[na] Let transfer weight change the chosen schedule

[arniber21]

Summary

Next slice of issue 484. Stage, movement, and idle weights stay fixed and positive. trap_transfer_weight then changes which verified one-gate orientation is chosen. A transfer weight below the one-atom move's extra movement and idle cost keeps the two-atom move (4 transfers, shorter move). A transfer weight above that cost keeps the one-atom move (2 transfers). Each verified total is lower under the weights that chose it.

Issue 484 stays open.

What changed

• The one-gate orientation fixture already scores transfer weight. This test holds the other three weights fixed, so the flip is the transfer weight crossing that cost rather than movement weight turning off.

Still open on issue 484

• Document cost_model fields, units, the canonical formula, and schema versioning.
• Movement weight and idle weight still lack a held-fixed proof of the same kind. Their existing test turns the competing weights off.
• Error-budget mode and the placeholder time objective still do not read arbitrary target weights.
• Diagnostics for unsupported or inconsistent weights beyond non-finite and negative values.
• Stats echoing the effective objective.
• A documented migration for time-only and error-budget CLI modes.
• Tests that omitted optional weights keep the documented default.

Test

cargo test -p quon_na --features mlir --lib -- trap_transfer_weight_selects

  

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
quon	Ready	Preview	Oct 1, 2026 7:27am UTC