Checked against the current Windows build. Where an article names a menu, a button or a message, that is the wording in the application, not a paraphrase.
A scan and a Word export are not the same problem
Before it compresses anything, Creasepoint counts the images on every page and works out how much of the file they account for. Below about forty per cent it stops and says the document is mostly text and vectors, that there is little here to re-encode, and to expect a few percent at best. The command line prints the same note above its result.
That is not a hedge. A font is already compressed and a vector line is a handful of numbers; neither has a quality setting to turn down. A JPEG of a scanned page does have one — it can be re-encoded at a lower quality and downsampled to a lower resolution, and that is where every megabyte a compressor saves comes from. So the first question about a file that will not shrink is not what size it is. It is whether anything in it is a picture.
The band has a floor, and a file under the floor is rejected too
Two PDF presets ship with the application, each showing where its numbers came from: NTA — NEET UG, a certificate between 50 and 300 KB, read from the 2026 information bulletin, and the Income Tax e-filing portal, an attachment up to 5 MB. creasepoint portals lists them. Above them in the panel sit the plain ceilings, from under 100 KB to under 5 MB.
A band produces four different answers and each has its own sentence. Already inside the band, left untouched, because re-compressing a file that fits only loses quality. Re-encoded and inside, with the resulting size. Below the floor, in which case compressing further makes rejection more likely rather than less. Or above the ceiling with nowhere acceptable left to go. Read that sentence rather than the size in Explorer — three of the four outcomes look identical from the file list.
The ceiling message is a finding, not a setting to argue with
The fitter searches a scale of compression settings for the gentlest one that still lands inside the band, so what you get is the best quality the band allows rather than the smallest file the encoder can produce. It also refuses to search past the point where text stops being legible. When even that setting overshoots, what comes back is the smallest acceptable result, its size, and the ceiling it missed: the smallest we can go without the text becoming unreadable is 1.4 MB, above the 300 KB ceiling.
There is no hidden setting behind that message. It is the answer for that document, and its value is that you now know the size of the gap. A file that stops at 1.4 MB against a 300 KB cap will not get there by re-running the same operation with a different preset selected.
What to do when the ceiling is genuinely out of reach
Send fewer pages. Most portals ask for one certificate, not the forty-page bundle you were handed. Extract pages builds a new document from the ones you select, and creasepoint extract file.pdf --pages 1,3,5-7 --out out.pdf does it from a script. This is the fix that works, and it is the one people try last.
Rescan, if the scanner is still in reach. A page captured at 600 dpi in colour carries several times the data of the same page at 200 dpi in greyscale, and no compressor recovers that difference without visible cost. Creasepoint cannot undo a scanning decision.
Then read the notice again for the file type. A PDF preset will not help you past a portal that wanted a JPEG, which is why the image presets are filtered out of the compression list rather than left sitting there to be picked on a deadline. And if the PDF came out of a document you still have, make a smaller one from the source. The size-limit guide covers the presets in full.
A folder of them, and what batch will not do
The Batch tool runs one operation over every PDF in a folder and writes the results into a folder you choose, each file handled in its own worker. Its compression job has a single fixed target, under 300 KB, and no way to give it a floor. So the 50–300 KB band across forty files is a command-line loop rather than a batch run, and the command line has the shape of one.
Whichever route you take, read the per-file results rather than the number of files that finished. If several come back carrying the mostly-text note, that is the same finding arriving forty times over, and the answer is still fewer pages rather than a harder setting.
The two places worth looking next.
Read the matching guide if you want the reasoning rather than the fix, and check limitations before spending any longer on it — some of what people look for here is absent on purpose.