Sixteen rows in the Ledger say we could not confirm the mechanism. This is each of them, with the one thing that would settle it.
UNVERIFIED in the Mechanism Ledger has a narrow meaning and it is not “we suspect this is false”. It means: we named the surface, we went to the primary documentation, and the documentation does not say what the claim requires it to say. Sometimes the value is real and Microsoft has never published a default for it. Sometimes the page that used to say it has been unpublished. Sometimes the only thing we could find was a user-authored forum thread, which is not documentation no matter which domain hosts it.
Each entry below carries three things: the question in a form that can be answered yes or no, what we checked, and the single piece of evidence that would close it. That third field is the point of this page. An open question with no stated closing condition is a shrug; an open question with one is a task.
If you hold one of these — a Microsoft page we missed, a capture that resolves, a document number — the address is the contact form. The row will be updated, the change will be dated at /mechanisms/changes/ with the old verdict kept beside the new one, and you will be credited in the body of the entry rather than in a footnote.
§1Why this page exists at all
Sixteen of sixty-eight rows is a lot of admitted ignorance to publish, and the commercial logic runs entirely the other way: uncertainty does not convert. A page that says “we could not establish this” sells nothing, ranks worse in the short run than a page that states an answer confidently, and gives a competitor an easy sentence to write about you.
It is published anyway for a reason that is not sentimental. The evidence base underneath this whole subject is far thinner than its reputation, and the thinness is the interesting finding. There is one fully verifiable independent measurement of consumer optimiser software in the public record, one more that is now unreadable at every address it ever had, and no academic or independent-lab study at all. Against that, a very large volume of confident assertion — including two fabricated statistics currently circulating from a machine-generated page, which we will not restate here because restating a fabrication is how it launders.
A reference built on that record either publishes its uncertainty or pretends to a certainty the record cannot support. There is no third option, and the second one is how everything else on this subject got written.
§2The sixteen Ledger rows, as questions
Ordered by Ledger identifier. Each links back to its row.
| Row | The question | What we checked | The single piece of evidence that would close it |
|---|---|---|---|
| M-03 | Does Microsoft document a default state for memory compression, and does turning it off change anything measurable? | The Enable-MMAgent and Get-MMAgent cmdlet references, whose entire description of the switch is circular. Memory Management Registry Keys. The Windows Server cache and memory manager pages, which cover page combining and never mention compression. RAMMap. There is no conceptual article about memory compression on learn.microsoft.com at all. | Any Microsoft page stating the default value of the MemoryCompression MMAgent setting on a client SKU. One sentence would do it. |
| M-07 | Does setting DisablePagingExecutive to 1 improve anything on a currently supported build? | The Windows Performance Toolkit reference, which sets it only to enable x64 stackwalking on Vista and Windows 7 and states Windows 8 and later do not need it. Memory Management Registry Keys, which does not list it. The Windows Server 2003 registry reference, which is the only page asserting a performance benefit and is twenty years out of scope. | A Microsoft page dated to Windows 10 or later that describes an effect of this value other than keeping kernel code resident for debugging. |
| M-08 | What does RAMMap’s Empty menu actually do, in Microsoft’s own words? | The RAMMap download page in full: it documents seven tabs, refresh and snapshot handling, and defers all terminology to Windows Internals. The Defrag Tools episode index, whose timeline lists an Empty-menu segment but supplies no written definition. The Windows Server cache and memory manager troubleshooting guide. Everything else Microsoft-hosted is a forum post. | Written Microsoft documentation of the Empty Standby List and Empty Working Sets commands — what they call, and what the memory manager does afterwards. |
| M-09 | What is the default value of EnablePrefetcher on Windows 11? | The Windows Embedded Standard 7 and Windows XP Embedded pages, which are the only Microsoft sources enumerating the 0/1/2/3 values, both archived and both written for write-filter devices. Memory Management Registry Keys. The Win32 memory reference. The commonly repeated “default is 3” has no Microsoft source we could find. | A Microsoft page giving the default for a desktop SKU. Without one, no tool can honestly claim to restore this value. |
| R-02 | Has anyone ever measured the effect of deleting orphaned registry entries, under a stated method, on a machine anyone can identify? | Microsoft’s unpublished registry-cleaner support policy, in its last readable archive capture. The current unwanted-software criteria. Registry Storage Space. The published record outside Microsoft contains one stopwatch test on three machines whose own conclusion was equivocal, and one 2011 comparison that is no longer readable anywhere — see §3. | One reproducible measurement: a named counter, a stated method, a machine specification and a date. Twenty-one years after the gap was first pointed out, it does not appear to exist. |
| R-03 | Can a scan result that reports a count and no per-item path be verified by anybody, including the vendor? | Microsoft’s criteria page, which classifies “misleading or inaccurate claims about files, registry entries, or other items” as grounds for an unwanted-software detection. We could not find a Microsoft page addressing the reporting format itself. The often-quoted line about optimisers needing to provide error details rather than a count is not on the criteria page; it is attributed to a 2018 Microsoft Security Blog post we have not read in full and therefore do not cite. | The 2018 Microsoft Security Blog post, readable, with the error-details language in it. That would turn a widely repeated paraphrase into a citation. |
| R-11 | Does Microsoft document the Max Cached Icons value, its type or its default? | learn.microsoft.com and support.microsoft.com searches for the value name. Every result was a user-authored Microsoft Q&A thread. The values people recommend — 2048, 4096 — trace to those threads and to third-party sites, not to a reference page. | A Microsoft reference page naming the value. Failing that, an authoritative statement that the value is obsolete, which would also settle it. |
| S-02 | Which services does a “recommended” optimisation list actually disable? | Microsoft’s service guidance for Windows Server 2016 with Desktop Experience, which is server-scoped by its own first paragraph and whose affirmative disable list is two Xbox services. The Windows IoT Enterprise guidance, scoped to fixed-function devices. We searched for a Windows 10 or 11 client equivalent and there is none. | The vendor’s list, published. This is preamble question 2 and it is answerable by any vendor in an afternoon; the fact that it generally is not answered is itself the finding. |
| S-12 | Is there any live Microsoft statement addressing the belief that Windows reserves 20% of network bandwidth? | The ADMX_QoS policy CSP, which is live and says the Packet Scheduler limits reservations to 80 percent — a ceiling on what programs may reserve, not a deduction, and not the number in the folklore. The archived Packet Scheduler pages, which discuss shaping and sequencing and give no percentage. KB 316666, which carried the sentence everyone quotes, returns 404. | A live Microsoft page, or a readable capture of KB 316666, containing the statement that all bandwidth is available unless a program explicitly requests a reservation. |
| B-04 | Does useplatformclock or disabledynamictick improve performance on any documented workload? | The BCDEdit /set reference, where both carry the identical note “This option should only be used for debugging.” The BCD boot options reference, which frames the whole element set as driver development and debugging. The BCDEdit command-line options page, which documents commands rather than elements. | Any Microsoft source describing a performance effect. As it stands the documentation points the other way, and a tweak guide recommending these is contradicting the only source that describes them. |
| B-05 | Where does Microsoft define BootTime, MainPathBootTime and BootPostBootTime? | The Windows ADK On/Off Transition Performance assessment, which defines Main Path Boot Duration and Post On/Off Duration but never mentions the event log or these field names. The WPT on/off transition recording page. The TechNet Wiki archive entry for event 100, which tabulates events 100 to 110 by cause and defines no timing field. The only Microsoft text relating them is an archived forum post carrying an explicit as-is disclaimer. | A Microsoft page defining the three fields in the Diagnostics-Performance event payload. Every boot-time report quoting them is quoting a field with no published definition. |
| D-06 | Did an AutoRun restriction update end the autorun.inf threat, and is KB 971029 readable anywhere? | The AutoPlay registry documentation, which covers NoDriveAutoRun and NoDriveTypeAutoRun and says nothing about a restriction update or about occupying the filename with a directory. Both live forms of the KB URL, numeric and GUID, which return 404. learn.microsoft.com searches. | A readable copy of KB 971029, or any current Microsoft page stating that AutoRun entries from non-optical media are no longer offered in AutoPlay. The claim is widely repeated and we could not source it. |
| D-10 | How large is reserved storage, in Microsoft’s own words? | The DISM storage reserve reference, which documents Get and Set and states no size. The consumer support page on how reserved storage works, which states no size. The Dism PowerShell module. The figure everyone quotes appears in press coverage and in a Microsoft-hosted blog post whose body we could not retrieve, and we will not cite a number we have not read on the page. | A Microsoft page stating the reserved size, or the formula it is derived from. Until then, “Windows is hiding N gigabytes” has no N. |
| A-05 | Does any consumer startup manager enumerate WMI permanent event subscriptions? | The CommandLineEventConsumer and event-subscription documentation, which establishes that a permanent consumer survives reboot and starts an arbitrary process as LocalSystem. The Autoruns documentation, which publishes coverage of this surface as the switch autorunsc -m. We found no published capability list from a consumer optimiser that names the root\subscription namespace. | One published capability list from a consumer product that includes it. Question 2 of the preamble, again, and the reason a “nothing starts automatically” verdict cannot currently be trusted from these tools. |
| A-10 | Does adding an entry to KnownDLLs change how quickly anything loads? | The dynamic-link library search order documentation, packaged and unpackaged, which places Known DLLs at step 5 and describes the system using its own copy. The Session Manager references. Nothing states or implies a preload, and nothing quantifies a saving. | A Microsoft statement about the timing effect of the list, as distinct from its position in the search order. The security consequence — changing which copy of a DLL every process resolves to — is documented; the performance one is not. |
| A-15 | Is the Drivers32 key documented as a subject anywhere? | The acmDriverAdd reference, which names the key incidentally while documenting an audio-compression API. The related multimedia reference pages. learn.microsoft.com searches for the key path. There is no documented layout, no enumeration of value-name prefixes, and no statement of when entries load. | A Microsoft page treating Drivers32 as its own topic. Autoruns enumerates it as the codecs category, which establishes that it matters and not what it contains. |
§3Three questions about the record itself
These are not Ledger rows, because they are not claim–surface pairs. They are questions about the evidence base and about this domain’s own history, and they are the three carried on the front page.
