Siddhartha Chatterjee on Darmok’s championship cryptic crossword system

Siddhartha Chatterjee on Darmok’s championship cryptic crossword system

In the arXiv abstract, Siddhartha Chatterjee presents Darmok, an automated system for British cryptic crosswords, where clues combine a disguised definition and wordplay. Chatterjee says Darmok solved all seven puzzles at the 55th Times Crossword Championship without errors and beat every human competitor’s time in every round. The paper’s central argument is that strong average performance is not enough for a dependable system: cloud API calls can stall, time out, refuse requests or loop, while tournament scoring punishes even one wrong or blank square. Darmok cross-checks clues with three independent models, uses verified crossing letters to help solve harder clues, then uses a whole-puzzle model and dictionary lookups for remaining gaps without overwriting confirmed words. Chatterjee reports near-zero-error performance over five months of daily evaluation, and says backup paths carried the solve on the worst operating days and during a severe slowdown of the primary model in the final. The author argues that as foundation models handle average cases, resilience in the worst case becomes a central challenge in building complex AI systems. The abstract reports these results; it does not establish how well the design transfers beyond crossword solving.

Publication: 6362b275-ebfb-4d8a-8096-900a56e2237e

attribution — pass: The reading attributes Darmok, its tournament results and its resilience argument to Siddhartha Chatterjee, the sole listed author, in an arXiv abstract. It keeps the results as the paper's reported claims rather than established facts.

Authors:Siddhartha Chatterjee (The University of Texas at Austin)

summary — pass: The summary leads with the actor and the development, and explains cryptic clues and the strict tournament scoring in plain language. It then gives the author's core thesis: average-case strength is not enough, and backup paths matter in worst cases. One optional fidelity improvement: add the abstract's point that removing backups barely changes accuracy under normal benchmarks, so the design looks over-engineered until failures occur. The omission does not misstate the argument. The question names a real open gap.

removing backup models, fallback steps, or elaborate prompt instructions barely changes accuracy, making the architecture appear over-engineered

date — pass: originalPublished 2026-11-10 matches the arXiv v1 submission date. The feed date of Nov 11 is correctly not used.

[Submitted on 10 Nov 2026]

support — pass: Each claim is directly supported by the abstract. This covers the seven error-free puzzles and faster times than every human in every round, the cross-checking, crossing-letter and fallback design, the five months of near-zero error, and backup paths carrying solves on the worst days and during the final slowdown. The limitations correctly note that the abstract does not quantify backup usage and does not establish transfer beyond crosswords. No benefits or costs are invented.

during both the worst days of daily operation and a severe slowdown of our primary model in the live Championship Final, the backup paths carried the solve

media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the most direct way to access the source.