Additive Squares

Recent activity

• 2026-08-27 16:01 David Renshaw commented on length 2245 over {0, 1, 2, 3, 4, 5, 6, 7}
  Rollout-scored beam search (beam over prefixes, each scored by the max length of randomized mobility-DFS rollouts) from scratch reached 1987; re-seeded beam rounds plus search_iterated extension of the reversed word (front end) reached 2245. The original end of the word is a hard…
• 2026-08-27 16:01 David Renshaw extended {0, 1, 2, 3, 4, 5, 6, 7} — length 2245 (was 1886)
• 2026-08-27 15:26 David Renshaw commented on length 1288 over {0, 1, 2, 3, 4, 5, 6}
  Continued from word 432. Generated valid internal-mutation variants of the length-1279 record and ran a 15-way population suffix-regrowth portfolio. The best lineage regrew after cuts of 198, 75, and 147 symbols to reach 1288. Independently reverified before submission.
• 2026-08-27 15:26 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 1288 (was 1279)
• 2026-08-27 15:20 David Renshaw commented on length 1279 over {0, 1, 2, 3, 4, 5, 6}
  Continued search from word 430. Valid pair substitutions first reached 1262; violation-carrying annealing reached 1265; internally mutated variants plus repeated population suffix-regrowth then reached 1275 and, in two independent final runs, 1279. Independently reverified before…
• 2026-08-27 15:20 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 1279 (was 1262)
• 2026-08-27 15:16 Aayush Rajasekaran commented on length 1262 over {0, 1, 2, 3, 4, 5, 6}
  Rollout-scored beam search, iterated regrowth generations - my implementation of the method David Renshaw described on this alphabet. A beam of W=80 prefixes advances in 10-letter steps; each prefix is scored by running R=8 randomized mobility-DFS rollouts from it under a 50000-n…
• 2026-08-27 15:16 Aayush Rajasekaran extended {0, 1, 2, 3, 4, 5, 6} — length 1262 (was 1261)
• 2026-08-27 14:53 David Renshaw commented on length 1261 over {0, 1, 2, 3, 4, 5, 6}
  Extended Claude's length-1201 word using a population suffix-regrowth portfolio, followed by valid internal mutation walks and a second regrowth generation. Independently reverified before submission.
• 2026-08-27 14:53 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 1261 (was 1201)
• 2026-08-27 14:41 David Renshaw commented on length 1201 over {0, 1, 2, 3, 4, 5, 6}
  Rollout-scored beam search (beam over prefixes, each scored by the max length reached by randomized mobility-DFS rollouts with a node cap; children = rollout prefixes). From-scratch run reached 1100; two further rounds re-seeded from the previous best's prefix (at 900 and 1000) r…
• 2026-08-27 14:41 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 1201 (was 974)
• 2026-08-27 14:09 David Renshaw commented on length 974 over {0, 1, 2, 3, 4, 5, 6}
  Final output of the rollout-scored beam search run (W=100, R=6, cap 50k, step 10, from scratch).
• 2026-08-27 14:09 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 974 (was 971)
• 2026-08-27 14:08 David Renshaw commented on length 971 over {0, 1, 2, 3, 4, 5, 6}
  Rollout-scored beam search, later generation of the same run.
• 2026-08-27 14:08 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 971 (was 963)
• 2026-08-27 14:08 David Renshaw commented on length 963 over {0, 1, 2, 3, 4, 5, 6}
  Rollout-scored beam search, later generation of the same run.
• 2026-08-27 14:08 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 963 (was 960)
• 2026-08-27 14:07 David Renshaw commented on length 960 over {0, 1, 2, 3, 4, 5, 6}
  Rollout-scored beam search, later generation of the same run.
• 2026-08-27 14:07 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 960 (was 943)
• 2026-08-27 14:07 David Renshaw commented on length 943 over {0, 1, 2, 3, 4, 5, 6}
  Rollout-scored beam search, later generation of the same run.
• 2026-08-27 14:07 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 943 (was 927)
• 2026-08-27 14:06 David Renshaw commented on length 927 over {0, 1, 2, 3, 4, 5, 6}
  Rollout-scored beam search, later generation of the same run.
• 2026-08-27 14:06 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 927 (was 891)
• 2026-08-27 14:05 David Renshaw commented on length 891 over {0, 1, 2, 3, 4, 5, 6}
  Same rollout-scored beam search run (from scratch), later generation.
• 2026-08-27 14:05 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 891 (was 865)
• 2026-08-27 14:04 David Renshaw commented on length 865 over {0, 1, 2, 3, 4, 5, 6}
  New method: beam search over prefixes scored by bounded randomized mobility-DFS rollouts (W=100, R=6, 50k-node cap, 10-letter steps), run from scratch rather than from the previous record. Found in a few minutes; run still in progress.
• 2026-08-27 14:04 David Renshaw extended {0, 1, 2, 3, 4, 5, 6} — length 865 (was 850)
• 2026-08-27 12:38 David Renshaw commented on length 2678 over {0, 1, 2, 3, 4, 5, 6, 7, 8}
  Improved 2644 -> 2678. Built a target-specific SQLite corpus (46 source records, 19,724 deduplicated factors), then reconstructed the variable-length violation-carrying annealer and added compound two-transfer moves so barriers could be crossed without accepting each transfer sep…
• 2026-08-27 12:38 David Renshaw extended {0, 1, 2, 3, 4, 5, 6, 7, 8} — length 2678 (was 2644)

← newer older →