What grace-period mechanism did “use free Windows 7 for 360 days” refer to?
You can use windows7 offers up 360 days free from this feature.Archived vendor help topic, tlwinset.com/wineb/wh25.htm, capture 2013-09-24. Checked 2026-08-05 · Ledger row R-12
What is documented. Microsoft states the activation grace period is “typically 30 days” and that “in Windows 7 and Windows Vista, there is a limit to the number of times Windows can be rearmed. Typically, a system can be rearmed only 3 times.” The support article on the corresponding sysprep error puts it as a hard statement: “The Rearm process can be run a maximum of three times in a Windows image.” Thirty days plus three thirty-day resets is one hundred and twenty days. The advertised figure is three times that.
What we checked. The unattend SkipRearm reference; KB 929828; the Windows 7 volume activation planning guide, which confirms the thirty-day grace period and the KMS exemption but states no rearm count; the slmgr.vbs options reference, which documents /rearm and names the SkipRearm registry entry but states no limit. No Microsoft source documents a 360-day figure for Windows 7, and every third-party page asserting one traces to another third-party page.
What would close it. A Microsoft document describing a supported configuration in which a Windows 7 installation runs unactivated for more than one hundred and twenty days. We expect there is none.
What we will not publish either way. The archived help page for this feature offers a checkbox for having already used it more than three times, which takes the count from four to twelve, and that is where the figure comes from arithmetically. We decline to describe, test or explain any method of exceeding the documented rearm limit. That refusal is the whole of the UNSAFE-CLAIM verdict on R-12, and it is not a moral position, it is a scope decision: this publication explains documented mechanisms, and this is not one.
Is the 2011 test that everyone cites as the registry-cleaner debunking readable anywhere?
Why it matters. A great deal of the confident writing on this subject rests, when traced, on a single comparison published by Windows Secrets on 2011-11-10, which reportedly set the built-in Disk Cleanup against two commercial cleaners and found the built-in tool did the job as well. The best-known consumer article dismissing this category runs no measurements of its own and rests its entire empirical basis on that test. If the test is unreadable, the chain of citation terminates in nothing.
What we checked, 2026-08-05. The Wayback CDX index for windowssecrets.com returns no captures under a /2011/11/ path. The site’s Fred Langa author archive has hundreds of captured index pages across 2011 to 2016, which is what makes the absence of the article itself notable rather than merely unlucky. An earlier and different piece, /2007/03/01/01-Are-Registry-cleaners-worthwhile, does have captures returning 200 between 2007 and 2010, and returns 301 from mid-2011 onward — that is the site migration, and it is the point at which article URLs stopped resolving. The successor publication at askwoody.com returns HTTP 403 to our fetcher.
What would close it. A readable copy of the 2011-11-10 article at any address, or a print or PDF edition of that issue. Anything that lets a reader see the method, the machines and the numbers rather than a paraphrase of a paraphrase. Until then this site does not cite it as a measurement, and neither should anyone else.
Where did PCLinuxOS Magazine link this domain, and in what context?
The situation. A backlink index records pclosmag.com, a Linux community publication, linking tlwinset.com/why-is-windows-10-better-than-linux with the anchor text “Linux is better than Windows 10”. The slug argues one side and the anchor argues the other, which is interesting enough to want confirmed.
What we checked, 2026-08-05. We fetched pclosmag.com/links.html directly: HTTP 200, 15,269 bytes, and it contains zero occurrences of the string tlwinset and zero of winset. Site-scoped searches surface nothing. The magazine publishes as per-issue HTML and PDF, so a citation inside an article body would be hard to surface either way, which means the negative result is weak evidence rather than a refutation.
What we did establish. The page the link pointed at is recoverable from the archive across 2019 to 2025: a listicle titled 10 Reasons Why Windows 10 Is Better Than Linux?, bylined James Phelps, published under branding that is not Tenglnet’s. It asserts that Linux’s voice assistant is called “Betty”. That is not a fact about Linux, and it makes the most economical reading of the contradiction — a Linux magazine citing a bad pro-Windows post in order to argue against it — considerably more plausible. It remains a reading.
What would close it. The issue number and page of the PCLinuxOS Magazine article containing the link, or a capture of the linking page. The link’s existence in our records comes from a commercial backlink index, not from our own inspection, and we say so on every page where it matters.
§4What happens when one of these closes
The row is edited, the verdict moves from UNVERIFIED to whichever of the other seven codes the evidence supports, and the entry leaves this page. The edit is appended to /mechanisms/changes/ with the date, the old verdict, the new verdict and the source that moved it. Nothing is silently revised and nothing is removed, so the record of having been wrong stays available to anyone who wants to check whether this publication is honest about it.
Send evidence to the contact form. A primary source is worth more than an explanation, a document number is worth more than a link that may rot, and “you are wrong and here is the page” is the single most useful message this address can receive.