Benchmark Wordle solvers with Jev word priors

01 · Role in the system

What Jev does here

Solving Wordle well needs more than entropy - you need to know which words are actually likely answers. Jev provides that prior: one typed scoring call rates how answer-like each of the 12,972 accepted words is, and the prior feeds plain information-theoretic search over the full 1,925-day history of real NYT answers. The methodology is the story: code, pinned raw inputs, per-day results, null results and verification are all archived, with a single bash script reproducing the follow-up. The README even corrects three of its own earlier shorthand claims (T0 is a heuristic baseline, not proven optimal; a category feature was mislabeled) - null results and corrections published, which almost no demo does.

02 · Control boundary

Where Jev sits

Jev as a calibrated word prior inside an information-theoretic solver experiment: one scored call over the accepted-word list feeds entropy search, and the whole study is archived with pinned inputs, per-day results, null results and a reproduce script.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• The Jev prior is one ingredient; the README itself documents which baseline claims were corrected.
• English Wordle only; the 12,972-word list is NYT-specific.
• Created 2026-09-30, 0 stars; single-author experiment.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• Atagoxx ↗Community · github · public · checked 2026-10-02