PubTrivia has five free tools on its marketing site. None of them has a backend, none of them has a <form> element, and nothing typed into them is ever sent anywhere, which is the promise each page makes in writing. That also means no HTML validation ever runs: a max="200" attribute is only enforced when something submits a form, and nothing here submits anything.
So I spent an afternoon lying to all five of them and writing down what came back. Every result below is reproducible on the live pages:
- /tools/quiz-night-profit-calculator
- /tools/scoresheet-generator
- /tools/qr-code-generator
- /tools/quiz-night-poster-maker
- /tools/team-name-generator
The policies, side by side
Five tools, four different answers to the same question.
tool field attributes what state does what the output does ------------------ ----------------- ---------------- --------------------- -------------------------- profit calculator 10 number inputs min/max/step stores any finite clamps percentages, number, 0 for the floors negatives, prints rest the rest verbatim scoresheet rounds/questions/ min/max clamps to the range cannot be made to teams on every keystroke exceed 12/20/40 scoresheet quiz name none stores every char draws it unwrapped poster maker 7 text fields maxlength cuts to the limit, silently drops anything strips characters outside Windows-1252 qr generator web address none stores every char accepts anything the up to the URL parser URL parser will swallow team names none none nothing to type n/a The calculator takes 1e9 teams
The profit calculator asks a landlord for ten numbers and tells them what a weekly quiz is worth. Its teams field carries min="0" max="200" step="1". Typing 1e9 into it is accepted by the number input, accepted by React state, and accepted by the arithmetic:
Per night £28,799,999,957 Over a year of weekly quizzes £1,497,599,997,754 Players in the room 4000000000, of whom 3999999994 would not have come without the quiz The field does go :invalid at that point. I checked in the console: document.getElementById("profit-teams").checkValidity() returns false, and nothing anywhere reads it. There is no form to block, no submit handler to intercept, and no styling hung off :invalid, so the only consumer of max="200" is the spinner arrows.
Three more results from the same field:
-
1e999parses toInfinity, the state setter rejects anything non-finite, and the controlled input re-renders as0. The field rewrites itself in front of you, which is the most honest behaviour in this whole post. -
-5stays on screen as-5. Every number derived from it is computed from0, because the arithmetic floors its inputs at zero. The number you can see and the number being used are different numbers, and nothing says so. -
1e21crosses the threshold whereNumber.prototype.toLocaleStringgives up on grouping, so the money line reads£28,800,000,000,000,000,000,000while the player count, which is printed raw, reads4e+21in the middle of an English sentence.
Percentages behave differently again. Margin accepts 500 and the prize share accepts -50, and both are clamped to the 0 to 100 range inside the calculation. You get a perfectly correct answer to a question you did not ask: £488 a night, computed at 100 per cent margin and a zero prize, while the two fields on screen still say 500 and -50.
A cost of £0.004 a month moves the break-even point
This one is not a fuzzing result, it is a real finding that fuzzing surfaced. Put 0.004 into "software or questions per month" and the break-even line moves from 1 team to 2.
That is correct arithmetic. It is also the tool admitting what its default answer means. With no costs at all, the night nets exactly £0 at one team, because four players against six regulars-who-would-have-come-anyway leaves nobody whose bar spend counts, and the entry money all goes back out as the prize. The search for the break-even point asks for netPerNight >= 0, zero qualifies, and the page reports "1". Add four tenths of a penny a month and the same night is 0.09 pence down, so the answer becomes 2.
A landlord reading "teams to break even: 1" will not read it as "at one team you make nothing". That is a copy fix, not a code fix.
The one that clamps
The scoresheet generator does the thing the calculator does not. Its limits live in the module next to a comment that says they are the bounds the inputs clamp to, so a typo cannot ask for a 900-page PDF, and the setter runs every keystroke through Math.max(min, Math.min(max, Math.round(value))).
So typing 60 rounds, 40 questions and 200 teams gives you back 12, 20 and 40 in the fields themselves, and a 13-page PDF of 104,206 bytes, which is the largest file the tool can produce. You cannot make it generate anything bigger, and the input tells you it refused by changing under your cursor.
