ZCET: The Agentic Selection Stack

The transition from “Navigational Search” to “Agentic Selection” has rendered traditional SEO obsolete. AI agents do not browse websites; they evaluate facts in latent space. Brands failing to provide machine-executable, verifiable data suffer from Navigational Debt, leading to structural exclusion from the agentic economy. ZCET solves this by refactoring brand data into a deterministic PICSA-A Selection Stack.

ZCET operates across three isolated trust zones to ensure PII containment and technical veracity:

Build-time / Publication Plane (Internal): Ingestion, Fact-Modelling, Vectorisation, and Signing. This plane holds source content and PII; it is never exposed to agents.

Runtime / Agent-facing Plane (Edge): The Agent Gateway and published artifacts (JSON-LD + C2PA assets). This is the only surface reachable by external agents.

Accountability Plane (Observation): The t3ai-Ledger. It observes the publication and transaction seams without sitting in the data path, providing a tamper-evident audit trail.

Presence: Active Ingestion Push (AIP) replaces passive crawling. Data is chunked, tagged, and pushed into high-authority corpora (Common Crawl, registries) to ensure agentic perception.

Interpretability - Pillar A: “The Lock”. Facts are modelled as triples (Subject → Predicate → Object) using Schema.org. A deterministic record_id (UUIDv5) is minted for every claim, ensuring referential integrity across the stack.

Interpretability - Pillar B & Selectability: “The Hook” generates semantic fingerprints (embeddings) proximal to target intent. Multi-dimensional tags (cost, sustainability, delivery speed) align the brand with the agent's objective functions (cost-min vs. quality-max).

Credibility: Every asset is signed with a C2PA manifest and embedded with watermarks surviving compression. A manifest_id links asset-level provenance to decision-level ledger records.

Selection without completion is a dead end. The actionability gate exposes the brand’s capabilities as machine-navigable interfaces:

OpenAPI 3.1 & MCP: Provides a documented, callable surface for agents to execute transactions (e.g., live pricing, bindable holds).

Non-Human Auth: Utilises OAuth 2.0 client-credentials and scopes under Infrastructural Sovereignty.

Gating Capability: A brand that cannot be transacted with programmatically is structurally excluded, regardless of its ranking.

The t3ai-Ledger provides the layer PICSA was missing: a signed, append-only record of AI-mediated decisions.

Publication Seam: Records what the brand published and who approved it.

Transaction Seam: Records what the agent did at the gateway, referencing the record_id consumed.

Verification: An independent verifier binary proves an unbroken hash-chain on exported archives, satisfying Article 12/26 requirements for high-risk AI systems.

ZCET replaces Click-Through Rate (CTR) with the Selection Probability Score (SPS). This cross-cutting metric measures the brand’s transition from “Invisibility” to “Logical Choice” across the entire funnel. SPS is reproducible from ledger and gateway data, providing the clinical ROI proof of implementation.

As of 23 August 2026, ZCET provides the only clinical framework for:

EU AI Act Art. 50: Mandatory machine-readable marking and provenance for AI-generated/mediated content.

GDPR: PII containment via crypto-shredding of object keys in the erasable F1 Content Store.

Sovereign Residency: Hosting on academic infrastructure (University of Patras) to mitigate FISA 702 exposure.