A PDF on a phone does not lose resolution — it loses the page number
Every reader app advertises reflow. The pitch is simple and true: a PDF is laid out for paper, a phone is not paper, so the text is re-laid into a single column that fits the screen. On a two-column journal article, the difference is dramatic. Without it you are pinching and panning across a page designed for A4; with it you read the way you read anything else.
The cost is almost never mentioned, and it is not visual.
What a page number actually is
In a PDF, a page is a physical object. Page 214 is a rectangle of fixed dimensions, and everything a citation needs — “see p. 214, second column” — depends on that rectangle existing and being stable. Two people with different phones, different font sizes and different margins will still agree on what is on page 214, because page 214 is not computed. It is stored.
Reflow computes pages. It takes the text stream, measures it against the current screen and font, and produces a new sequence of pages that has nothing to do with the original. On a 6-inch phone at a comfortable size, a 300-page paper can become 900 pages. The reading experience improves enormously. The citation becomes meaningless: “p. 214” in the reflowed view is somewhere around page 640, and it will be a different number on the next device, and a different number again if the reader changes the font size.
This is not a bug in any particular app. It is what reflow is. The question is what a reader should do about it.
The decision
We keep the original page model as the source of truth, even when the text is being displayed reflowed.
Concretely: the reader knows which original page each reflowed paragraph came from. The progress indicator, the bookmark, and the “go to page” field all speak in original page numbers, not in the computed ones. The page counter at the bottom of a reflowed document therefore shows something slightly odd — it does not count the pages you are turning through — but it shows the number you would write in a footnote.
The alternative is what most readers do: show the computed page number because it is what the reader sees on screen. It is more intuitive for ten seconds and useless for every purpose a page number has.
Where this gets uncomfortable
Half-measures are worse than either extreme here.
If the citation says page 214 and the reader jumps to “page 214” of the reflowed document, they land in the wrong place. So the jump has to resolve against the original pagination and then translate back into the reflowed position — which means a jump can land mid-paragraph rather than at the top of a page, because the original page break happened in the middle of a sentence that reflow has since moved.
We accept that. A jump that lands one line off is recoverable; a jump that lands forty pages away is not.
The second uncomfortable case is figures. A figure in a two-column paper is anchored to a column, and reflow moves the text around it. Some readers re-anchor figures to the paragraph that references them, which is usually right and occasionally catastrophic: a full-page plate that originally appeared after the references gets pulled into the middle of section 3, and the reader concludes the document is corrupt.
Our rule is conservative: figures stay near their original page position, and if that means a figure appears later than the text that discusses it, so be it. Following a “see Fig. 4” reference backwards is a normal reading skill. Reconstructing a document that has been silently rearranged is not.
What reflow does to hyphenation
Single-column reflow narrows the measure, and a narrow measure with justified text produces rivers of white space unless hyphenation is on. Most systems turn it on automatically.
We do not, by default. Automatic hyphenation is only as good as the language dictionary behind it, and a paper in English with a German abstract will hyphenate the German incorrectly for as long as it takes the reader to notice. Ragged-right text with a few loose lines is a minor aesthetic loss; a document where every fourth line breaks a word in the wrong place is a reading problem.
The setting exists, it is just off until asked for.
The thing we will not do
We will not renumber the document and call it a page.
There is a temptation, once reflow is good, to present the reflowed pagination as the document’s pagination — to hide the original entirely, the way a word processor hides the difference between a draft and a proof. It makes the interface simpler and it quietly breaks the only contract a PDF has with the outside world: that page 214 is a place you can point at.
A reader that cannot be cited is a viewer. We are trying to build the other thing.
A note on the two engines
OmniShelf carries two PDF engines, and the choice is not cosmetic. One is the reference implementation used by most open-source readers; the other is faster on complex vector graphics but differs in how it handles font substitution and in what it exposes about text positioning.
Reflow needs the second kind of information — character positions, not just glyphs — so the reflow path and the plain rendering path do not always agree on where a line ends. When they disagree, the original page model wins, because that is the one the citation depends on. It is a small invariant, enforced in one place, and it is the reason the page number in the corner can be trusted.
The next note is about the shelf itself: why a novel and a scanned paper share one grid, and what that costs.