The Flush Margin Isn't Proof of a Program
A block of plain text where every line ends flush on the right margin, no stretched gaps, no trailing hyphen, looks like it came out of a formatter. Ninety-nine times out of a hundred it didn't. Monospace text has no fractional space to distribute. You can't nudge a word two pixels left to make a line land. Hyphenation is the other lever every justifier has, and in monospace it helps for a moment and then stops: a break still falls on a column boundary, so whatever columns are left over go back into the gaps anyway, and the hyphen you spent to get there is itself a seam sitting at the edge of the block. What remains is stretching the gaps between words until the line reaches the margin, and past a certain point that stretching stops looking like typography and starts looking like a mistake, big ugly rivers of whitespace running down the page. You've seen the artifact: four words on a line, canyons between them, the eye tracking the gaps instead of the sentence. The algorithm's compromise is visible because the algorithm runs out of moves and still has a line to fill.
A person facing the same constraint has a different move: throw out the word entirely and pick a different one that happens to be shorter or longer. Synonym, rephrase, reorder the clause, cut a sentence to a fragment, whatever gets the last character to land on the margin. That's not a formatting operation, it's an editing operation, and it can be repeated as many times as the writer has patience for. The plaintext game walkthroughs of the late nineties are where I first noticed this. Fixed column width, no formatter anywhere in the pipeline, some of them running the length of a novella, and here and there one where the right margin lands flush the whole way down. No tool did that. Someone chose words carefully, section after section, past any point where a normal person would call it done.
That's the general shape of the thing worth noticing: when a constraint has an obvious automated solution and the automated solution has a known, visible failure mode, and you see a result with none of that failure mode showing, the automated solution probably isn't what produced it. The algorithm is stuck with a small, fixed set of levers, and when those levers run out, the residue shows up as a gap, a rounding error, a slightly-off placement, a compromise you can point to. A person isn't stuck with those levers. A person can swap the whole unit, not adjust it, and try again as many times as it takes.
Crossword construction has the same tell. An autofill tool given an open grid will happily drop in a run of glue words nobody wants to see, ENE, EPEE, ONO, whatever satisfies the crossing letters, because its only goal is a valid fill and it will take the first one that works. A constructor who wants a livelier grid will reject that fill, rework the black-square pattern, try a different seed word in the corner, and do it again. And again. Some constructors will spend most of a day on a single stubborn quadrant because they're not solving "produce a valid grid," they're solving "produce a grid I'm not embarrassed by," which is a different problem with no shortcut through it. A grid with no glue anywhere, top to bottom, is a grid somebody refused to ship until it stopped looking assembled.
None of this is really about text or crosswords. It's about what a floor does to the evidence. An automated process has a fixed toolkit and no shame about using the worst tool in it once the good ones run out, so its failure mode is legible if you know to look, a stretched gap, a lazy fill, a seam. A person grinding through the same constraint by hand has no floor at all, just diminishing patience, and patience doesn't leave a signature the way a bad algorithm does. So the absence of a visible compromise isn't proof that no compromise was needed. It's often proof that someone was willing to keep redoing the corner until the compromise disappeared.
Next time something looks suspiciously seamless, the question isn't what tool made this. It's who had the patience not to need one.