Then there is the quiz name next to it, which has no maxLength at all.
pasted name 480 characters rendered width at 16pt 1,193.5 mm printable width of A4 180 mm characters that fit 72 download filename 492 characters jsPDF's text() does not wrap, so the other 408 characters are drawn off the right edge of the page and off the next five pages' worth of imaginary paper. The filename is built by slugifying the same string, so the browser is handed a 492-character name for a file on a filesystem whose per-component limit is 255 bytes, and something downstream has to truncate it.
Two tools, two definitions of "web address"
The QR generator and the poster maker both take a web address, both assume https:// when you leave the scheme off, and both hand the result to new URL(). Only one of them checks the result.
The QR generator accepts whatever the URL parser accepts, and the URL parser is more forgiving than you would like:
typed encoded in the QR code ------------------------- ---------------------------------------- localhost https://localhost/ 0 https://0.0.0.0/ not a url https://not%20a%20url/ The Red Lion quiz https://the%20red%20lion%20quiz/ ftp://files.example.com https://ftp//files.example.com tel:12345 https://tel:12345/ .. https://../ Every row there produces a real, scannable, downloadable QR code. The poster maker adds one line, requiring a dot in the hostname, and that single line rejects localhost, not a url and The Red Lion quiz while still cheerfully accepting 0, because 0.0.0.0 has three dots in it.
A caution if you go testing this yourself: the two runtimes disagree. In Chrome, new URL("https://not a url") percent-encodes the spaces and returns a URL. In Node 24, the same expression throws ERR_INVALID_URL. I wrote half of this section from Node output and had to throw it away, because the code runs in a browser and the browser is the lenient one.
Neither tool imposes a length limit, so a 2,000-character address is fine by both, and the QR encoder quietly absorbs it: version 40, 177 modules, 33,839 bytes of PNG.
The poster deletes characters and does not mention it
The poster maker writes into a PDF through jsPDF's built-in Helvetica, which can draw Latin-1 plus the extra punctuation in Windows-1252 and nothing else. So the layout code strips everything else before measuring or drawing, which is the right call: the alternative is mojibake, and an earlier post about the scoresheet PDF measured exactly that happening in the other tool, where a Japanese quiz name is written into the file as raw bytes and comes out as 0¯0¤0º0Ê0¤0È.
The trouble is the silence. Measured live on the preview:
typed into "venue name" input keeps poster shows --------------------------- ------------ ------------------------- The 🍺 Red Lion 🎉 18 chars "The Red Lion" 居酒屋クイズナイト 9 chars nothing, the line is gone " " (three spaces) 3 chars nothing, the line is gone 60+ chars of venue name 60 chars wraps to 2 lines at 11.5pt A venue with a Japanese name gets a poster with no venue name on it, and the only hint is that the preview quietly has one fewer line of text than it did a moment ago. The 60-character limit, by contrast, is visible: the input stops accepting characters, and the name shrinks and wraps to fit rather than running off the page, which is the behaviour the whole layout module was written to guarantee.
What this does not show
None of this is a security finding, and it would be dishonest to dress it up as one. There is no server to protect: the tools are client-side, the input never leaves the browser, and the worst case for the site is a wasted CPU cycle on somebody else's laptop. I also could not make any of the five throw. No stack traces, no blank screens, no hangs, including at a billion teams and 2,000-character URLs.
What it does show is a cost paid by the person using the tool. A landlord can be shown a number computed from a value the screen contradicts. A pub with a non-Latin name can print 50 posters without its own name on them. A quiz called something long gets a PDF with its title off the edge of the paper and a filename the operating system has to cut. Those are not crashes, which is precisely why nobody finds them.
The cheap fixes, in order
- Export the scoresheet's
clampand use it in the calculator. The limits are already written down in themaxattributes; they just need a reader. - Put a
maxLengthon the quiz name, and cap the slug used for the filename at something like 60 characters. - Move the URL rule into one function in
liband have both tools call it, so there is one answer to "is this a web address" instead of two. - When the poster's sanitiser drops characters, say so under the field. One line: "some characters cannot be printed and were left out".
- Change "teams to break even: 1" to say what it means when the night nets zero.
If you want to reproduce any of it, the tools are at pub-trivia.app/tools, there is no sign-up, and the Network tab will stay empty while you do it.

