Best 6 Letter Starter Words: Measured Results

For this site’s current pools, SATIRE ranks first by entropy among accepted six-letter guesses with no repeated letters. SORTED ranks second, but it performs better on two other risk measures: its expected remaining candidate count is 7.932551 rather than SATIRE’s 8.052786, and its worst feedback bucket contains 24 answers rather than 34.

That tradeoff is the useful answer. There is no metric-free “best” starter. SATIRE produces the highest entropy in this experiment; SORTED leaves a slightly smaller candidate set on average and caps the worst branch more tightly. Neither result is claimed to be official or universal.

What the experiment measures

The generator evaluates all 1,485 accepted guesses against each of the 682 possible answers in this site’s fixed pools. For every guess-answer pair, it computes six feedback tiles. Exact-position matches are consumed first; misplaced matches are consumed second, so a repeated letter cannot be credited twice.

Each guess partitions the 682 answers into feedback buckets. A bucket contains all answers that would produce the same color pattern for that guess. The experiment records:

• Entropy: how evenly the answers are split across feedback patterns. Higher is better under this objective.
• Expected remaining candidates: the average bucket size after weighting each possible outcome by its probability. Lower is better.
• Worst-case bucket: the largest branch the guess can leave. Lower is better for a conservative strategy.
• Distinct feedback patterns: how many different outcomes occur in the 682-answer run.

The exact formulas, data fingerprints, tie-break order, and two-pass feedback version are published in the word-list methodology. The JSON table above is generated at build time rather than copied into Markdown, so a changed experiment cannot leave an old Top 10 visible unnoticed.

SATIRE versus SORTED

SATIRE has 7.065155 bits of entropy, 8.052786 expected remaining candidates, and a worst-case bucket of 34. SORTED has 6.992969 bits, 7.932551 expected remaining candidates, and a worst-case bucket of 24.

Choose SATIRE if this experiment’s entropy objective is the priority. Choose SORTED if the slightly lower mean and substantially lower worst branch matter more. SORTED is also in this site’s answer pool; SATIRE is an accepted guess but not an answer candidate here. That distinction affects whether a first guess can solve immediately, although this experiment does not assign frequency or prior probability to answers.

Accepted guesses are not the answer pool

The broad accepted-guess pool contains words the game can allow as input. The smaller answer pool contains the candidates this analysis treats as possible targets. Every accepted guess is useful to evaluate, even when it cannot be an answer, because a probe can still divide the answer set efficiently.

The editorial table focuses on the 774 accepted guesses with six distinct letters. Full-pool ranks remain in the generated JSON, including repeated-letter guesses, but the primary table keeps the first-guess comparison on six newly tested letters. This is a declared scope choice, not evidence that repeated letters are always wrong.

Boundaries of the result

The analysis assumes every one of the 682 answers is equally likely. It does not use English corpus frequency, a hidden official schedule, player history, or today’s puzzle. Changing the answer pool, accepted list, feedback rules, priors, or optimization metric can change the order.

Use the candidate solver after the first feedback arrives. If duplicate colors are confusing, the repeated-letter strategy shows exactly how green and yellow tiles consume the available letter count.