June 24, 2025

Context Windows: Why 'It Can Read the Whole Filing' Is Not the Whole Story

Vendors advertise how many pages a model can hold in a single request. That number answers a real question, but not the one that determines whether the output is trustworthy.

What a large context window solves

Being able to pass an entire 10-K, or several years of filings at once, into a single request means a model can compare figures across periods without a person manually assembling excerpts first. That is a genuine, practical improvement over the earlier pattern of feeding a model small chunks and hoping nothing important got left out.

What it does not solve

Taking in more text does not guarantee the model weighs every part of it equally. Details buried in the middle of a long document are more likely to be missed or under-weighted than details near the beginning or end. A model that technically 'read' a footnote on page 140 may still fail to mention it when summarizing.

The practical fix

Do not rely on a single pass over a long document for anything that matters. Ask a targeted follow-up about the specific sections that carry the most business-development signal, auditor changes, related-party transactions, material weaknesses, rather than trusting a general summary to have caught them on its own.

A practical habit for long documents

When a filing runs past fifty or sixty pages, do not trust a single request asking for 'a summary of anything relevant.' Break the request into sections that map to where BD-relevant signal typically lives, risk factors, controls and procedures, related-party transactions, subsequent events, and ask about each specifically. This costs a few extra minutes and produces a meaningfully more complete result than a single broad request over the whole document.

This habit matters more as filings get longer, which they generally do over a company's life as more disclosure requirements accumulate, making the single-pass approach progressively less reliable over time.

A note on how this affects turnaround time

Breaking a long-document review into targeted sections adds a small amount of time compared with a single broad request, typically a few extra minutes rather than a few extra hours. Firms sometimes assume thoroughness and speed are in sharp tension here; in practice, the added time is modest relative to the value of not missing a footnote that a general summary would have skipped.

Where this leaves a firm

None of this is complicated in principle, which is exactly why it gets skipped under deadline pressure. The question worth returning to before treating matching the right model to the right task as settled is what a careful reader would actually notice if the firm got it right. On the point raised above under “what a large context window solves,” the answer is usually specific rather than clever: a large context window lets a model compare data across an entire filing at once. Firms that build this expectation into how they train new associates find it easier to sustain once experienced staff move on, because the standard lives in a documented habit rather than in one person's memory. The gap between a firm that talks about matching the right model to the right task and a firm that actually practices it shows up over several quarters, not in any single engagement, and it tends to show up most clearly in the small, unglamorous checks that a client never sees directly but benefits from anyway.

It also helps to name, plainly, who is responsible for keeping this working once the novelty of a new tool wears off. Someone should own the point raised under “what it does not solve,” check it periodically rather than assume it stays true on its own, and be the person a colleague asks when a new situation does not fit the pattern described here. Put simply: ask targeted follow-up questions about high-signal sections rather than trusting one pass. That kind of ownership, named and specific, is a small addition to a firm's process, and it is usually the difference between a good idea that is followed for a month and a standard that actually holds up over a year of real client work.

None of this needs to be elaborate to be effective. A short, dated note in a shared file, reviewed at the next quarterly check-in, is usually enough to keep the responsibility from quietly disappearing when the person who first cared about it moves on to something else.

VIPMarketing Editorial