Methodology

Exactly how ratings, experience, and coins are calculated.

Rating

Ranked play uses the Elo system, devised by Arpad Elo for chess and used by rating bodies since the 1960s. The idea is that a rating is a prediction: the gap between two ratings says how likely each player is to win. Play to that prediction and your rating holds. Beat it and your rating rises by the amount the result was a surprise.

After a ranked match ends, the server computes an expected score for each player and adjusts both ratings by the difference between what happened and what was expected.

expected  E = 1 / (1 + 10 ^ ((opponent rating − your rating) / 400))

  change  Δ = round(K × (S − E))

           S = 1 for a win · 0.5 for a draw · 0 for a loss

A 400-point gap means the stronger player is expected to score about ten times as often as the weaker one. Because E is subtracted from what you actually scored, beating a stronger opponent moves your rating further than beating a weaker one, and losing to a weaker opponent costs more than losing to a stronger one. A draw between equals moves nothing.

Where ratings tend to settle. Everyone begins at 1200 and spreads out from there — winning against the prediction moves you right, losing to it moves you left. This curve is drawn from the shape the math produces, not from real games.

The constants

Value	What it is
1200	The rating every player starts on in a pool they have never played.
K = 40	Used while a rating is provisional — your first ten games in that pool. Ratings move quickly at first so a new player reaches roughly the right level fast.
K = 24	Used from the tenth game onward, once the rating has settled. One result can then move it by at most 24 points.
10 games	The point at which a rating stops being labeled provisional. The two players in a match are counted separately, so one can still be provisional while the other is not.

What that works out to

Situation	Result	K	Change
You are rated 1200, your opponent 1200	Win	24	+12
You are rated 1200, your opponent 1200	Draw	24	0
You are rated 1200, your opponent 1200	Loss	24	−12
You are rated 1200, your opponent 1400	Win	24	+18
You are rated 1200, your opponent 1400	Loss	24	−6
You are rated 1400, your opponent 1200	Win	24	+6
You are rated 1400, your opponent 1200	Loss	24	−18
Your first nine games, evenly matched	Win	40	+20

A few notes

There is no floor and no ceiling. A rating is not clamped at either end, and no result is capped beyond what K already limits it to.
A match is not always zero-sum. When one player is provisional and the other is not, they are on different K values, so the points one gains need not equal the points the other loses. Nothing is created or destroyed deliberately; it falls out of letting new players find their level faster.
Every rating change is written to a permanent ledger recording the rating before, the rating after, the change, and the expected score that produced it. Settlement runs once per match and is keyed so that a repeat cannot double-count.

Why there are forty separate ratings

A rating only means something when everyone holding it played the same game. A forty-five minute match and a one-minute match ask for different skills, and a rating that mixed them would describe neither. So each combination keeps its own rating, entirely independent of the others.

There are ten clocks, two modes — OG and GO — and Hard Mode on or off. That is forty ranked Practice pools, plus a separate pool for each of the two ranked Daily modes. Your rating in one says nothing about your rating in another, and you can be provisional in one while established in another.

Clock	Budget	Applies to
Untimed	No clock at all	Whole match
1 minute	60,000 ms	Whole match
3 minutes	180,000 ms	Whole match
5 minutes	300,000 ms	Whole match
10 minutes	600,000 ms	Whole match
20 minutes	1,200,000 ms	Whole match
45 minutes	2,700,000 ms	Whole match
1 day	86,400,000 ms	Every move
3 days	259,200,000 ms	Every move
7 days	604,800,000 ms	Every move

The first seven are a budget for the whole match: run out and you lose on time. The last three are a budget for each move, which refills every turn — those are the correspondence games, played over days.

Everything else about a ranked match is fixed, so that the pool name describes the game completely: every ranked match is five letters, expert difficulty, and a ranked GO match is always five puzzles. Unranked and private matches keep every option; it is only ratings that need the comparison to be fair.

How a match is decided

Ratings move on the result, and the result comes from points. Each guess scores on the evidence it produced, and solving the word is worth far more than the letters that led there — so a player who solves in three guesses beats a player who almost solves in six.

For	Points
Each letter in the right place	5
Each letter in the word but the wrong place	2
Each letter not in the word	0
Solving the word	100
Each guess you did not need, when you solve	10
Solving in Hard Mode	15

In a GO chain the attempts tighten as you go: six on the first puzzle, then five, four, three, and two from the fifth puzzle onward. Fewer attempts also means more unused ones are unavailable, so later puzzles are worth less by way of the solve bonus, not by a separate rule.

