Skip to content

hex-roots: Phase-4 bench re-scaffolding to clear the inconclusive verdicts #8750

Description

@kim-em

reports/hexroots-performance.md (PR #8749) records 9 inconclusive verdicts, all diagnosed as benchmark-design findings, not wrong-asymptotic implementations: the seeded family's per-degree root geometry makes driver wall time non-monotonic (replace with a smooth-difficulty family, e.g. fixed root-separation products); the compare group's 5-rung n=2..6 schedule is too narrow for a slope fit (widen); runTaylor's fixed non-integer centre 1/4+i/8 sits in a ~n^2.25 transition band (use an integer centre or push into the n^3 regime); witnessCheck/newtonSquare/certify run in the startup-dominated microsecond band (hoist per-call overhead or grow the schedule); refineTo's Newton-doubling quantises achieved precision (parametrise by achieved prec). Each Concern in the report carries its resolution. Phase-4 exit for hex-roots is blocked on these plus the time-budget issue.

🤖 Prepared with Claude Code

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions