LTO’s published durability is “17 nines.” Real-world tape lands closer to four — and the people who build the format have said so, in public.

Open almost any LTO marketing page and you’ll meet one of the most impressive numbers in all of storage. The LTO program’s own website states that LTO-9 delivers “at least 17 nines of durability.” [1] IBM, in its long-running Busting Four Myths About Tape, dispatches the myth that tape is less reliable than disk by pointing to tape’s bit error rate — by IBM’s math, ten times more reliable than enterprise SAS disk and a hundred times more reliable than enterprise SATA. [2] The message is consistent and, on its own terms, accurate: by the numbers printed on the box, tape is the most reliable storage medium money can buy.

But those numbers are not the reliability you actually get, and you do not have to take a skeptic’s word for it. You can take the word of the people who design the drives.

First, what a “nine” is

Durability in storage is often expressed in “nines” — the probability that a given piece of data survives. Four nines (99.99%) means roughly one failure in ten thousand. Eleven nines — the figure cloud object stores like Amazon S3 made famous — means one in a hundred billion. Each additional nine is a tenfold improvement, so the distance between four nines and seventeen isn’t a rounding difference. It’s thirteen orders of magnitude.

When LTO says “17 nines,” that figure is derived from a single underlying spec: the uncorrectable bit error rate, or UBER. For LTO-9 and LTO-10, UBER is rated at better than one uncorrectable error in 10^20 bits — astonishingly low, and the engineering behind it is genuinely world-class. The catch is in what UBER does and doesn’t measure.

UBER is a calculation, not a measurement — and it assumes the one thing reality breaks

UBER describes the quality of the recording technology under a specific set of assumptions: that errors are random and uncorrelated, on a properly stored cartridge, read back under archival conditions. [4] It’s a statement about the magnetic recording channel — head, media, signal processing, and the drive’s built-in C1/C2 error correction — under laboratory idealizations. It is not a statement about what happens to tapes in a working data center.

That built-in correction is formidable. LTO uses a two-dimensional Reed-Solomon scheme — C1 across each track, C2 across channels — that can absorb burst errors, random errors, and even the total loss of several channels and still hand back clean data. It’s why the UBER number is so small. But all of it is computed against that founding assumption: that errors arrive independently, sprinkled at random across the medium. And random is exactly what real-world tape errors are not.

The number the format’s own engineers report from the field

In September 2025, Turguy Goker — Quantum’s advanced-development lead for LTO and the chair of the INSIC tape-technology roadmap — presented field data at a public IEEE technical talk. [3] This was not a simulation. The dataset: more than 500 petabytes written, 350 petabytes read, roughly 190,000 tape loads, across 200-plus LTO-9 drives and some 20,000 cartridges.

The conclusion, stated plainly on the slide: “operational reliability typically settles at ~4 nines per cartridge based on real-world non-random errors.” And, set directly against the spec: “UBER is 17 nines, so real world is much worse than UBER spec because operational errors are not random.”

The 17-nines figure on the marketing page and the ~4-nines figure from the field are not in conflict because one of them is false. They measure two different things, and the company that helped write the spec said as much, in public, to a room of magnetics engineers.

Why the gap is thirteen orders of magnitude wide

The same presentation is specific about why. Tape errors are rare — on the order of 0.01% or less — but they’re positionally correlated. They cluster: at the beginning and end of the tape, and at specific wraps where the medium has seen more handling stress, tension variation, and exposure to contamination. When a drive hits one, it retries — backing the tape up and re-reading, sometimes repeatedly over the same stretch. Those repetitive retries can accelerate wear in exactly the spot that was already marginal: a small feedback loop that can nudge a correlated soft error toward an uncorrectable one.

None of that is captured by a model that assumes each bit fails independently of its neighbors. Which is why, when the same field data is re-normalized, the answer swings wildly with how much you assume you lose per error: score it as the loss of a single small dataset and you get eight or nine nines; score it as the loss of a full cartridge and you get one to three. Average across real outcomes and it lands near four. The lab number and the field number were never describing the same event.

Even the nines you do see in marketing aren’t always the same nines

One trap to watch for: when a tape-library vendor advertises “five nines,” check what it is actually counting. Spectra Logic, for example, markets five-nines figures on hardware — but that’s five nines of availability [5], the system’s uptime, not the durability of your data. Availability, durability, and UBER are three different measurements that all get rendered in the same “nines” shorthand, and the brochures rarely tell you which one you’re looking at. That isn’t a tape failing; it’s a marketing-language failing — but it’s yours to untangle when you’re sizing a data-protection strategy.

