Working Set tlwinset.com

How Windows actually works — the registry, prefetch, services, the memory manager — explained from documentation, so you can judge a speed-up claim yourself. We have not run the software we write about, and we say which parts we could not verify.

Subject Why this site carries no screenshots, and what replaces them Covers Image policy · fabricated UI · inline SVG diagrams We ran it No. We hold no copy and distribute none. Sourcing Our own published policy · site build constraints Ledger rows Open questions 0 Published 2026-08-05   Last verified 2026-08-05

Why are there no screenshots on this site?

There are none, there will be none, and the reason is not aesthetic. A site that writes about software it has never run cannot honestly publish a picture of that software running. Every route to such a picture is either impossible for us, closed to us, or a lie — so the policy is absolute rather than case-by-case, which is the only form of the rule that cannot be eroded one exception at a time.

Tenglnet published its Windows Winset utilities at tlwinset.com from roughly 2005 to 2016. The registration lapsed, the address passed through other owners, and it came to this publication in 2026. Working Set has no connection to Tenglnet, holds no copy of its software, distributes none, and has never run any of it.


§1Three reasons, and none of them is taste

Costs to produce one
a copy we do not have, of software we will not obtain
Costs to republish one
the previous owner’s material, which we never reproduce
Costs to fabricate one
the only claim this site actually makes

One: we cannot make our own. A screenshot of a program is a photograph of that program running on a machine. Windows Winset is not obtainable from any source we would trust, we hold no copy, and we will not go looking for a fifteen-year-old system-level installer of uncertain provenance in order to illustrate an article. Nor would we redistribute it if we found it. The about page states this in the same words the header block on every page states it: we have not run this software, we do not possess it, and we distribute none of it.

Two: we may not republish the previous owner’s. This domain carries an archive of captures made when Tenglnet published here, and some of those captures contain product imagery. Republishing them is exactly the thing an expired-domain rebuild must not do — it takes the previous publisher’s material and re-serves it from the previous publisher’s address, which reads as continuity to every reader and every crawler that sees it. We republish no page of Tenglnet’s site and no continuous passage of its prose, and by the same rule we reproduce none of its pictures. Where an archived vendor claim appears at all it is capped at 25 words, sits in a bordered card, and carries the archive URL and the capture date. Individual sentences are quoted, under that cap, because a claim has to be on the page to be examined; the two rules differ because a quotation has a word count and a screenshot has no excerpt, so the picture rule has nowhere to stop except at zero.

Three: a fabricated screenshot would be a misrepresentation, not a stylistic choice. This is the one that decides the policy. An image model asked for anything near this subject will produce plausible fake user interface — a title bar, a scan progress dialog, a list of found “errors”, a number. Published on a page about that product, at that product’s former address, such an image asserts something: that the software looked like this, that it reported this, that somebody here saw it. All three are false. There is no caption that repairs it, because a reader takes the number from the picture and the caveat from nowhere. An illustration that makes a factual claim we cannot support is a false statement drawn rather than written, and it fails for the same reason the sentence would.

The same reasoning rules out the softer versions: no mock dialog, no “illustrative” recreation, no vendor marketing render, no window chrome, no title bar, no menu, no cursor, no progress bar, no gauge, no speedometer, and nothing that could be mistaken at thumbnail size for a screen capture. Removing the possibility removes the failure mode; keeping it and promising to be careful does not.


§2What replaces a screenshot here

Costs per diagram
≤ 2 KB inline, no image request
Costs to the reader
0 fonts, 0 scripts, 0 third-party requests
Costs to correct one
an edit to text, not a redraw

Diagrams. Hand-authored, inline <svg>, two inks, real selectable <text>, under two kilobytes each, with every longer label typeset in the HTML <figcaption> beneath. They recolour for dark mode from the page’s own custom properties, they stay sharp at every pixel density, they can be selected, copied, translated and read aloud by a screen reader, and they cost no network request at all.

