The Four-Layer Stack: Why Retrieval Simulation Needs Entity Data to Mean Anything
You cannot model how AI systems retrieve your content without first knowing which entities they have extracted from it. The SiteNexis intelligence stack enforces this dependency order — and it matters for every result you see.
SiteNexis produces scores across four intelligence tiers. The scores are not independent — each tier depends on the output of the tier below it. Layer 1 (crawl and structure) must complete before Layer 2 (semantic intelligence) can begin. Layer 2 must produce entity data before Layer 3 (AI Visibility) can compute retrieval readiness. Layer 3 scores are required inputs to Layer 4 (Machine Trust), which runs the retrieval simulation, machine trust analysis, temporal authority model, recommendation surface mapping, and synthetic entity detection. This dependency order is not a technical constraint imposed by our architecture — it is a logical constraint imposed by how AI retrieval systems actually work.
Why Retrieval Simulation Requires Entity Data
The Retrieval Simulation Engine models six stages of AI retrieval: chunk extraction, ranking pressure, summarisation degradation, context truncation, answer formation, and citation eligibility filtering. At every stage, entities are the unit of analysis. Chunk extraction quality depends on whether entity mentions create natural chunking boundaries. Ranking pressure is estimated using entity authority scores. Summarisation degradation is measured by how many entity-attribute pairs survive compression. Without a complete entity graph from Layer 2, the simulation produces numbers that are structurally meaningless — there is no entity foundation to anchor the retrieval model to.
Why Machine Trust Requires Retrieval Scores
Machine Trust analysis determines whether AI systems would believe the content that retrieval surfaces. To assess believability, the system needs to know which content actually reaches the retrieval layer. A page with high extraction fidelity but low retrieval rank contributes less to the trust model than a page with both high extraction fidelity and high retrieval probability. The trust analysis weighs entity credibility signals in proportion to the retrieval probability of the pages that carry those signals. Without retrieval scores from Layer 3, the trust model weights every page equally — which misrepresents how AI systems actually evaluate domain trust.
▲Layer 4 analysis on SiteNexis is gated to Pro and Agency plans not only because of compute cost but because it is genuinely meaningless without the Layer 1-3 foundation. Running a retrieval simulation on a site with unresolved entity inconsistencies produces unreliable results. The tier gate enforces the logical dependency order.
The Architecture as a Dependency Graph
The SiteNexis agent architecture enforces this dependency order through its execution phases. Phase 1 runs the crawl agent — no analysis can start without crawl data. Phase 2 runs technical SEO and schema agents in parallel. Phase 3 runs retrieval, entity, and performance agents — which depend on Phase 2 outputs. Phase 4 runs citation and semantic trust agents. Phase 5 runs all Layer 4 agents simultaneously, each consuming outputs from earlier phases. Phase 6 runs the visualization agent on the completed perception graph. Phase 7 generates the PDF report. Every phase is a logical dependency boundary, not an arbitrary sequencing choice.