Quick Pick / rank a list
Rank a list by picking pairs
Ranking twenty things at once is hard. Choosing between two is easy. Paste your options, answer "this or that?" a few dozen times, and get an order with an honest range for each place. Send the list to your team and combine everyone's picks.
1. Your options
The pictures stay on your device: the page opens them locally and never uploads them. Their file names become the list, so a teammate opening your link attaches the same files on their device.
Attach the pictures
This list is pictures. They never travel in the link, so choose the same files on this device (all at once is fine; they are matched by file name).
Sort first: one tap each
If you can roughly tell top from bottom, a quick sort first saves picks: in simulations, 20 options took about 64 taps and picks after a careful sort instead of 93 picks. Aim for about a third in each pile; it does not need to be exact. If you are unsure, skip it: a careless sort costs picks rather than saving them.
Keys: 1 top, 2 middle, 3 bottom, U back.
Which do you prefer?
How many picks a list takes, measured
Keys: ← left, → right, ↓ about equal, U undo, F look closer.
Which do you like most?
Tap the one you like most.
Keys: 1 to 4 pick the option in that place (most first, then least), U undo.
Your order
| # | Option | Rank range | Top pick |
|---|
The rank range is where 90% of the orders consistent with your answers put the option. A wide range means your picks have not separated it from its neighbours yet: answer more to narrow it.
3. Combine the team's picks
How it works
- The model is Bradley-Terry: each option has a hidden score, and the chance you pick A over B rises with the gap between their scores. After every answer the page refits all your answers at once (Newton's method, with a Laplace approximation for the uncertainty), the same code the Quick Pick survey app runs.
- The next question is the pair whose answer is expected to tell the most about the order: two options the model cannot yet separate. Which side each appears on is random, so screen position cannot creep into the result.
- When to stop: "settled" is how closely the orders your answers still allow agree with the order shown (the average Spearman correlation over draws from the model). It stops at 90%. On simulated respondents who answer with ordinary noise, the order shown at that point matched their real order with a Spearman correlation of about 0.94. If many of your picks are close calls, it checks again with the model fitted to how strongly you seem to feel (an estimate from your own answers) and keeps asking until that read passes too, because a model that assumes strong likes reads noise as a settled order (on 25 simulated near-coin-flip options it had claimed 90% for an order that was right at about 0.56). Two options you find genuinely close may never separate; the rank range shows that, and you can stop at any time. The results' ranges and chances use the same fitted model, so if your picks were close calls they come out wider and less sure than they would for someone who answered every pair firmly.
- Full order or favourite? Full order spreads questions across the list and stops when the whole order is 90% settled; that can leave the top two close. "Be sure of the favourite" asks only among options that could still be the favourite and stops when one has a 90% chance. In simulation it took about a third more questions and named the right favourite 12 times out of 12, against 6 to 8 for the full-order stop.
- Find my top k is for a shortlist: it asks only about options that could still be either side of the cut and stops when the shortlist is 90% settled. On simulated people who answer with ordinary noise, choosing the top 3 or 5 of 15 to 30 options, it named exactly the right shortlist 56% of the time against 26% for the full-order stop, at about 1.6 times the questions; for people who never waver it costs about a fifth more questions, and on lists of 10 it gains little.
- Sort into piles first (lists of 8 or more, optional): one tap per option, near the top, in the middle or near the bottom. The piles become each option's starting estimate, allowing for a misplaced option, and the picks then go mostly to options in the same pile, with every 4th pair chosen as if there were no piles so a misplaced option still gets tested. If your picks keep disagreeing with your piles, the page notices and leans on the picks. On simulated people with ordinary noise it took 64 actions for 20 options after a careful sort, 87 after an ordinary one and 120 after a careless one, against 93 picks without; the order came out about as good in each case.
- Best and worst of four (optional mode): each screen shows four options and you tap the one you like most, then the one you like least, the question MaxDiff surveys ask. It is scored with the exact best-worst model (the best is a choice among the four, the worst a choice among the other three), not split into implied pairs, which would count two taps as five independent answers and make the ranges too narrow. The next screen is the four whose answer is expected to tell the most. On simulated people with ordinary noise, 15 options took about 27 screens (54 taps) against 70 picks for an order about as good, and 8 options 13 screens against 38 picks. A screen takes longer to read than a pair, and how much longer is not measured yet.
- Team results refit each person separately, then average their centred scores, so someone who answered 80 questions counts the same as someone who answered 30. The page shows how much people agree and where they split.
Ready lists to start from: a feature backlog, your personal values, a name shortlist, a group trip, team goals, customer pain points, a to-do list, or all ready lists.
Used to MaxDiff surveys? "Best and worst of four" is that question, scored with the exact best-worst model; how it compares with pairs.
Ranking pictures (photos to cull, logo drafts, design options)? Use "Or rank pictures" above; how picture ranking works.
Choosing between a few options on several criteria (rent, commute, space)? Weigh criteria by picking pairs: a decision matrix with no typed weights.
Running a survey with many respondents, demographic splits and an admin view? The full app is open source: deploy your own copy.
Privacy and usage counting
Everything runs in your browser. Your list and answers are never uploaded; we have no accounts, tracking cookies or analytics vendor. Your progress is kept in this browser's local storage so a reload does not lose it. The page counts anonymous events in local storage (lists started, picks made, results shown, codes copied, team combines), nothing about who you are or what you ranked, and the site counts one page view per load without any identifier.