AdaptiveOrder

AdaptiveOrder

AdaptiveOrder (-o 14) is GraphBrew's runtime selection boundary. The validated path is a frozen deterministic rule, not a machine-learning model.

./bench/bin/pr -f graph.sg -s \
  -o '14:_:_:_:allkernel-lowreuse-rule:best-endtoend:<reuse>' \
  -n 3

<reuse> must be 1 or 2.

Figure 1. AdaptiveOrder has no intrinsic permutation. The strip shows the output of one selected GraphBrew arm on the shared catalog input; a different feature decision can select a different arm.

Runtime contract

The rule:

• samples structure from the new graph;
• models the selected kernel's property footprint relative to LLC;
• uses kernel identity and declared reuse;
• evaluates one frozen predicate; and
• chooses the promoted GraphBrew composition or Boost Rabbit.

It does not run candidate orderings first, train at runtime, use graph names, or query prior benchmark rows. The branch decision is deterministic; a selected cd_parallel GraphBrew mapping can still be schedule-sensitive.

Unsupported kernels, small graphs, and reuse above two use the fallback.

Timing

Public portfolio evidence accounts for chosen mapping cost plus reused kernel time. The binary reports Adaptive Feature Time separately; fully deployed timing must include it.

For the predicate, supported kernels, exact arms, graph examples, confidence intervals, and limitations, use All-Kernel Low-Reuse Selector.

Legacy model modes remain only for offline compatibility.