A fair word about who’s telling you this

It’s worth being clear about why Quantum, of all companies, is the one putting the 4-nines figure on a slide. Quantum sells the fix: an erasure-coding system — its ActiveScale Cold Storage, built on a patented two-dimensional scheme — that it markets at better than eleven nines of durability. [6] The reliability gap this article describes is exactly the problem that product is built to solve, and the presentation makes that case directly. Far from undercutting the number, that context reinforces it. Consider the incentives: the consortium that markets “17 nines” has every reason to make tape’s reliability look as high as possible; a vendor selling a durability upgrade has every reason to make the baseline look honestly low — and no reason to understate it. When the side inclined to oversell and the side inclined to flag the gap both arrive at “operational errors are not random,” you’re about as close to a triangulated answer as this industry’s self-reported data gets. None of which is a knock on Quantum. In an ecosystem where nearly all the field telemetry lives inside the companies that sell the hardware, even the honest numbers tend to come wrapped in a sales motive — which is precisely why an independent read is worth having.

What this means if you actually run tape

First, the reassuring part, because it’s true and it matters: four nines per cartridge is genuinely good. On the honest, field-adjusted numbers, tape remains more reliable — and far cheaper per terabyte — than the alternatives for long-term archive, with power and cooling advantages disk simply can’t match. The point of this article is not that tape is unreliable. It plainly isn’t.

The point is that you should plan with the operational number, not the spec number. For a great deal of data, four nines plus a second copy is entirely sufficient. But for the data now driving archive growth — irreplaceable scientific and AI training sets, compliance records, national archives — four nines may not be enough. And the distance between four nines and the eleven-plus nines those workloads demand cannot be closed by the drive’s built-in error correction, because correlated, positional, wear-driven failures aren’t the kind of error that correction was modeled to catch.

Closing that gap takes protection above the drive. Three moves matter. Real verification rather than assumed verification — the read-while-write check that happens as data is first written confirms the write looks good in that instant, on that head; it is not the same as reading the tape back later, ideally on a different drive, to confirm the data is genuinely recoverable and its fixity intact. A second copy, the discipline tape shops have always practiced. And erasure coding that spreads protection both across tapes and within each tape — the modern path to eleven nines and beyond, and a subject worth its own article.

Why we think you should hear this

The most striking thing about the 4-nines number is not the figure itself, but where it came from. This was not surfaced by a tape skeptic or a disk vendor with an axe to grind; it was presented by Turguy Goker — the chair of the INSIC tape roadmap and a senior engineer at Quantum, one of the three companies that make the format — from his own company’s field telemetry, at a public engineering conference. The information is out there. It just doesn’t make it onto the marketing page next to “17 nines.”

We started the LTO Show because the people who buy, run, and depend on tape deserve the same picture the engineers have — not a worse one, and not a glossier one. Seventeen nines is a real measure of how good the recording technology is. Four nines is a real measure of how that technology behaves in your data center. You need both to make a sound decision. Now you have them.

Sources

1. LTO Program, What Makes LTO Technology So Darn Reliable? lto.org, August 2022. https://www.lto.org/2022/08/what-makes-lto-technology-so-darn-reliable/
2. IBM, Busting Four Myths About Tape. https://www.ibm.com/support/pages/busting-four-myths-about-tape
3. Turguy Goker (Quantum; chair, INSIC tape-technology roadmap), Tape Roadmap and Challenges with using new High Areal Density Tapes, IEEE technical presentation, September 12, 2025. https://events.vtools.ieee.org/m/491103
4. INSIC, Global Trends, Applications and Use Cases for Tape Adoption (2024 Tape Technology Roadmap). https://insic.org/roadmap/
5. Spectra Logic OSW-2400 (“five-nines availability”), StorageNewsletter, January 2025. https://www.storagenewsletter.com/2025/01/29/spectra-logic-transforms-tape-connectivity-with-1u-osw-2400-optical-sas-switch/
6. Quantum Corp., U.S. Patent 11,216,196 B2, Erasure coding magnetic tapes for minimum latency and adaptive parity protection feedback. https://patents.google.com/patent/US11216196B2/en

Pete Paisley is the host of the LTO Show, the premier podcast for leaders in the LTO tape storage hardware community. Reach out with story ideas or comments at pete@ltoshow.com. Copyright 2026 The LTO Show and Pete Paisley. Linear Tape-Open, LTO, the LTO logo, Ultrium and the Ultrium logo are registered trademarks of Hewlett Packard Enterprise, IBM and Quantum. All product and company names are trademarks of their respective holders.