Skip to content

Lantern: Re-evaluate metric coefficients on the 2026 golden trace dataset #17438

Open

Description

@connorjclark

Summary

Lantern's static linear combination weights (intercept, optimistic, pessimistic) in FirstContentfulPaint, LargestContentfulPaint, SpeedIndex, and Interactive were originally tuned on the 2018/2019 HTTP/1.1 & HTTP/2 dataset:

  • FirstContentfulPaint: intercept: 0, optimistic: 0.5, pessimistic: 0.5
  • LargestContentfulPaint: intercept: 0, optimistic: 0.5, pessimistic: 0.5
  • SpeedIndex: intercept: 0, optimistic: 1.4, pessimistic: 0.4
  • Interactive: intercept: 0, optimistic: 0.45, pessimistic: 0.55

Next Steps

With the 2026 golden trace dataset (#17411), evaluate whether re-tuning these coefficients improves accuracy (MAPE, rank correlation, and bias) for modern HTTP/3 multiplexing and contemporary JS bundle patterns using the analysis tools in core/scripts/lantern/.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions