Every Usenet provider brags about retention. “5,000 days!” “6,000 days!” Bigger must be better, right?

Not necessarily. Retention is the most misused number in the entire industry, and it’s the one most likely to make you overpay. But there’s a second, quieter piece of the puzzle that almost nobody explains: PAR2 repair science. Retention is the window in which content exists; PAR2 is what actually rescues you when content inside that window is incomplete. This guide covers both — why the headline retention number misleads you, and why your downloader’s ability to repair a broken file often matters more than a couple of extra thousand days of storage.

What Retention Actually Is

Retention is simply how long a provider stores every article on its servers before deleting it. Measured in days. If a provider has 5,000 days of retention and an article is posted today, it’ll be purged roughly 5,000 days from now — assuming the provider doesn’t expand its spool in the meantime (they usually do, which is why these numbers climb year over year).

That’s the whole definition. Everything else in this guide is about where that definition gets fudged.

Wait — why isn’t retention “forever”?

Storage costs money. Usenet ingests an enormous volume of data every day — on the order of hundreds of terabytes per day across the network. Keeping all of that for 16 years is an expensive proposition. Providers trade off retention against how much hardware they want to pay for. That’s why different backbones offer different retention.

This is also the root of a specific UK angle: because you’re physically downloading gigabytes across undersea links and your ISP’s infrastructure, a deep-retention spool that’s hosted across the Atlantic is often slower in practice than a shallower one hosted closer to home — no matter how many days its marketing page quotes. More on that in the large-transfer stability section of the downloader guide.

The Crucial Distinction: Full vs “Up To” Retention

Here’s the trap. Providers quote retention as “up to X days.” That’s the maximum age of the oldest articles they’ll ever serve. It does not mean every article that age is actually there.

In practice, providers usually have two retention layers:

  • Full retention — the articles the provider is genuinely storing in full and can serve on demand.
  • Popular retention — older articles that are still listed in the headers but may be progressively thinned, reprocessed, or only partially available.

The headline “6,000 days” is almost always the maximum, and the complete-coverage number is meaningfully lower. This is a huge reason cheap providers can quote giant retention numbers while quietly having terrible gaps on older content. My honest take in the provider roundup explains how this shows up in practice.

Retention vs Completion: The Part Everyone Skips

You’ll hear retention and completion used interchangeably. They are not the same thing, and completion is the one that actually affects you.

  • Retention = how long articles are stored.
  • Completion = how complete a given binary is when you download it — i.e. what percentage of its articles your provider can actually serve.

A provider can advertise 6,000 days of retention and still hand you a 97% file on a popular post that’s three weeks old, because chunks of it were never propagated to their spool or got taken down. Retention tells you the window; completion tells you the reality inside that window.

For everyday use, completion matters far more than the retention number. A provider with 5,000 days and 99% completion beats one with 6,000 days and 96% completion, almost every time.

Why completion differs between backbones

Completion differences come down to:

  • Peering/uptake. How completely a backbone ingests the full global feed. Bigger backbones (Omicron) ingest a large share of the total feed, so they complete better across the board.
  • Takedown policies. Providers that aggressively honour removals lose articles faster than those that don’t, which hits completion on anything affected.
  • Propagation. With tens of terabytes a day in motion, glitches mean some articles simply never land on a given spool. More independent backbones in one account can catch what another misses — which is exactly the logic behind multi-backbone providers like NewsgroupDirect’s Grand Slam (NGD + Supernews + Usenet.Farm + ViperNews in one account). See the best providers comparison for why that’s worth money, and the Omicron vs UsenetExpress head-to-head for the two backbones behind most accounts.

The PAR2 Repair Science Section You Won’t Find Elsewhere

Here’s where retention guides usually stop — and where they should dig in. Because when a provider hands you a 97% file, the question isn’t “which backbone was it?” It’s “can my downloader repair the missing 3%?”

What PAR2 actually is

PAR2 (Parity Archive, volume 2) is an error-correction scheme layered on top of a release. Poster software (SRR, QuickPar, or the tools that build a release) generates parity files alongside the main binaries. Those parity files encode redundant data: enough that your downloader can reconstruct any missing or corrupt article from the parity alone.

Think of it like a crosswords grid. Every row and column has checksum clues (the parity). If one cell (article) is missing, you can still deduce what belongs in that cell from the other clues in its row and column — you don’t need the cell itself. PAR2 works the same way, but with far more redundancy at scale.

