--- type: Article name: What happens when your data disagrees with itself. id: "https://selfe.ai/insights/trust-and-commerce-data/what-happens-when-your-data-disagrees-with-itself" url: "https://selfe.ai/insights/trust-and-commerce-data/what-happens-when-your-data-disagrees-with-itself" publisher: "did:web:selfe.ai" description: "A product can be in stock on the website, sold out on the warehouse system, and back-ordered on the marketplace listing, all at the same moment, and an AI agent checking any one of the three has no way to know it is only getting a third of the picture." datePublished: "2026-09-15T05:32:19.119Z" author: The Selfe team --- # What happens when your data disagrees with itself. > A product can be in stock on the website, sold out on the warehouse system, and back-ordered on the marketplace listing, all at the same moment, and an AI agent checking any one of the three has no way to know it is only getting a third of the picture. ## In short - The same product can show three different stock statuses across three channels at once, and each one can be individually true at the moment it was last updated. - Three sources disagreeing is more informative than one confident page, because it signals that at least one of them is stale before anything else is known. - The lasting fix is one shared source every channel reads from directly, not tighter syncing between three separate copies. The same product can be in stock on one channel and sold out on another at the exact same moment, with neither page lying. What separates a usable fact from a stale one here isn't confidence, it's which source was touched most recently. ## The disagreement itself is the useful signal. A skincare brand sells its best-known serum through its own site, a marketplace listing and a wholesale partner's shop, all pulling from the same physical stock. A busy afternoon sells through the last units, and the brand's own site updates within the hour, but the marketplace listing, synced overnight, still says in stock the next morning. An agent checking the marketplace at ten o'clock has no way to know it is reading yesterday's truth. Three sources disagreeing on the same underlying stock level tells an agent something specific: that at least one of them is stale, before it even works out which. A single confident-looking page gives no such warning. Three pages quietly contradicting each other is, oddly, more informative than one page stating a clean number nobody has checked recently, because the contradiction itself is a signal the single page never gives. ## Freshness is what breaks the tie. None of the three sources here is lying, and each was accurate at some point; what separates them is how recently each was touched. An agent that can see a timestamp, or infer one from how the source behaves, updated within the hour versus synced overnight, has a real basis for preferring one over the other. An agent that only sees three flat claims with no sense of age has nothing to break the tie with except guessing, which is exactly the outcome the site's pillar piece argues an agent should never be forced into. This is why a timestamp, even an approximate one, is worth more than it looks. It turns three competing claims into one ranked list instead of three equally weighted guesses, and ranking is a problem an agent can solve. ## What an agent does with three answers instead of one. A well-built agent does not simply average three numbers or pick the middle one. It treats the disagreement as a reason to prefer the freshest source it can identify, and where none of the three carries a clear timestamp, it treats the fact as unconfirmed rather than picking one arbitrarily. That is a more cautious outcome than acting on any single page's confident claim, and it is the right kind of caution: a wrong "in stock" costs a business a cancelled order and an annoyed customer, while a brief "checking availability" costs almost nothing. ## The fix is one source of truth. The instinct to keep all three channels individually accurate, tightening the sync schedule on each, treats the symptom. The actual fix is structural: one stock figure that every channel reads from directly, rather than three separate copies that each need to be kept in step by hand. A brand that gets this right does not have a marketplace listing that is usually close enough, it has a marketplace listing that is never more than seconds behind the number that is true, because there is only ever one number. This is not a problem unique to skincare, or even to ecommerce, any business selling the same thing through more than one channel runs into the identical version of it, whether the channel is a marketplace, a wholesale partner, or a booking platform standing between a guest and a room. A hotel with the same room bookable through its own site, a wholesaler and a metasearch listing is running the identical risk on availability instead of stock, three calendars that are each honest about what they last knew and quietly wrong about what is true this minute. The harder version of the fix is organisational, not technical: someone has to decide which system is the one everything else defers to, and stop treating the other two as equally authoritative copies worth maintaining in parallel. That decision is usually more uncomfortable than the engineering, because it means admitting that two of the three channels were never really sources of truth to begin with, just places the same number happened to be displayed. One shared stock and pricing system, connected directly so every surface that checks in, an agent included, reads the same live number rather than a copy that drifted out of step, is what Selfe builds. ## Why would a business's own data contradict itself like this? Because different channels update on different schedules, one live, one synced overnight, one updated by hand, and each is honestly reporting what it last knew. ## What should an agent do when it finds contradicting facts? Prefer whichever source it can tell is most current, and where that is not clear, treat the fact as unconfirmed rather than picking one version arbitrarily. ## Act on this - [Check whether an agent can buy from you](https://selfe.ai/agentic-commerce/buyability-check) — the free readiness scan. - [Discover venues](https://selfe.ai/api/registry/discover) — `POST`, semantic browse across the registry. - [Match a bookable answer](https://selfe.ai/api/registry/match) — `POST` with dates and party size. - [Verify Selfe's identity](https://selfe.ai/.well-known/did.json) — `did:web:selfe.ai`. - [Agent card](https://selfe.ai/.well-known/agent-card.json) — how to connect over A2A or MCP.