Newest first. Every entry below is something that actually happened on the date it carries.
This site is new and has not been through a public release yet, so what follows is its build history rather than a record of things that changed under anyone's feet. Each date is the day the work landed.
When a change alters an answer this site gives — where the same list would now produce a different result — the entry says Changed an answer. That is a different kind of event from a new tool or a visual tidy-up, and it is the kind a returning visitor might actually need to know about. Nothing here has needed that marker yet, which is worth stating plainly rather than leaving you to notice its absence.
The coin page already explained that a run of heads does not make tails due. It now answers the question people are really asking, which is whether a long run is strange at all: in a hundred flips, a run of six or more turns up most of the time, and the longest run you are more likely than not to see is seven. The figures are calculated exactly rather than estimated, and the calculation is checked against a count of every possible sequence.
Type everyone taking part and draw: nobody gets themselves, and you can mark which people live together so they never draw each other. Pass one device around and each person taps their own name to see only their own. If the households you set up make a draw impossible, it says so straight away and names the people involved rather than trying and failing quietly.
The site is now published at its own address, and there is somewhere to write to. The contact page also sets out plainly what cannot be helped with — a lost list, or a result from last week — because nothing about a draw is kept anywhere, by anyone, including us.
The fairness page publishes the measurements behind everything else here: how the shuffle was tested, over how many runs, and what a deliberately broken version of it scores on the identical check. It also sets out the things we deliberately do not claim, which is the part most worth reading.
Flip up to a hundred coins at once and keep a tally that survives the tab closing, or draw numbers from a range with or without repeats. The number tool will happily give you ten different values from a range of a billion, and it tells you before you press the button when a request is impossible — ten different numbers cannot come out of a range that only holds five.
Any number of dice from four to a hundred sides, with each individual face shown rather than only the total, an optional modifier, and a history that is still there tomorrow along with the dice you set up.
The headline tool. Spin a wheel over a saved list, cross each name off as it is drawn, give some names a bigger share of the wheel, and share the whole list as a link that carries the names in the address itself rather than uploading them anywhere. The pick is made before the wheel starts moving, so the result is announced as text too and the animation can be skipped entirely.
Split a list into a set number of teams, or into teams of at most a given size. Sizes are balanced and, when the split is uneven, which team gets the extra person is decided at random — otherwise team one is the biggest every single time, which is a real unfairness in a tool whose whole promise is fairness.
The order shuffler, plus the part every later tool reuses: lists you type once, save under a name, and find again next time. Duplicate names are kept rather than merged, because two people in a group really can share a first name and quietly removing one of them removes a real person from the draw.
Three of the ways this page could be wrong are checked automatically, and the build fails rather than publishing them: an entry naming a tool that does not exist, a date that is not a real date, and entries out of order. The date at the top of each group is the entry's own date rather than a separate “last updated” line, because two dates for the same thing can only ever disagree.
What no check can tell you is whether a sentence above is true. Nothing can read prose and know whether the thing it describes happened, so these are written from the repository rather than from a plan — and saying where the automated part stops seems better than implying it goes further than it does.
The most recent change here was on .