David was three weeks into his first large-scale document review as a junior associate at a national corporate law firm. The task was straightforward but relentless: review emails for responsiveness to a subpoena related to a client’s construction contracts. The review criteria were narrow, and the deadline was short. David’s supervising partner was known for his strictness. As he put it during the kickoff meeting, "Speed and accuracy are the focus. Don’t get bogged down. We’re not here to editorialize.” David took this to heart, wanting to make a good impression.

In a way, David enjoyed the exacting, if tedious, nature of the task, and he was proud of his efficiency. Late one evening, David was preparing to finish for the day when he encountered a short email exchange between two mid-level managers discussing “soil compaction rates” at a site that did not appear on the list of projects covered by the subpoena. Under the review guidelines, the emails were plainly outside scope. They involved a different project, a different entity, and no obvious contractual issues. David was about to mark the thread non-responsive and move on, but something caused him to hesitate. As he read a few emails in, the language struck him as unusually tense for what appeared to be routine technical correspondence. The managers seemed careful—almost evasive—in how they referred to the issue. There was nothing overtly incriminating, but the exchange felt incomplete, as though it assumed knowledge not reflected elsewhere. Against his better judgment, David set the emails aside and resolved to dig deeper.

Returning to work the next day, David felt drawn back to the strange thread. In the firm’s review software, he toggled on the Concept Map, a tool that visually grouped documents by topic rather than keyword. He noticed a small, isolated cluster of documents labeled "Sub-Surface Stability" that sat far away from the main Construction Contracts hub. When he hovered over the cluster, the system flagged a series of attachments that had been "near-duplicated"—the AI had identified them as almost identical to the project reports David was supposed to be reviewing, yet they carried a different project ID.

David suddenly realized the soil compaction emails weren't just a random tangent. Leveraging his software, he was able to track the hidden metadata and saw that the managers were systematically moving technical data about a geological fault line into a folder in a new domain for a different subsidiary that hadn't been disclosed in the initial discovery phase. The AI hadn't flagged it as responsive because the keywords were buried in technical jargon, but the data patterns almost certainly indicated a deliberate effort to keep these files out of the main production set. David wasn't sure if this was illegal or simply reckless, but the pieces pointed to a deliberate effort to hide a truth that could have a significant impact on the outcome of the case.

Raising the issue would complicate the review, require additional time, and potentially expand the scope of discovery—outcomes the client (and his supervising partner) had explicitly sought to avoid. Ignoring it, however, meant treating the document review as a mechanical exercise rather than a professional judgment call.

David saved his notes and returned to the queue, unsure whether what he had found warranted escalation—or whether he had simply strayed beyond the task he had been assigned.