About SongRank

How it works

Start from a ready-made list — all-time greats, a genre, a decade, or what's charting right now — or build your own by pasting a list or searching for songs one at a time. A ready-made list is just a starting point: it opens in the same build screen, where you can add to it, drop anything you don't know and rename it. Each song is matched against Apple's music catalogue to pull in artwork and a preview clip.

Then you vote. Two songs at a time, with a short clip of each so you can refresh your memory before deciding. When a matchup is genuinely too close to call, “flip a coin” records it as a tie instead of forcing a pick — neither song gains or loses ground, and you move on. When the matchups are done you get a full ranking, not just a winner, and you can export it.

Sharing and comparing

Rankings are private unless you say otherwise. Make one public and it appears on Browse, where anyone can read it, copy its song list, and rank the same songs themselves — their votes are their own, and nothing they do can change your ranking. Their copy is a template, not a duplicate: they can add and remove songs, swap any of them for a different recording, rename it and choose how thorough their own ranking should be, rather than inheriting your answers.

Once you have both ranked the same list, comparing the two shows how closely you agree, which songs you disagree most sharply about, and every song side by side. Following someone is one-way and simply puts their public rankings first on your Browse page. Pick a username and friends can find you by it — neither of you ever has to hand out an email address.

Why pairwise ratings, not Swiss or a bracket

A knockout bracket eliminates half the field every round. That finds a winner quickly, but it tells you almost nothing about the rest: a song knocked out in round one by the eventual champion is ranked no higher than one knocked out by the worst song in the field. Most of your votes end up discarded.

SongRank used to run on the Swiss system — the same format Magic: The Gathering tournaments use — and Swiss is a real improvement over a bracket. But it still isn't enough, and the reason is arithmetic rather than opinion. Producing a provably correct order of n items needs at least log₂(n!) comparisons — that is how many yes/no answers it takes to distinguish between all the possible orderings of the list. For 256 songs, log₂(256!) ≈ 1,684. Eight rounds of Swiss on 256 songs is only 8 × 128 = 1,024 comparisons — short of that floor before you even count that Swiss also wastes comparisons re-pairing songs whose relative order a smarter system would already treat as settled.

SongRank now uses an adaptive pairwise ranking engine instead. Every song carries a rating and an uncertainty score, both updated after each vote — a surprising result (an underdog winning) moves a rating more than an expected one, and a song's uncertainty shrinks the more it plays. Rather than following a fixed bracket or round schedule, every single matchup is chosen fresh: whichever pair currently has the closest ratings and the most combined uncertainty, because that is the comparison whose answer teaches the engine the most. A blowout between the best and worst songs on your list would confirm what the ratings already predict; two closely-matched songs are where a vote actually moves the needle.

How many matchups, and when it stops

On the build step you pick a depth — Quick, Balanced, or Thorough (the default), each shown with its estimated matchup count before you commit. The target scales with your list size as roughly 1.25 × n × log₂n matchups at Thorough — 36 for 8 songs, 486 for 64, 2,566 for the full 256. A list of 6 songs or fewer just plays every pair once, which is exact and cheaper than being clever about it.

You are never required to reach that target. The ranking is valid after any number of votes — stop whenever you want and you still get a complete, ordered list, it just gets more confident the longer you play. A progress readout shows what fraction of all the pairs in your list the engine can already order confidently — read it as how much of the final answer is decided if you stop right now. A long list finishes short of 100%, which is honest: a few thousand comparisons genuinely cannot pin down 200 songs to the last adjacent pair. A small or lopsided list, on the other hand, can finish well before its estimated matchup count, once every neighbour in the ranking is clearly separated. A close, evenly-matched list tends to use its full budget, because that is exactly when more votes keep being informative.

Every one of those matchups is chosen the same way, right to the end. A ranking used to finish with a short playoff between the leading few songs, and that was dropped: six matchups among four songs settles a podium, but ordering the whole chart is the job, and those same six comparisons do more good spread across the list. Measured over a hundred simulated 60-song rankings, moving them back into the main phase left the whole-chart ordering slightly more accurate, and every quarter of the list level or better.

Where the audio comes from

Clips come from the iTunes Search API, which is free and public and returns a 30-second preview for most tracks. SongRank plays the whole preview — there is no more audio to be had, so there is nothing to gain by playing less of it. A progress bar and an elapsed/total readout show how far through you are, and you can replay a clip as often as you like before deciding.

Playback level is evened out between songs, because mastering loudness varies enormously between eras and a quieter recording otherwise loses votes for reasons that have nothing to do with the song.

Some tracks have no preview available. Those songs still appear and are still fully votable — you just have to go from memory.

Accounts are optional — saving isn't

Building a list and voting through it works fully without signing in. But saving and resuming a ranking is a signed-in feature: without an account, your progress lives only in the current browser tab, and closing or refreshing it loses your place. A Thorough ranking of a large list is thousands of matchups, so that is worth knowing before you start, not after — the build page says so up front if you are not signed in.

Signing in with Google adds saved history: a ranking follows you between devices and you can come back to a finished ranking later. See the privacy page for exactly what gets stored.