What to guess second, and the rule that decides it
6 min read
The opening-guess post ended by admitting it had done the easy half. Guess one has a single right answer you can memorise. Guess two has 17 right answers, one for each tally the opener could have come back with, and no amount of memorising will cover them.
What you can carry instead is a rule for choosing. So we computed the whole second move: open with SLATE on the regular board, split the 646 answers by what the tally would have been, and for each of those 17 branches find the sharpest possible follow-up. Here is what it says.
First, the good news
| Board | Answers | After guess one | After guess two |
|---|---|---|---|
| Regular (SLATE) | 646 | 74.8 | 8.8 |
| Easy (RATE) | 600 | 93.44 | 15.26 |
| Hard (SEATER) | 1117 | 106 | 9.86 |
The opener looks feeble — 646 down to 74.8 is a 88% cut and still leaves you staring at seventy words. Guess two cuts what is left by another 88%, down to 8.8.
That is the shape of this game and it is worth internalising, because it is the opposite of how it feels. Guess one feels wasted. It is not: it is what makes guess two enormous. The run is decided in the middle, not at the start.
The real question: probe or answer?
At guess two you face a choice that does not exist at guess one. You can play a word that could still be the answer — and win outright if you are lucky — or a pure probe, a word you know is wrong, chosen only because it splits the field better.
Everyone has an instinct here and the instincts disagree. So, measured across every branch, weighted by how often each branch actually happens:
| Board | Best probe leaves | Best possible answer leaves | Probe's edge |
|---|---|---|---|
| Regular | 8.8 | 9.47 | 8% |
| Easy | 15.26 | 16.1 | 6% |
| Hard | 9.86 | 10.39 | 5% |
The probe wins, and by almost nothing. On the regular board it leaves 8.8 candidates against 9.47 — a difference of 0.67 of a word. And the word that could still be the answer carries something the probe cannot: a chance of simply ending the run right there.
The exception, which is worth memorising
Averages hide things. Break the same data out by branch and one group behaves completely differently: the outcomes with several hits and no nears.
| Opener said | Left | Best probe | Leaves | Best live answer | Leaves |
|---|---|---|---|---|---|
| 2 hit, 0 near | 51 | CRAPS | 5.67 | SPARK | 7.86 |
| 3 hit, 0 near | 26 | CHAMS | 3.62 | SPARE | 5.69 |
| 0 hit, 2 near | 100 | MOREL | 11.86 | MERIT | 12.54 |
| 0 hit, 1 near | 116 | BATON | 14.84 | COUNT | 15.74 |
After 3 hit, 0 near the probe leaves 3.62 and the best live answer leaves 5.69 — the probe is 57% better, which is nothing like the 8% average.
The reason is clean once you see it. Three hits and no nears means three of SLATE’s letters are sitting in the right places and the other two are absent entirely. Every surviving answer therefore shares those three letters in those positions — so guessing one of them re-tests three letters you have already nailed down and asks about only two new ones. A probe made of fresh letters asks about five.
- Lots of hits, no nears — the survivors all look alike. Spend the guess on a probe with new letters.
- Nears but few hits — the survivors are spread out and a live answer divides them about as well as anything. Take the free win chance.
What actually happens most days
Worth knowing where you will usually find yourself. The most common reply to SLATE is 0 hit, 1 near, on 17.96% of days, and it leaves 116 of 646 answers standing — the most likely outcome is close to the least informative one.
| Opener said | Share of days | Answers left | Best follow-up |
|---|---|---|---|
| 0 hit, 1 near | 17.96% | 116 | BATON |
| 0 hit, 2 near | 15.48% | 100 | MOREL |
| 1 hit, 1 near | 14.71% | 95 | SPORT |
| 1 hit, 0 near | 12.23% | 79 | GRIDE |
| 1 hit, 2 near | 8.36% | 54 | PARLE |
| 2 hit, 0 near | 7.89% | 51 | CRAPS |
Notice that the top six branches account for roughly three-quarters of all puzzles, and that none of them leaves you with fewer than fifty candidates. If you have ever felt that you were playing badly at this point — you were not. This is where everyone is.
The rules, short
- Play a word that could still be the answer. It costs you 0.67 of a candidate and might just win.
- Unless you got several hits and no nears. Then the survivors are near duplicates of each other; probe with fresh letters instead.
- Do not chase the hits. A high hit count feels like progress and is often the branch that leaves you the most work — on the easy board, 2 hit, 0 near leaves 59 answers but the best follow-up still leaves 13.61, worse than branches that started with more candidates.
- Expect guess two to do the heavy lifting. Seventy candidates becoming nine is the single biggest step in the run.
After guess two
We stopped here on purpose. Beyond guess three the branches multiply past the point where a table helps anyone, and the honest advice becomes the boring advice: keep a written note of what each tally ruled out, and use the scratchpad rather than your memory. The game gives you one because holding five simultaneous constraints that name no letters is genuinely beyond most people, including the people who built it.