The physics of a repair

This is the bit that surprises people, and it’s directly relevant to retention:

  • The fraction of a post’s data you can repair from PAR2 is capped by the parity volume included in the release. Posters typically add 5–10% parity. So if a file is missing more than ~10% of its articles, PAR2 cannot save it — the repair will fail.
  • The further out of the retention window you pull, the more articles tend to be thinned or missing, eating into that parity headroom. This is the real, physical union of retention and repair: a deep-retention spool gives you the same 5–10% repair budget; it just means you’re more likely to be pulling from complete coverage in the first place.
  • PAR2 repair is CPU-bound and I/O-heavy. During a big repair, SABnzbd or NZBGet runs a Reed–Solomon decode across the whole file. On a low-RAM NAS or an underpowered Pi, this is exactly where a fast download stalls at the finish line. See the SABnzbd vs NZBGet guide for which downloader handles this science better. For the practical side — the exact settings and what to do when a repair fails — see the practical repair guide.

Why “just get more newsgroups” philosophies mislead

A common (and wrong) framing is that completion is purely a function of how much you can grab. In reality, repair is bounded by parity, not by retrieving more raw articles. If a release ships with 6% parity, no amount of extra fetching replaces articles that are missing network-wide from the spool — PAR2 can only reconstruct up to that 6% cap. This is precisely why paying for broad, high-completion backbones (which keep you inside working coverage) beats paying for a single deep-but-thinly-peered spool.

The practical rule

  • Recent releases with 5–10% parity rarely need repair at all — completion is near-total.
  • Older, less-popular content is where both retention gaps and missing articles stack up. Here your PAR2 budget is your lifeline, and a provider with better completion covers more of that budget than a provider with more headline days.

So when someone tells you “just buy the spool with the most days,” the PAR2-aware (and honest) answer is: buy the spool most likely to give you near-complete articles, so your limited parity budget is spent only on genuine stragglers — not on making up for systemic gaps.

How Retention Should Affect Your Buying Decision

How deep do you actually need to go?

Work backwards from your content age:

  • Recent releases: you need barely any retention. Even a cheap, modest-retention provider handles this fine, and PAR2 rarely even activates.
  • Content up to a few years old: ~2,000–4,000 days is comfortable.
  • Ten-year-old archives: you want maximum retention and top completion, because old, obscure content is exactly where thin spools fall apart first — and where your PAR2 budget can’t bail you out.

Overpaying for a giant retention number you’ll never use is the single most common budget mistake. If you mostly grab new releases, spend your money on completion and speed, not an extra 2,000 days of retention you’ll never touch.

Blocks are retention-safe

Because good block accounts never expire, a block from a deep-retention backbone is a cheap way to hedge: you buy access to the deep spool once, and it’s still valid when you eventually need an old article. My cheapest providers guide costs this out.

Does Retention Grow Forever?

No, and this is worth understanding. Retention climbs because providers keep expanding storage, not because articles magically persist. There are physical and financial limits, and the rate the network ingests data keeps growing. At some point providers may hold retention flat or even trim it on less-popular groups. This is manageable by most users, but “big number now” isn’t a guarantee for the future — which is another quiet argument for having access to more than one backbone.

A Note on the UK ISP Angle

Retention is a storage promise, not a speed promise. In the UK, whether a deep spool is actually usable is also a question of where that spool lives and how your ISP routes to it. UK ISPs have historically throttled Usenet traffic, and while modern SSL on port 563 defeats the throttling that was once baked into the network, routing quality to different backbones still varies by provider and by the time of day. A spool with better European peering can genuinely out-download a “deeper” spool routed badly in the UK. This is another reason the backbone (and its peering) matters more than the brand — covered in the best providers UK comparison.

The Bottom Line

  • Retention = how long articles are kept.
  • Completion = how complete the ones you actually download are.
  • PAR2 = the repair budget that rescues incomplete downloads — but it’s capped by the parity a release ships with (usually 5–10%), so it can’t fix systemic gaps.
  • The headline retention number is a maximum, not a guarantee of full coverage.
  • For most people, completion and a working PAR2 budget beat raw retention.
  • Buy retention depth only if you genuinely pull old content.

Further Reading