When a match settles, the server does not trust the running total it was shown. It recomputes every move’s points from the recorded evidence, checks the declared winner against the totals it derived, and refuses to settle at all if the two disagree. A match that cannot be verified does not move anyone’s rating.

How the leaderboard is ordered

Within a pool, players are ranked by rating, highest first. Where ratings are equal the order is decided by games played, then by peak rating, then by which rating changed most recently, and finally by profile identifier — so the order is always definite and never reshuffles between visits.

Peak rating is the true historical high, read from the rating ledger rather than from the current value. To appear at all you need at least one settled game in that pool and a public profile. Provisional players are listed and labeled rather than hidden.

Experience and levels

Experience comes from Solo games — Practice and Daily.

Outcome	Experience
Win	40, plus 20 for each puzzle solved, plus a speed bonus of 10 minus the number of guesses you made — nothing if you made ten or more.
Loss	10 for each puzzle you solved before the game ended.

Levels are a square curve: level n begins at 100 × (n−1)². So level 2 arrives at 100 experience, level 3 at 400, level 4 at 900, level 5 at 1,600. Each level takes longer than the last, deliberately — the number should keep meaning something after a hundred games.

COMBAT awards no experience. Competitive play is measured by rating, and paying twice for the same match would let one mode inflate a number meant to describe another.

Coins

Coins also come from Solo games.

Outcome	Coins
Win	8, plus 4 for each puzzle solved.
Loss	1 for each puzzle solved, up to 4 — so a long chain that ends badly is still worth something, but not much.

What they buy:

Item	Cost
Reveal one letter	25
Remove five incorrect letters	40
Unlock a past Daily	60
Another guess	Varies — see below

A balance can never go below zero, and no single transaction may exceed 10,000 coins.

The cost of another guess

Buying another guess is the one price that is calculated rather than fixed. Two things move it: how close you already are, and how many extra guesses you have already bought in this game.

H = half the word length, rounded up
C = how much of H your best row has already earned
n = extra guesses already bought in this game

cost = (H − C + 3) × (n + 1)          minimum 1

Being closer makes the next guess cheaper, because C rises with the proportion of the word you have placed correctly — measured on your best row for the puzzle you are currently on, not across the whole chain. Buying repeatedly makes it steeply more expensive: the second extra guess costs twice the base, the third three times. For a five-letter word:

Letters placed	Progress	1st extra	2nd	3rd
0	0%	6	12	18
1	20%	6	12	18
2	40%	5	10	15
3	60%	5	10	15
4	80%	4	8	12

Longer words start dearer, because H is larger: a ten-letter word with nothing placed costs 8 for the first extra guess where a five-letter word costs 6.

The daily streak

Your streak counts consecutive days on which you finished a Daily. The rules are short enough to state completely:

Either Daily counts. The OG and the GO are two ways to keep the same streak, not two streaks. Finishing both on one day counts once.
Finishing counts, not winning. A Daily you lose keeps the streak. Showing up is the thing being measured.
The day is your local day, the same one the Daily calendar uses, so your streak turns over at your midnight rather than anybody else’s.
Miss a day and it lapses. Your next finished Daily starts a new streak at one. The panel shows the streak as zero from the moment a day has gone by without one.
Unlocking a past Daily cannot repair it. Playing an older date records the result and pays the usual coins and experience, but the streak only ever moves forward. A streak you have lost cannot be bought back.

What’s checked and where

Some of this is enforced by the server, and some is calculated by the app on your device.

Server

Ratings: Points are recomputed from the recorded evidence, the winner is checked against them, and the rating change is written in the same transaction as the result.
The price of a reveal, a removal, and your balance: The server knows the two prices and refuses a purchase you cannot afford.
Unlocking a past Daily: The server holds the price, takes the coins, and records the unlock in a single step. Neither half can happen without the other, and the record of what you have unlocked is kept where the app cannot write it.
The coins a finished game pays: The server reads the game you just recorded and works out what it is worth, rather than being told an amount. Each game pays once.

App

Experience earned: Applied through an operation that is keyed so it cannot be applied twice.
The daily streak: Computed on your device from the date of the Daily you finished and written to the same saved record as your experience. It is keyed to the day, which is what makes finishing both Dailies count once.
The price of another guess: The server checks only that you have the coins. The formula above is what the game charges, but it is the app that applies it. This one stays here on purpose: the price depends on the state of a Solo game, and Solo games are played on your device rather than on the server.
What a game records about itself: Your history is written by the app, so the coins a game pays are bounded by what that record says. A game cannot pay more than the most generous real game would.