable, the architectural pivot produced SENTINEL™ — same methodology, production-grade infrastructure.
Product-market fit was discovered organically. PHOENIX™ audited real client properties and produced real intelligence. Two of three January 2026 findings were remediated by the client within days of notification. The intelligence produced during the PHOENIX™ period represents genuine, governed analysis of Learning People's web properties using the same principles that underpin production delivery.
Appendix
Appendix: The Engine — and How We Govern It
SENTINEL™ applies the same transparency and evidence standards to itself that it applies to audited websites. Show your working.
MODELS
  • Primary: Gemini 3 Pro-Preview
  • Secondary: Gemini 3 Flash-Preview
INFRASTRUCTURE
Frontend: Next.js 16.1.3, React 19.2.3, TypeScript (strict), Tailwind CSS 4, Framer Motion 12. Hosted on Netlify.
Backend: Express.js v4.1.0 on Cloud Run gen2 (europe-west2). Browser automation: Playwright (Chromium). Storage: Google Cloud Storage. Real-time: SSE with 5-second polling fallback. Auth: Bearer token.
Build: 0 TypeScript errors.
TWO COGNITIVE MODES:
SENTINEL™'s pipeline stages operate in two distinct modes, calibrated to each stage's function:
Thinking Mode — AUDITOR and JUDGE
Extended reasoning with visible thought chains. These stages run Gemini 3 Pro-Preview with thinking enabled, producing the structured internal reasoning that feeds ThoughtPulse™. You watch them think through the evidence because they are configured to think out loud. The thinking budget (8,192 tokens of internal reasoning per call) is the precision control — not temperature.
Standard Mode — all other stages
Precise, low-variance execution tuned to each task. The analytical stages (LIBRARIAN, JURY, GOVERNOR) use Gemini 3 Pro-Preview for depth. The output stages (SYNTHESISER, SANITISER) use Gemini 3 Flash-Preview for speed. All run at low temperature (0.1) — the task is well-defined and doesn't benefit from exploratory reasoning.
The GOVERNOR applies its scepticism through an adversarial prompt, not through a different cognitive configuration. It uses the same precision model as the stages it oversees.
The same pipeline that audits your properties is itself governed, version-controlled, and adversarially tested.