Skip to content

Port fidelity: seven differences from a record engine, and a five-clue regression

Our port of a record-style backtracker differed from the published engine in seven places; all seven were fixed and tested, and the node sequences match the earlier port wherever the legacy switches are on. With all five clues the faithful defaults do worse than the old ones (deepest depth 18 to 54 against 33 to 215 on eight seeds), so the fix is not an improvement for that setting.

by Raphaël Anjou

experimentmeasurementmeasuredmatched edgesBacktrackingGo fasterUpdated 2026-11-14
Reproduceseeded: reproduces with the given seedreruns the searchBudget: node-capped runs (10^7 to 10^8 nodes, one restart); about 1.5 to 3.8 s of user time per 10^8 nodes on the loaded host
Hardware & run
Native runCPU only
0.0011core-hours
Cores
1
RAM
16 GiB
GPUs
0
CPU
Apple M1
Machine
8-core Apple M1 laptop, shared with another job during the re-checks, up to 7 single-thread runs at once
Budget
node-capped runs (10^7 to 10^8 nodes, one restart); about 1.5 to 3.8 s of user time per 10^8 nodes on the loaded host
Start
seeded; no-clue seeds 101 to 108, five-clue seeds 501 to 508, centre-only and short-run seeds 1 to 6
Reproduce this result

seeded: reproduces with the given seed

just experiments single-core-benchmark

The recipe reruns the benchmark grid, whose Blackwood-style row now measures the faithful engine; it does not rerun the fidelity comparisons, whose raw outputs and scripts are committed under the experiment's fidelity-results folder.

Keep exploring

Referenced by

Page sourceView as Markdown