Changelog
Every version of Dark Route, newest first. One or two plain sentences each.
Version 0.47
Fix for a false positive in the shared-opening and shared-ending check. Handles on two accounts often share a joining word — "the_map_nerd" beside "the_trail_runner", or "i_am_ann" beside "i_am_jo" — and the check was calling that a family resemblance. A shared fragment is now put through the stop-word list first and only counted when a word of its own is left, so a shared "the_" or "i_am_" is no longer reported while a shared "nightowl" still is.
The clean line under that check was reworded to say what it actually measured: no two handles share a long opening or ending with a word of its own in it.
Known limits: the filler test is the same stop-word list the bio checks use, so a shared fragment built from a common word that is not on that list (say "official" or "real") is still reported. And as with every change here, this was reasoned from the code and walked through on paper, not watched in a browser — there is none in this workspace.
Version 0.46
Fix for a false positive in the word-in-bio check: when only one bio is pasted, a six- or seven-letter word from a handle or display name is no longer reported as a link. With one bio there is no second bio to show that a word is just the subject of the writing, so only words of eight letters or more are matched there. Two or more bios are read exactly as before.
When those shorter words were left out, the result now says so, instead of a clean line that would have claimed more than the run checked. The hint under the bios box names the rule.
Known limit: as with every change here, this branch was reasoned from the code and walked through on paper, not watched in a browser — there is none in this workspace.
Version 0.45
The hint under the handles box now names the word-order check added in 0.44, with the same example it uses ("mapnerd_leeds" beside "leeds_mapnerd"). Until now that check could fire without the page ever having said it looks for reordered words. No check, score or wording in the result changed.
Version 0.44
New check: two handles built from the same words in a different order — "mapnerd_leeds" against "leeds_mapnerd". The old near-identical check walks past these, because almost no letters sit in the same place, but a search for either word still finds both accounts. It fires only when the two handles hold the same set of words (four letters or more each) and the order really differs, so one shared word is still left to the shared-fragment check above.
Known limit: this reading was reasoned from the code and walked through on paper, not watched in a browser, because there is none here.
Version 0.43
Phones: the top of the page is tighter — smaller page gutter, closer headings, a slightly smaller lead line — and the lead is now one sentence instead of two, so the first input box is reachable with less scrolling. No wording about how the checks work was removed, nothing is hidden and no check changed.
Known limit: this spacing was chosen by reading the CSS, not by resizing a real phone window, because there is no browser here.
Version 0.42
The word-in-bio check now also says when the matched word sits inside another of your handles or display names too, because that one word is then doing the work on more than one account. Only the finding text changed; which words fire and the score are unchanged.
Known limits: a common 6-letter word inside a single bio could still be flagged (fixed in 0.46), and the check names up to two such words before it says "and N more".
Version 0.41
Changelog trimmed: entries below 0.30 condensed to one or two sentences each. No version was dropped and no product code changed. The page is about 32 KB smaller than it was.
Version 0.40
The 0.38 word-in-bio check no longer fires on a word that also turns up in another bio — that reads as a subject word, not a link. When the set-aside word was the only match, the clean line says so instead of claiming nothing turned up.
Known limit: a common 6-letter word inside a single bio could still be flagged (fixed in 0.46).
Version 0.39
The button row no longer uses a fixed 30rem break point: it is now an auto-fit grid, so buttons are full width per column and fall back to one per line on very narrow screens.
Version 0.38
New check: one long word of a handle or display name written inside a bio (6 or more letters), reported as a medium signal.
Version 0.37
Section 1's two preamble paragraphs merged into one, so the first box sits higher.
Version 0.36
New check: a run of four words copied from one bio into another, in the same order.
Version 0.35
Dropped a hint paragraph that repeated the 20-line cap already stated under each box, and shortened the tip under the buttons.
Version 0.34
The "no finding" paragraph no longer repeats the untested-link detail that the note above it already names.
Version 0.33
When no handle or display name could be compared between boxes, the "no finding" line now says so: no finding among the checks that ran.
Version 0.32
The page now names the entries that were too short to compare between boxes, instead of staying silent about them.
Version 0.31
The cross-box clean line is only claimed when at least one pair was actually testable.
Version 0.30
The cross-box clean line is only claimed when two or more boxes have entries.
Version 0.29
New check comparing boxes: a handle equal to a display name, or a handle or display name written inside a bio. The clean example was renamed so it still reads clean.
Version 0.28
The "compared N entries" line moved under the verdict and is announced with it.
Version 0.27
First changelog trim: the longest entries lost a paragraph.
Version 0.26
The low-band verdict no longer repeats the entry count; it points at the line below instead.
Version 0.25
New line near the verdict: "Compared N entries — x handles, y display names, z bios".
Version 0.24
The bar explanation and the low-bar caveat merged into one note, and the key gained the "work down the checklist in order" sentence.
Version 0.23
"What looks separate so far" lists at most six checks and says how many more came back clean.
Version 0.22
First attempt at the button row on a phone (a 30rem break point, replaced in 0.39).
Version 0.21
Announcements moved off the whole results block: the verdict, the compared count and the findings are announced; standing hints and the checklist are not.
Version 0.20
The empty result explains itself and points at the clean example.
Version 0.19
Second button: "Load clean example" fills all three boxes with a pair built to read clean.
Version 0.18
The reading key and the low-bar note became boxes, so they stand apart from ordinary hints.
Version 0.17
A low bar is not proof of safety — a note says what the bar counts and what it cannot see.
Version 0.16
A key explaining "strong signal" against "signal", and what the bottom list means.
Version 0.15
Hints state that near-miss handles, close display names and near-copy bios are checked, and that the bar is a rough count.
Version 0.14
Header mark: a small inline shield-and-route drawing. No image file.
Version 0.13
New check: two bios sharing most of their distinctive words.
Version 0.12
New check: display names where one sits inside the other or is nearly identical.
Version 0.11
Checklist item: post a re-exported copy of a photo rather than the original file.
Version 0.10
An on-screen note that Ctrl+P prints just the findings and the checklist, and how to check the preview.
Version 0.9
Clearer empty states, including "only one box was filled" with what that leaves untested.
Version 0.8
Print stylesheet: paper gets the findings and the checklist, not the buttons or the hints around them.
Version 0.7
Warning when a box holds lines past the first 20, which were read nowhere and dropped quietly before.
Version 0.6
Findings and their checklist items share one strongest-first order, and long runs of repeated items are capped.
Version 0.5
Section "What this page cannot check": reused photos, photo metadata, posting times, a reused email or phone number, forgotten old accounts.
Version 0.4
Copy the checklist as text.
Version 0.3
Optional label per line: write twitter: nightowl_92 and findings name the site and the line.
Version 0.2
Findings quote the entry they came from and name the box and line, instead of just a number.
Version 0.1
First version, built by an AI agent. Three boxes (handles, display names, bios) and checks for repeated handles, look-alike letter and digit shapes, shared openings and endings, near-identical handles, repeated display names, repeated number runs and reused bio words, with a rough linkage bar and a base checklist. Nothing is saved between visits and nothing is sent anywhere.