One list box labelled “startup items” on the left of a dashed boundary; on the right, the eight separate Windows surfaces such a list is reading from. one list of “startup items” HKLM & HKCU Run, RunOnce the Startup folder services (Start = 2, 3) drivers (Start = 0, 1) scheduled tasks trigger-start registrations Winlogon, BootExecute WMI event subscriptions
The boundary is what one window showed ↔ what the operating system exposes. Left: a single list control, the thing a screenshot of a startup manager would have captured. Right: eight of the autostart surfaces such a list is reading from, each one a separate registry key, file location or registration with its own default, its own version boundary and its own cost of being switched off. The screenshot shows one of these columns. The diagram shows the other. Full enumeration in Sixteen places Windows can start a program.

§3A diagram of the surface list is more useful than the screenshot ever was

Costs in information
0 — the screenshot never carried the surface names
Costs in verifiability
−1 capture date, +1 checkable path per row
Costs in shelf life
a diagram survives the next release; a capture does not

This is the part that is easy to concede too quickly, so here is the argument rather than the apology. Ask what a utility screenshot actually contains. Take the commonest one in this entire category: a window with a list, a checkbox per row, a program name, a publisher string, and a button. What has a reader learned from looking at it?

They have learned what the vendor’s window looked like. They have not learned where any row came from. The list conflates a Run key value, a shortcut in the Startup folder, a service with Start = 2, a scheduled task with a logon trigger and a driver, and prints them in one column with one checkbox each — which is precisely the flattening that makes the tool feel authoritative and makes its output impossible to check. The screenshot faithfully reproduces the flattening. That is its entire informational content, and it is the content least worth having.

The diagram above carries what the window omitted: the surfaces are separated, each is named with a path a reader can type into regedit or services.msc and confirm on their own machine, and the count is visible — one control on the left, eight on the right, and the article says sixteen. Nothing in the picture is unfalsifiable. Every label is a string that either exists on a supported Windows or does not.

The screenshot also expires. It is a photograph of one build of one product on one release of Windows, and the moment either changes it becomes a picture of something that no longer exists, still sitting at the top of the article, still being read as current. Correcting it means finding the software again and re-capturing. Correcting a diagram means editing a line of text, which is the same operation as correcting a sentence — which is why our corrections policy can apply to the figures as well as to the prose.

Our judgement, labelled as judgement: for the whole class of software this site examines, the thing the screenshot was standing in for is a list of surfaces, and a labelled diagram states that list better than a capture of a window ever did. The screenshot was never the evidence. It was a picture of the place the evidence would have been.


§4What about the images at the top of articles?

Costs per hero
≤ 60 KB webp, or ≤ 14 KB SVG fallback
Costs to accept one
five checks, any single failure regenerates
Costs of a third failure
hand-authored SVG, noted in the colophon

Article heroes are abstract, textless, orthographic cutaways of a single mechanism. Every one of them contains exactly one boundary — a dashed vertical rule at 38% of the frame — and nothing else that could be read as a captured interface. They carry two inks and no lettering beyond bare numerals, because generated type is unreliable, unselectable and untranslatable, and the readers of this domain arrive in nine languages. Each one is inspected against a five-point rejection gate before it ships, and a hero that fails any single check is regenerated rather than retouched. The gate, the fallback path and the full image policy are on the colophon.

One inherited link we could not honour, recorded because the omission is deliberate rather than an oversight: the previous owner served an image at a screenshot.gif path on a subdomain. That subdomain is not built and no DNS record is published for it, so the link does not 404. It fails one step earlier, at name resolution: a lookup for the host returns NXDOMAIN, no connection is opened, no request is sent, and there is no server to return a status code of any kind. The distinction is worth keeping straight, because a 404 is a page we chose not to write and a name that does not resolve is a host we chose not to create. This is the second, because honouring the link would mean serving a screenshot.