Skip to content

bug(D4): concentric AP haze banding at 50–100 km altitude (deferred, has bisection notes) #5

Description

@DennisSmolek

Tracking the ROADMAP.md D4 defect as an issue.

From ~50–100 km camera altitude, smooth concentric arcs (iso-distance contours around the nadir) appear in AP-LUT-driven haze over the planet surface. Clean at ground level; clean above blendEndKm (pure raymarch).

Known facts from bisection (full notes in ROADMAP.md § D4):

  • Bands live in the AP-LUT branch: hazeMode=raymarch is clean; blendEndKm directly modulates visibility; TLUT/MS resolution has no effect.
  • Tried and reverted (4827ff097e3003): ±half-texel W dither + 64-sample jittered LUT build — bands persisted AND added noise. So the W-interpolation-kink hypothesis is wrong or incomplete.
  • Next suspects: 32×32 screen-space XY interpolation (froxel rays diverge strongly at altitude; the underground-froxel correction rewrites ray dir/tMax per voxel, so neighbouring XY texels store integrals of genuinely different rays), or banding baked into stored values by the correction itself.
  • Discriminator before any fix attempt: ?debug=ap-alpha screenshot at a banding altitude vs ?debug=rm-alpha.
  • Possibly related: reference scales AP samples per slice (see the per-slice SPP issue) and caps coverage at 128 km vs our 1024 km in demo 05.
  • Workaround shipping today: hazeMode=raymarch at altitude, or lower blendEndKm so the auto ramp hands off earlier.

🤖 Generated with Claude Code

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions