The best opening guess in Wordtallyo, measured
6 min read
Every word game collects this question, and most of the answers are folklore — a word someone liked, repeated until it sounded official. This one has an answer you can check, because the board is small enough to compute exhaustively. The regular board has 646 possible answers and 8,636 legal guesses. Scoring every guess against every answer is 5.6 million comparisons, which takes a few seconds.
So we did it, for all three boards. Here is what came back, plus one genuinely surprising thing that fell out on the way.
What “best” means here
A guess is not good because it feels clever. It is good because of how finely it cuts the field. Before you guess, any of the 646 answers is possible. After you guess you get a tally — so many letters in the right place, so many in the word somewhere else — and every answer that would have produced a different tally is gone.
The answers that would have produced the same tally are what you are left with. So the measure of an opener is simply how many answers are still standing afterwards, on average. Lower is sharper. It is the same number the game’s own Assist mode uses to score a word you are already considering, so the blog and the game agree by construction rather than by coincidence.
The answers
Regular board, five letters, 646 answers to cut into:
| Opener | Answers left | Bits | Outcomes |
|---|---|---|---|
| PRASE | 74.37 | 3.37 | 17 |
| SLATE | 74.8 | 3.36 | 17 |
| CRANE | 75.57 | 3.31 | 16 |
| BLARE | 75.73 | 3.33 | 15 |
| BRACE | 76.55 | 3.30 | 16 |
| PRATE | 76.66 | 3.32 | 15 |
| STARE | 77.18 | 3.38 | 16 |
| FLARE | 77.7 | 3.28 | 16 |
Easy board, four letters, 600 answers:
| Opener | Answers left | Bits | Outcomes |
|---|---|---|---|
| RATE | 93.44 | 2.88 | 12 |
| TARE | 93.8 | 2.87 | 12 |
| RALE | 94.3 | 2.85 | 11 |
| LATE | 95.43 | 2.87 | 12 |
| TALE | 95.85 | 2.86 | 12 |
| RASE | 98.09 | 2.82 | 10 |
Hard board, six letters, 1117 answers:
| Opener | Answers left | Bits | Outcomes |
|---|---|---|---|
| SEATER | 106 | 3.67 | 20 |
| CERATE | 110.32 | 3.58 | 21 |
| CAREER | 110.97 | 3.53 | 19 |
| CARTER | 111.25 | 3.61 | 21 |
| RETINE | 111.32 | 3.63 | 22 |
| SERINE | 112.02 | 3.60 | 21 |
If you want one word to memorise and then stop thinking about it: SLATE on five letters. It sits within half a word of the top of the table, and unlike PRASE — a mineral almost nobody has heard of — you will still remember it tomorrow.
The part nobody expects: anagrams are not interchangeable
You are never told which letters landed. It follows, most people assume, that the order you type them in cannot possibly matter. Same five letters, same information.
That is wrong, and not by a little.
| Same letters, different order | Answers left |
|---|---|
| SLATE | 74.8 |
| STALE | 79.5 |
| LEAST | 88.88 |
| SETAL | 102.92 |
| TALES | 107.33 |
| STEAL | 109.46 |
The reason is the first of the two numbers. A tally reports how many letters are in the right place — and “the right place” is a fact about position, even when the position itself is never named. Put an S where answers rarely start and the hit count barely moves. Put it where answers often end and the same letter starts pulling the pool apart.
So the arrangement is doing real work. You are not told which positions were right, but you choose which positions get asked about. PRASE beats ASPER by sixty per cent on exactly this, holding an identical bag of letters.
What a bad opener looks like
The floor is more instructive than it sounds. The worst openers are JUJU, BUZZ, FUZZ on four letters and MUUMUU, BOOBOO, HUBBUB on six. Every one of them spends its letters twice.
MUUMUU asks about three distinct letters across six slots and leaves 552.61 of 1117 answers standing. It is barely a guess. The rule that falls out of this: no repeated letters in your opener, and prefer letters that are common without being universal. A letter that appears in every answer eliminates nothing. Neither does one that appears in none.
Two honest caveats
- The two metrics disagree slightly. The tables rank by expected answers remaining, but information theory would rank by bits, and the orders are not identical. STARE yields more bits (3.379) than PRASE (3.369), yet leaves more answers standing. Bits reward an even spread across outcomes; expected-remaining punishes one oversized bucket. For choosing an opener the difference is noise — anything in these tables is a fine guess — but quoting one number as the answer would be overselling it.
- This covers guess one and nothing else. It says nothing about guess two, which is where the game actually gets hard, because by then you are intersecting two constraints that each name no letters at all. A follow-up owes you that.
Why every number here looks so bad
If you have played a game with coloured tiles, the figures above should look broken. A good opener there leaves you with a dozen candidates. Here, the sharpest five-letter opener in the entire dictionary leaves 74.37.
That gap is the whole game. A tally can come back only 17 distinct ways on this board. A grid of coloured tiles can come back 104. Fewer distinct replies means less information per guess, which is why these boards hand you more guesses to spend.
We measured that gap properly, and it is wider than we expected — that is the next post.
The short version
- Five letters: open with SLATE. Four: RATE. Six: SEATER, or CASTER if you would rather guess a real word.
- Never repeat a letter in your opener.
- The arrangement matters even though positions are never revealed — worth up to 46% of the field.
- Do not expect to feel close after guess one. Seventy candidates still standing is what a good opener looks like here.