We Ran a 6,000-Page File Through Guidon. Here Is Exactly What Happened.

Guidon Insights · September 2026 · A capacity benchmark, published with its methodology and its failures — because that's how we'd want a vendor to answer us.

In a meeting this month, the head of a veterans service organization told us something we already half-knew: the files are getting enormous. His officers regularly see records that run six thousand pages. He asked, reasonably, whether our software could handle that. We told him we don't answer capacity questions with promises — we'd run the test and send him the number. This is the number.

The result

What	Measured
Document	6,000-page PDF (synthetic veteran record, ~11.6 MB of text)
Reading the PDF into text	27 seconds — all 6,000 pages
Conditions in the record	24 seeded · 24 found — nothing missed, nothing invented
Full analysis, end to end	1 hour 42 minutes, unattended
Output	Every condition mapped to the rating schedule with candidate percentages, evidence gaps flagged, service-connection pathways assessed, prioritized work plan — all drafts for an officer's review
Human attention required during the run	None. Upload it, go serve veterans, come back.

For scale: an officer who reads two pages a minute would need fifty hours to read this file once. The point of the software is not that it is fast — it's that the officer's fifty hours go to judgment, veterans, and claims instead of page-turning.

Methodology, because a benchmark without one is marketing: we generated a synthetic 6,000-page medical record — varied progress notes, labs, C&P excerpts, and consult notes across 24 distinct conditions, no two pages identical, clearly labeled synthetic throughout. No real veteran's data is ever used for testing. The run used the same production system our pilot offices use, the same day they were using it. One test, reported as measured. Larger files mainly mean proportionally longer runs: the system reads in fixed-size chunks (this file took 119) and picks up where it left off if anything interrupts it.

The part where it broke

Here's the part a brochure would leave out. Six days before this test, an officer in one of our pilots uploaded a 2,900-page record — the biggest real file we'd ever seen — and the analysis failed at 4:58 PM. The cause wasn't the file; it was a daily processing-capacity ceiling at our cloud provider that we, a young company, hadn't raised yet. Two record-breaking files in one afternoon hit it.

Our monitoring emailed us the failure automatically. By the next morning, two fixes had shipped: the software now waits out capacity limits instead of dying (the officer's re-run rode through 115 of them without losing a page), and analysis now fails over across multiple secured data-center regions — all inside the same HIPAA-eligible, Business-Associate-Agreement-covered environment — so one region's ceiling can't kill a veteran's case. His analysis resumed from its checkpoints, nothing re-uploaded, report ready at 9:21 AM.

During the 6,000-page test, that failover engaged for real — the run rotated across three regions under live capacity pressure and finished without a hiccup. The failure last week is the reason this week's test couldn't fail the same way.

Why we publish this: every vendor demo works. The questions that matter are what happens on the biggest file, on the worst day, and whether the vendor tells you the truth about both. Our accuracy page publishes our misses; this page publishes our outage. If a vendor has never told you about a failure, it isn't because they haven't had one.

What this means for an office

If your officers see 500-page files, analyses come back in about 15 minutes. A 1,000–3,000-page record is a coffee-and-casework wait. A 6,000-page monster is: upload it, work your morning appointments, and read the finished work plan after lunch. In every case the officer — not the software — decides what's pursued, what's signed, and what's filed. The output is a draft until a service officer says otherwise.

The record's size should determine how long the software works — never whether the veteran gets helped.

Questions about a specific file size or format, or want to see it on your own office's kind of caseload? The pilot is free, and honest numbers are the deliverable.