This piece walks through what actually happens between the moment someone asks Verety to check a counterparty and the moment a verdict comes back — not as a feature list, but as the sequence of decisions the system makes along the way, and why each step exists.
Step 1: Resolving the entity
A request to Verety typically starts as a name, sometimes alongside a registration number or a website. The first job isn't investigation — it's resolution: confirming which real-world entity is actually meant. A common name can resolve to several real companies across jurisdictions, and getting this step wrong undermines everything after it. Verety's resolution step checks registries for exact and close matches, weighs supporting signals (a provided website, a jurisdiction hint, an industry), and either resolves confidently to one entity or surfaces the ambiguity rather than guessing silently.
Step 2: Dispatching specialist agents
Once an entity is resolved, the system doesn't run one generic check — it dispatches a set of specialist agents, each responsible for one kind of evidence: a registry agent pulling the official corporate record (status, officers, registered address, filing history), a sanctions agent screening the entity and its known principals against relevant lists, a media agent searching for adverse news coverage, and a domain agent examining the entity's own web presence for consistency with what the registry says. Each of these runs largely in parallel and reports back independently — the point of specialization is that each agent is built to be good at one kind of source, not a generalist skimming everything shallowly.
Step 3: Building the graph, not just the record
This is the step that separates an investigation from a lookup. As each specialist agent reports findings, the system builds a graph of the entity and everything connected to it — other companies sharing a director, a registered address shared with entities that have their own flags, a beneficial owner who appears, one or more hops away, behind other parties in the request. The graph isn't a side feature; it's where the organized-fraud and hidden-risk patterns that a single-entity check would miss actually become visible, consistent with the scale/connections/outliers framework described in our writing on organized fraud detection.
Step 4: Cross-checking, not just collecting
Raw findings from five sources aren't a conclusion — they're inputs that need to agree with each other, or need their disagreement understood. If the registry lists a different address than the one on the entity's own website, is that a recent move the registry hasn't caught up with, or a sign the entity isn't where it claims to be? If adverse media references a company with a similar but not identical name, is that the same entity under different branding, or a genuine false match? This cross-checking step is where the investigator LLM does real reasoning work — reconciling sources, resolving ambiguity where it reasonably can, and flagging it clearly where it can't.
Step 5: Scoring — and showing the work
The output isn't a single number pulled from a black box. Verety separates risk (what could go wrong — sanctions exposure, adverse findings, structural red flags in the network) from compliance posture (is this entity in good standing, properly registered, appropriately documented) because conflating the two hides information a real compliance decision needs. Every score traces back to the specific findings that produced it, with their sources — the point of an auditable system is that a compliance team (or a regulator) can see exactly why a verdict came out the way it did, not just trust that it did.
Step 6: Being honest about what wasn't checked
Not every jurisdiction has the same source coverage, and not every check succeeds. When a source is unavailable, thin, or inconclusive for a given entity, that gap is reported explicitly as part of the result — not silently treated as a clean pass. A confident wrong answer is worse than an honest "we couldn't fully verify this" one, and the whole value of an auditable system collapses if its gaps are invisible.
Step 7: Staying current after the first verdict
An investigation doesn't end when the first report is delivered. For customers using continuous monitoring, the same specialist agents keep watching the entity and its network going forward — a new sanctions listing, a director change, a newly discovered connection to a flagged entity — and surface what changed rather than requiring the next check to start from zero, closing the gap between "something changed" and "someone noticed" that a point-in-time check structurally can't close.
Why this sequence, not a simpler one
It would be faster to skip straight from "resolve the entity" to "return a score" using one or two sources. The reason Verety doesn't is the same reason a skilled human analyst doesn't: the useful signal is disproportionately in the parts that take more work — the network, the cross-checks, the honest handling of what's uncertain. A system built to skip those steps would be faster and worse at exactly the cases where getting it right matters most.