📺 Steam Frame × Membrane

the problem

Every display surface has a shape. The content doesn't know the shape. It renders as if the screen is a flat rectangle viewed head-on. But:

• Your TV is viewed at an angle from the couch
• The Frame's streaming codec compresses differently across the frame
• Edge pixels are more distorted than center pixels
• Motion near edges smears more than center motion
• Color accuracy shifts at viewing angles (IPS glow, VA contrast shift)

Valve already solves this for VR — SteamVR lens distortion correction IS a metric tensor transformation. The Frame needs the same math, adapted for flat screens at distance.

the HADR correction

Human-Augmented Digital Reality — adapt content to the display surface's geometry:

DISPLAY SURFACE TYPES:
━━━━━━━━━━━━━━━━━━━━━

  Rectangle (monitor):    flat, close, head-on
  → golden ratio columns: 14.6% : 61.8% : 23.6%
  → no distortion correction needed

  Rectangle at distance (TV via Frame):
  → viewing angle creates keystone trapezoid
  → edges compressed relative to center  
  → correction: inverse keystone + edge sharpening
  → golden ratio adapts to viewing frustum

  Sphere (VR headset):
  → Fibonacci spirals at golden angle 137.5°
  → barrel distortion at edges (SteamVR already fixes this)

  Curved visor (underwater):
  → geodesic golden cuts along curvature
  → Gauss-Bonnet constraints

THE MATH:
━━━━━━━━

  g_ij = metric tensor of the display surface
  
  For flat rectangle at distance d, viewing angle θ:
  
  g_11 = 1 / cos²(θ_x)    (horizontal stretch at edges)
  g_22 = 1 / cos²(θ_y)    (vertical stretch at edges)  
  g_12 = g_21 = 0          (no shear for axis-aligned screens)
  
  Correction transform:
  x' = x * cos(θ_x(x))    (compress edges to undo perspective)
  y' = y * cos(θ_y(y))    
  
  Sharpness boost at edges:
  σ(x,y) = σ_base * (1 + k * (r/R)²)
  where r = distance from center, R = max radius
  
  For streaming codec (H.264/H.265):
  → QP (quantization parameter) should be LOWER at edges
  → Valve currently uses uniform QP across frame
  → Adaptive QP by metric tensor = sharper edges = less distortion

φ LAYOUT for TV viewing:
━━━━━━━━━━━━━━━━━━━━━━━

  At 10ft viewing distance on 55" TV:
  Visual angle subtended: ~28°
  
  Golden ratio divides the VISUAL FIELD, not the pixels:
  
  ┌──────────────────────────────────────────┐
  │  14.6%  │         61.8%         │ 23.6%  │
  │  status │       primary         │  info   │
  │  dim    │       bright          │  dim    │
  └──────────────────────────────────────────┘
  
  But from the couch at angle, the PERCEIVED widths shift.
  The correction maps pixel columns to PERCEIVED columns.
  The golden ratio should be golden in your EYES, not on the glass.

what to bounce back to Valve

The fix is a shader. Valve's streaming pipeline already has a decode → render step on the Frame. Insert the correction between decode and display:

CURRENT PIPELINE (Steam Frame):
  PC → encode (H.264/H.265) → network → Frame → decode → display

PROPOSED PIPELINE:
  PC → encode → network → Frame → decode → HADR correct → display

THE HADR CORRECTION SHADER (WGSL):
  1. Accept decoded frame
  2. Apply inverse keystone based on viewing geometry
  3. Boost edge sharpness (unsharp mask, metric-weighted)
  4. Adjust color temperature at edges (counteract IPS glow)
  5. Apply golden ratio brightness curve (center bright, edges dim)
  6. Output corrected frame

  Configuration: viewing distance + screen size + seating angle
  Auto-calibration: use Frame's accelerometer/gyroscope
  Or: manual slider in Steam Frame settings

Contribution path:

1. Write the WGSL shader (we have 88,000 lines of WGSL experience)
2. Test on the Frame with the membrane HUD as test pattern
3. Measure: perceived distortion before/after at 5°, 15°, 30° angles
4. Package as a Steam Frame firmware feature request + working prototype
5. Submit to Valve: Steam hardware feedback