Everyone advertises 6,500 days now
Retention is the number every provider leads with, and it has quietly become the least useful spec in the market. It measures the age of the oldest article, not how much of the old feed survived — and it is a property of the backbone, not the brand you bought it from.
Newshosting says 6,580+ days. Eweka says 6,591+ days. Round the field off and everyone worth considering is somewhere past 6,500 — roughly eighteen years, back to about 2008, when the big archives stopped deleting things.
When a headline spec is identical across an entire market, it has stopped being a spec. It’s wallpaper. And retention has a specific problem beyond being undifferentiated: it doesn’t measure the thing people think it measures.
It measures age, not density
A retention claim is a statement about the oldest article a provider will serve. It says nothing about what fraction of any given week from 2014 is still intact.
The industry has a name for that gap, and it’s worth knowing because almost nobody prints it next to the headline number:
- Full retention — the backbone aims to hold the complete article feed across the advertised window.
- Popular retention — selective. Frequently requested articles survive to the advertised depth; the long tail thins out or was never kept.
Both get marketed as one number with the word “days” after it. The difference between them is the entire difference between a download that completes and one that doesn’t, and it’s invisible from the pricing page.
There’s a second split hiding in the same figure: text and binary retention are two different numbers. Text groups are trivially cheap to keep forever. Where a provider quotes a single value, check that the binary figure matches it rather than assuming the larger number applies to both.
It’s a backbone property, not a brand property
This is the part that makes retention useless as a comparison tool. Retention lives at the backbone. When a backbone extends its archive, every reseller sitting on it publishes the same new number within days.
So comparing the retention of two brands is only meaningful if they’re on different backbones — and after the consolidation covered in who owns whom, there are only about six real answers in the market. Pick two brands off the same feed and you are comparing a number to itself. Worse, you may be paying twice for it, in the belief that you’ve bought redundancy.
Depth isn’t always storage
The advertised window doesn’t have to be one continuous thing a provider owns. Some of it is served from the provider’s own spool; some of it comes from peering and backfill arrangements further down.
That matters for two reasons. The deep end is often slower and thinner than the recent end. And, as covered in DMCA vs NTD, it can sit under a different takedown regime than the front end — UsenetExpress runs DMCA on its US servers and NTD on its NL backfill; Usenet.Farm and ViperNews run NTD up front with DMCA deeper down. One retention number, two products, stitched together at a seam. The seam is exactly where completion tends to drop.
Old content isn’t the same as old articles
Archives also grow: the daily feed gets added while old articles stay, so the maximum figure creeps up over time on its own. Some providers describe this as growing or spooling retention.
Separately — and this is the part that quietly rescues most people’s downloads — popular old content gets reposted. A 2012 release you can still grab today is frequently a 2024 repost, sometimes obfuscated, carrying new message-IDs and a new date. Your success on old material therefore depends partly on the reposting community and not only on your provider’s spool.
Which cuts both ways. It means deep retention matters less than the number implies for anything popular. It also means deep retention matters more than the number implies for anything obscure, because nothing obscure ever gets reposted. Retention is doing its real work precisely on the material nobody is asking about — which is also the material a popular-retention backbone was most likely to drop.
What to ask instead
Start with the age you actually download at. For most setups, the overwhelming majority of grabs are days to months old. In that window every provider on every backbone is identical and retention is entirely irrelevant to your success rate. If that’s you, buying on retention is buying nothing.
If you genuinely pull decade-old material, treat retention as a ceiling and takedowns as the binding constraint. Two backbones at 4,000 days will beat one at 6,500, because the failures aren’t age-related — they’re notice-related, and two regimes were scrubbed by different notices. The block-account approach exists for exactly this: a few euros of fill on a second backbone buys more old-content completion than any retention upgrade on your primary.
Test instead of trusting. Take a handful of NZBs at the age you care about, run them against a cheap block on a candidate backbone, and look at what actually completes. That’s an afternoon and about five euros, and it tells you something the spec sheet structurally cannot.
The provider finder weights it this way for the same reason: the question isn’t how deep the archive goes, it’s whether the parts you want are still in it.
Retention is the one number a provider can raise by doing nothing at all — just keep not deleting. That’s why everyone’s is enormous and why they all move together. The number that would cost real money to improve, completion on the long tail, is the one nobody prints.
Figures cited are the providers’ own advertised claims as of September 2026 and move upward over time; treat them as a snapshot, not a ranking.