Editorial Policy

A policy page usually defends a publication. This one describes a discipline, and it is the same discipline the bench articles follow: name the mechanism, name the source, and stop writing where the evidence stops.

What the desk prints, and what it refuses

The desk publishes mechanisms, readings and records a reader can check somewhere else. It does not publish rankings, scores, prices or brand recommendations, and it does not publish copyable settings: no voltage, no target frequency, no firmware value, no temperature threshold written as an instruction. A number that can be carried to another machine is a number the desk cannot stand behind, because the desk has never seen that machine.

Two habits follow. A claim and its limit travel in the same sentence. And where a figure is missing, the page says so instead of supplying one from habit. The table is the short version of both rules.

Why does a claim carry its ceiling in the same sentence?

Because a bench result describes one machine on one afternoon. A memory test that finished clean says that those patterns, on those modules, at that speed, at that temperature, in that slot order, produced no detected error during that run, and nothing more. Remove the conditions and the sentence starts certifying, which is the one promise the desk never makes. The bound cuts the other way too: a drive that reports nothing wrong is reporting what its firmware chose to count, and the desk reads that table as a vendor's account, not a verdict.

Where does the Overclockix record actually live?

Not here. This desk is not the Overclockix project and does not maintain, distribute or speak for it. The record belongs to other people: a first release, 3.2 on a Knoppix base, published by Arkaine23 in 2003; a project quiet after 2005; a takeover by mbentley in 2011 on a Debian base; a last published version, 7.8.0, dated June 1, 2015 according to DistroWatch. Eight releases stand in that record, on a DistroWatch page dated February 8, 2026 and consulted here on September 5, 2026.

The source sits in a public GitHub repository, and this site serves zero disk images of any of it: no ISO, no checksum, no mirror. The figures live in the release record.

Citing a toolkit the desk does not ship

The same restraint decides how another project gets cited. When the live systems pages describe a toolkit, the desk does not republish its package list, its mirrors or its checksums; it names the page that maintains those things. SystemRescue, the rescue toolkit once called SystemRescueCd, is the working example: its own site documents the bootable medium, the tools shipped with it, and guides for disk partitioning and LVM, under GPLv3 with sources on GitLab. Memtest sits in that list, and memory is the most blamed component on this desk, which is why the reader should take the list from the project rather than from a summary here. The desk sends readers to the project's homepage for that, and keeps its own pages on which toolkits still get maintained to purpose and blind spots.

How does a correction reach the page?

As a change to the sentence that was wrong, not as a note of regret under it. Every figure on the desk already names where it came from, so a correction has something concrete to name: which source, which sentence, which reading. What the desk refuses is the quiet edit, a number swapped while the old one keeps circulating in a reader's notes. If a claim here seems to outrun its source, read the sentence against that source first, then write to the desk with the line in question. The same route works for a figure the desk failed to bound, which on any bench is the likelier error.