Most of what surfaces in any search is noise. A name match that is the wrong person, a record that has been wrongly attributed, a coincidence of timing that suggests a pattern where none exists. A search that treats every finding with equal weight produces a long document of dubious value. The work is, in part, the work of separating signal from coincidence.

The first check is corroboration. A finding that has a single source is weaker than a finding with two independent ones. A claim that appears in one database and is contradicted by another is rarely a finding; it is, more often, an indication that one of the databases is wrong, and the work of determining which is its own undertaking.

The second check is plausibility. A purported finding that would, if true, be inconsistent with the rest of the picture is treated with suspicion. The world is not always self-consistent, but in the ordinary case, the records align. An outlier that contradicts the surrounding context is more often an error than a revelation.

The third check is replicability. A finding that can be confirmed by repeating the search through a different method or a different starting point is sturdier than one that depends on a particular sequence of queries. A finding that disappears when the search is rerun is rarely worth holding onto.

· · ·

The fourth check is provenance. A finding that comes from a source the desk knows and has worked with before is held differently from a finding from a source whose reliability is untested. The work of building confidence in sources is the slow part; it cannot be substituted for by speed.

The desk's output is shaped by these checks. The assessments are shorter than they would be without them, because the marginal findings have been filtered. The remaining content is more reliable. The client receives less material, and what arrives is what the desk is prepared to defend.