--- title: "After probing 86,599 x402 endpoints, we're clarifying where x402station sits in the stack" slug: signal-layer-after-86k-probes canonical: https://x402station.io/blog/signal-layer-after-86k-probes date: 2026-05-15 tags: [x402, agentic-commerce, positioning, signal-layer] --- # After probing 86,599 x402 endpoints, we're clarifying where x402station sits in the stack Two weeks of continuous probing have taught us something concrete about our seat at the table. This post is a public update on what we are, what we are not, and why we're tightening our positioning now rather than later. ## What we measured Between 2026-05-01 and 2026-05-15, x402station.io probed **86,599 active x402 endpoints** drawn from the agentic.market catalog. We probe each endpoint every 10 minutes, merge results with CDP settlement data, and classify against a five-signal taxonomy: zombie (CDP-confirmed zero-pay across 30+ days), decoy (HTTP 300+ second anti-scraper traps), concentration risk (single-provider catalog dominance), never-paid catalog ghosts, and settlement-recency drift. Two findings stand out from the fresh export: - **88.07% of the public x402 catalog is concentrated in two providers.** Not 86% -- 88.07%, and the gap to the next eight is wide. - **Six of the next eight providers by share have zero confirmed settlements in 30+ days.** They are catalogued, they look operational, agents will pay them -- and the on-chain evidence says nothing settles. These are the kinds of signals an independent probing layer can produce. They are not signals a marketplace, a facilitator-aggregated catalog, or a policy engine produces natively. Those layers consume signals; they do not generate them from probes. ## Where we sit in the stack Until this week our public description has been **"infrastructure for x402 agentic commerce"**. Accurate, but unhelpfully broad. As the ecosystem has matured -- Linux Foundation chartered the x402 Foundation in April, AWS shipped AgentCore Payments preview in May, and Ampersend (built by Edge & Node) has shipped a Control Layer with policies, approvals, and audit trails for the agent economy -- the stack has resolved into clearer roles. We have a specific seat in it, and ambiguity helps no one. So, the update: > **x402station.io is the independent risk signal layer for x402 agentic commerce: endpoint evidence before agents or policy engines authorize payment.** Concretely, that means: - **We probe and classify.** We do not decide whether your agent should pay. We tell you what we measured about the endpoint. - **We surface evidence policy layers query.** Whether the policy engine is Ampersend's Control Layer, AWS AgentCore Payments' spend-limit checks, or a one-off `if (preflight.warnings.includes("never_paid_zombie")) refuse()` in an agent's own code, they query us for endpoint truth before making their decision. - **We author open conventions for the data.** `x402-signals` v0.2.0 (CC0) is our standard for refund and fulfillment vocabulary, with ReloadPI as first field implementer on Base mainnet. Anyone implementing the spec, including future policy layers, gets a shared semantic for outcome state. ## What we are not We are not a spending firewall. We are not a control plane. We are not a policy engine. Those positionings exist in the stack and are already occupied by teams with substance, Ampersend in particular, with their `ampersend-sdk`, AWS AgentCore integration, and CrewAI hookups. Trying to be everything at once would mean copying a positioning we cannot defend today. The signal layer is what we have actually built. Two weeks of probing made the distinction visible. Our job is to be best in class at it. ## Why this is a maturation, not a pivot A useful test for a positioning update: does the work change, or does the description of the work change? Our daily probing cadence, signal classification taxonomy, paid endpoint surface, open spec authorship, and Verified-cert flow are unchanged from yesterday and the day before. What changes is how we describe the seat in the stack: sharper words, the same engineering underneath. If anything, the update commits us harder to the work. Signal layer is a layer with a moat: 30+ day probe history and CDP settlement merge are not replicable on a one-month timescale. Marketplaces and policy layers do not natively measure these things. That is our defensibility, and it is the part of the stack we are committed to deepening. Next on our roadmap: proxy and resell detection, continued spec authorship on v0.3 with cross-protocol vocabulary, and a thin client-side policy helper SDK for agents that want a one-call decision tree on top of our preflight. ## What this means for you If you are building an agent or framework that calls x402 endpoints, call `/api/v1/preflight` before signing PAYMENT-SIGNATURE. The trial endpoint is free at `https://x402station.io/api/v1/preflight-trial` (10 calls per IP per UTC day). Five recipe pages cover AgentKit, Cloudflare Agents, LangChain, Lucid, and Vercel AI SDK at `/guard/recipes/...`. If you are building a policy engine, control layer, marketplace, or any other layer that needs endpoint truth before authorizing payment, our signal layer is exactly what you query. Drop us a line at `hello@x402station.io` and we'll align on integration patterns. If you are a researcher, journalist, or Foundation member following the x402 ecosystem, the public concentration data is at `https://x402station.io/reports/provider-concentration.json`. The Hugging Face dataset snapshot is at `https://huggingface.co/datasets/x402station/preflight-dataset-v0_1`. Refresh to the current 86k row count is in flight. Counter-arguments, corrections, and signal categories we should be probing are all welcome. The independence of the signal layer depends on getting the measurement right, and the measurement gets better when the people who depend on it tell us where it is wrong. **Team (x402station.io)** hello@x402station.io https://x402station.io https://github.com/sF1nX/x402-signals