Skip to content

LADDER

Throw hundreds of cheap short searches at the board, keep only the deepest starts, and promote the survivors through longer and longer rounds.

by Raphaël Anjou

experimentsolvermeasurednew basin451/480matched edgesLocal searchUpdated 2026-07-21
Reproduceseeded — reproduces with the given seedreruns the search (See below)Budget: ~300 s × 8 finishers (exploratory run, not the standardized single-core bench)
Pipeline
  1. 1
    restart tournament

    Flood hundreds of short probes from random seeds, rank by perfect-prefix depth, ratchet-pin the deepest and probe beyond it

  2. 2
    exact tail

    From the deepest banked prefixes, a 14-cell exact finisher converts an opening into a full board

Complexity
Time
round 1: hundreds of 5 s probes; finish: 300 s × 8 exact-tail-14 runs from the deepest prefixes
Space
keeps only the banked prefixes and their live continuations between rounds

Successive halving: total effort is roughly constant per round, but each round spends it on fewer, deeper survivors, so the cost concentrates where it pays.

Hardware & run
Native runCPU only
0.67core-hours
Cores
8
RAM
16 GiB
GPUs
0
CPU
Apple M1
Machine
MacBook (Apple M1, 8 cores)
Budget
~300 s × 8 finishers (exploratory run, not the standardized single-core bench)
Start
stochastic; see the run's repro command

Reproduce this result: seeded — reproduces with the given seed

Keep exploring

Referenced by

Page sourceView as Markdown