# How to read an agentic payments dashboard without being misled Agentic payment dashboards usually show accurate figures over populations they do not define. Before reading one, establish which chains it covers, what window, what evidence standard decided what counts, whether self-dealing was removed, and how stale the data is. Without those, a chart is not comparable with any other chart. Published: 2026-10-14 Last reviewed: 2026-10-14 Publisher: Roundhouse AI Ltd. Source: https://roundhouseai.io/resources/reading-an-agentic-payments-dashboard --- ## Why are correct numbers still misleading? Almost every dashboard in agentic payments is computing something correctly. The misleading part is that the thing being computed is not the thing the label says, and the label is the only part most readers see. "Settled volume" can mean gross transfers with an x402-shaped marker, or facilitator-relayed settlements, or either of those net of self-payments. Those are three different quantities, often differing by an order of magnitude, under one heading. ## Six things to establish first **1. Which chains.** Pooled multi-chain totals cannot be compared with single-chain ones, and chain shares in this market move fast. Solana was reported at roughly 70% of transaction activity by mid-August 2026 while Base held the cumulative lead, and weekly shares shifted repeatedly through the year. **2. What window.** Cumulative since launch, trailing 30 days and last week differ enormously. A chart with no stated window is usually cumulative, which flatters. **3. What evidence standard.** Every gasless transfer, or only those a named facilitator relayed. [This difference is large](https://roundhouseai.io/resources/what-proves-a-transfer-was-an-x402-payment) and it is almost never stated. **4. What was excluded.** Self-payments, concentrated flows, low-counterparty sellers, or nothing. Independent analyses in 2026 found this exclusion alone roughly halving the figure. **5. How stale.** How far behind the chain tip. A dashboard that does not publish its lag is asking you to assume it is live. **6. Transactions or dollars.** They move independently. A protocol with a few-cent median payment looks healthy by count and thin by value, and the same data supports both stories. > [!NOTE] > A figure that states its basis, its window and its coverage can be compared with another figure that does the same. A figure without them cannot be compared with anything, including an earlier version of itself. This is why the metadata is the useful part rather than the fine print. ## Charts that mislead honestly Three shapes to watch for, all of which can be produced without anyone intending to deceive. **A total next to a filtered breakdown.** A headline computed over everything, sitting above a breakdown computed over a verified subset. Both correct, and the reader adds them up. The fix is to filter at the top so every figure on the page describes one population. **A ranking where most entries are unranked.** A leaderboard sorted by a score most rows do not have. The order looks meaningful and is mostly arbitrary. A ranking should say what fraction of the page is actually scored. **A total counted off a capped page.** A system reads a page of rows, hits a row limit, and counts the rows it got. The result is the limit, not a count of anything, and it is stable and plausible-looking. Any total that happens to be a round number like 1,000 deserves suspicion. ## What does a good dashboard do? Four things, and you can check for them in under a minute. - **Every figure states its basis and window** in or next to the label, not in a methodology page nobody opens. - **Empty and broken look different.** A failed read must not render as zero. On a payments index, zero reads as "this business never traded", which is a claim rather than a blank state. - **Coverage is published.** Which chains, from which block, how far behind the tip. - **Uncertainty is kept rather than resolved.** A three-valued evidence grade that keeps "unestablished" separate from "disproven" is more honest than a binary flag, because the binary forces every uncertain row into one of two wrong buckets. ## How should an agent read one? If you are a program consuming these figures rather than a person looking at a chart, the same rules apply with one addition: prefer the API to the page. A well-built index publishes the applied parameters in the response rather than echoing what you asked for, so you can tell when your window was silently widened or your filter ignored. It also distinguishes an empty result from a refused read, which matters far more to a program than to a human who can see the page is broken. Figures without that metadata should be treated as indicative, not as inputs to a decision about money. ## Frequently asked questions ### Is a bigger number always the inflated one? Usually, but batched payments invert it: a deferred scheme settles once for many calls, so a settlement-based figure understates activity for those sellers. ### Should I trust a facilitator's own figures? They are useful and partial. Each facilitator sees its own traffic, so none of them sees the market, and nobody else can check the numbers. An index built from the chain sees all of them under one method. ### How do I compare two dashboards? Only like for like: same chain, same window, same evidence standard. When that is done the agreement is often good. Roundhouse's Base figures for a 90-day window in 2026 landed within about 1% of Token Terminal's for the same chain and window, where the blended comparison had been off by a huge margin. ### What if a dashboard does not publish its method? Treat its figures as unusable for comparison. That is not a judgement about the team; it is what the absence of method means. ### Why do figures change retroactively? Reorgs, backfills and reclassification. An index that improves its evidence standard will restate history, and one that quietly does so without saying is worse than one that does it openly. ## Where to check this yourself [/stats](https://roundhouseai.io/stats) states the basis, window and chain for every figure and separates the verified band from the total. [Coverage and confidence](https://roundhouseai.io/docs/data/coverage-and-confidence) documents what each evidence grade asserts, and [/status](https://roundhouseai.io/status) publishes how far behind the chain tip the index is, which is the number most dashboards omit. --- Roundhouse indexes x402 settlements and ERC-8004 identities into one public record. - Every article: https://roundhouseai.io/resources.md - Every surface on this site: https://roundhouseai.io/llms.txt - The index, endpoints and schema: https://api.roundhouseai.io/v0 - A key of your own, $0.01 over x402, no account: https://api.roundhouseai.io/v0/test/x402 - Limits and the SQL surface: https://roundhouseai.io/docs/api