Ranked Choice Picker

Ranking ten things at once is genuinely hard, and most people who try produce an order they immediately want to change. Choosing between two is easy. This asks you to do the easy thing repeatedly and assembles the answer: which do you prefer, this or that, about twenty-three times for ten options — and the order that comes out is exactly the one your answers imply.

Your options

One per line. Everything stays in your browser — nothing is uploaded, stored or put in the address bar.

Your ranking will appear here once you have answered the questions.

What this generator does

Takes your list and ranks it from your own answers to a series of two-way questions. It runs a merge sort in which every comparison is a question put to you, so it asks far fewer questions than comparing every pair would: about 23 for ten options against 45, and about 112 for thirty against 435. The order it produces is exactly the one your answers imply — not a score, not an estimate — and it also offers to test whether those answers are consistent with each other, which is the one thing this method cannot guarantee on its own.

How to use this tool

Enter your options, one per line.
Press Start ranking.
Answer each question by choosing the one you prefer. Undo if you change your mind.
Read the finished order, and use Check my answers to see whether your preferences hang together.

Understanding the controls

Your options: One per line, in any order — the starting order does not affect the result. Blank lines are ignored. Two identical lines are kept as two separate options rather than merged, because deciding they are the same thing is your call and not the tool's.
Which do you prefer: The only question it ever asks. There is no scoring, no scale and no ties: every answer moves the sort forward by one step, which is what keeps the number of questions down.
Undo: Steps back one answer at a time, as far as you like. Nothing is recomputed or replayed — the tool keeps each state as it goes, so undoing is exact rather than an approximation of where you were.
Check my answers: Asks about pairs the sort never needed to ask about, because your other answers already implied them. Agreement means your preferences form a consistent order. Disagreement is common and is not an error — it means they do not, and it is worth knowing before you treat the result as final.

Worked examples

Ten options: About 23 questions, against 45 for every pair.
Twenty options: About 64 questions, against 190.
Thirty options: About 112 questions, against 435.
What a check finds: 'You said tacos beat pizza, but the ranking put pizza first' — a cycle in your own preferences.

Common use cases

Putting a shortlist of names, ideas or candidates into a defensible order
Settling what a group actually prefers rather than what the loudest person said
Ordering a backlog when everything feels equally urgent
Working out your real favourite when you have too many options to hold in your head
Producing a ranking you can show someone, with the number of comparisons behind it

How this generator works

The list is sorted by merge sort, with one change: where a sort would compare two values, this asks you. Merge sort splits the list into single items, then repeatedly merges sorted runs in pairs, and merging two sorted runs only ever needs to compare their current front items — which is why so many questions can be skipped. Once you have said A beats B and B beats C, no question about A and C is ever asked, because the answer is already determined. Measured over 200 random starting orders, ten options take 22.6 questions on average and thirty take 111.9, against 45 and 435 for comparing every pair. Information theory says no method can do better than about 22 and 108 respectively, so the saving is close to the most that is available. The finished order is read straight off the last remaining run.

Randomness and fairness

Nothing about the ranking is random. The questions asked follow deterministically from the list and your answers, and the same answers always give the same order. The only place randomness appears is in choosing which unasked pairs to offer as consistency checks, which uses the browser's cryptographic random number generator and does not affect the ranking at all.

For how randomness is produced across the whole site, see how Generate Random works.

Assumptions this tool makes

Each question is answered on the same basis. Switching what you are judging partway through — price for some pairs, taste for others — produces an order that means nothing, and no tool can detect it.

Limitations and good to know

It assumes your preferences are transitive — that liking A over B and B over C means you like A over C. Sorting cannot work without that assumption, and real preferences often break it. Where they do, the order depends partly on which questions happened to come up. The consistency check exists to tell you when this has happened rather than to hide it.
There are no ties. Every question needs an answer, and two options you genuinely rate equally will still be separated by whichever you clicked.
It ranks what you give it. It knows nothing about the options themselves, so it cannot tell you that one of your brand names is already a trademark or that one of your candidates is unavailable.
Lists are capped at 40 options. Beyond that the question count stops being something anyone will sit through, and a ranking abandoned halfway is worth nothing.
The result reflects the answers you gave on the day. Ask again next week and a different order is entirely possible — that is a fact about preferences, not a fault in the sort.

Common mistakes

Trying to answer as though ranking the whole list: Answer each question on its own. The point of the method is that you never have to hold the full order in your head — that is the tool's job, and second-guessing it produces worse answers, not better ones.
Treating the result as objective: It is exactly as objective as the answers you gave. What it adds is consistency and a record of how many comparisons it rests on, not authority.
Assuming a disagreement in the check means a bug: It almost never does. It means your preferences contain a loop — you prefer A to B, B to C and C to A — which is normal and which no ranking can represent. Run it again, or accept that the top few matter and the middle is genuinely unsettled.

Practical tips

Trim the list before you start. Twelve real contenders produce a better ranking in a third of the questions than thirty with padding in it.
For a group, have one person read the questions aloud and take the room's answer. It keeps the discussion on one comparison at a time, which is the whole benefit.
If the check disagrees, look at where. A loop among the bottom three does not matter; a loop involving your top pick means the decision is not made yet.

Troubleshooting

It is asking more questions than I expected: The count shown is an estimate of the worst case, and the sort usually finishes a little under it. Ten options take about 23 questions and twenty take about 64 — if it feels longer than that, the list is probably longer than you think.
I clicked the wrong one: Press Undo. It steps back exactly one answer and can be pressed repeatedly.
The ranking is not what I expected: Run the consistency check. An unexpected order usually means either that a genuine preference surprised you — which is the point — or that your answers contain a loop, which the check will find.

Privacy and your data

Your options never leave the browser. They are not uploaded, not stored, and never written into the page address — which matters more here than on most tools, because a list of candidates, names or priorities is often something you would not want in a URL you might paste somewhere. Analytics records only that the tool ran and how many options there were, never the options themselves or the order you put them in. Start over clears everything immediately.

Frequently asked questions

Why not just ask me to rank the list directly?: Because people are measurably worse at it. Ordering ten things at once means holding ten comparisons in mind simultaneously; choosing between two means holding one. The method trades your effort per question, which becomes tiny, against the number of questions, which stays manageable because most comparisons can be inferred rather than asked.
How many questions will it ask?: About n log n of them. Measured over 200 random starting orders: ten options take 22.6 questions on average, twenty take 64.3, and thirty take 111.9. Comparing every pair would take 45, 190 and 435. The theoretical minimum for those sizes is about 22, 62 and 108, so this is close to the best any method could do.
What if my answers contradict each other?: The sort cannot notice, which is why the check exists. Once you have said A beats B and B beats C, it will never ask about A and C — so if you would in fact have picked C, the order you get depends on which questions came up. Check my answers asks exactly those skipped questions and tells you how many disagreed. A disagreement is not a fault: it means your preferences contain a loop, which is normal and which no ranking can fully represent.
Is any of this random?: No. The same list and the same answers always produce the same order. This page used to shuffle your options into a random order, which was a different tool with the same name; it was replaced because the thing people search for here is help ranking, not a shuffle. If a random order is what you want, the random task order generator does that properly.