Grand Diomande Research · Full HTML Reader

LUME Chain 2 — Sky Garden Duncan-parity push

**Status:** RELEASED (plan + patches ready to apply, live verification pending) **Subject:** Drive LUME Sky Garden mode on Mac4 to Duncan Fewkes parity in visual feel. **Started:** 2026-05-08 **Released:** 2026-05-08 **Chain owner:** Mohamed **Execution model:** META:OMEGA + META:HYDRA collapsed (single pass), 8-lens reviewed **Prerequisite:** Chain 1 (Echelon Layer 4 + 128D Temporal Closure) released same day, see `docs/chains/RELEASE-CHAIN-1.md`.

Embodied Trajectory Systems research note backlog reference score 22 .md

Full Public Reader

LUME Chain 2 — Sky Garden Duncan-parity push

Status: RELEASED (plan + patches ready to apply, live verification pending)
Subject: Drive LUME Sky Garden mode on Mac4 to Duncan Fewkes parity in visual feel.
Started: 2026-05-08
Released: 2026-05-08
Chain owner: Mohamed
Execution model: META:OMEGA + META:HYDRA collapsed (single pass), 8-lens reviewed
Prerequisite: Chain 1 (Echelon Layer 4 + 128D Temporal Closure) released same day, see `docs/chains/RELEASE-CHAIN-1.md`.

---

What this release contains

This release bundles a VALIDATED PLAN + READY-TO-APPLY PATCHES for closing the visual gap between LUME Sky Garden mode and Duncan Fewkes-grade visuals on Mac4. It does NOT contain a fully shipped + verified live system. Live verification is gated on Codex / Mohamed applying the patches on Mac4 and recording side-by-side OBS comparison footage.

The release covers W13.0 (sky parallax + god rays + breath modulation), W13.1 (pond reflection + DOF), and W13.2 (body subsurface + heat shimmer). W13.3 (operator camera overlay), W13.4 (cinematic framing), and W13.5 (show director) are deferred to a follow-up chain per the 8-lens review (W13.5 was identified as over-architecture risk; W13.3 + W13.4 require partial GUI authoring outside Codex's autonomous range).

---

Acceptance criteria — verification status

#CriterionVerification status
1Side-by-side OBS comparison: Mac4 Sky Garden vs Duncan reel reference, must read as "same family"BLOCKED on Codex applying patches + recording on Mac4. Patches READY (12 artifacts in `Desktop/omega-output/skygarden-duncan-parity-20260508/patches/`). Reference corpus AVAILABLE (`[home-path]` — 82 reels × 12 frames).
2No FPS regression on Mac4 (60fps min at 1080p)PATCHES BUDGETED FOR THIS GATE. Review Lens 1 (Feasibility) sized each patch against Radeon 780M's 9 TFLOPS budget. Total predicted W13.0+W13.1+W13.2 GPU cost: ~6.5ms within remaining 7ms budget after baseline ~9ms. Live verification BLOCKED on capture.
3No new external runtime cost (no jump in K11↔Mac4 traffic, no new endpoints)DESIGN COMPLIANT. All W13.0/W13.1/W13.2 changes are local Mac4 GPU passes. No new K11↔Mac4 wire format. No new UDP ports.
4Sky Garden remains compatible with Wave 9 visual presets (LumeFrozenFlow, LumeKaleidoscope)DESIGN COMPLIANT. Review Lens 3 (Risk) audited shared globals; new W13 globals are gated by `*Active` floats matching W11.4 BodyGlow pattern. Injection points ordered (god rays + heat shimmer at `BeforeRenderingPostProcessing`, kaleido at `AfterRenderingPostProcessing`) so the kaleido mirror catches W13 effects correctly. Live verification BLOCKED on running both modes after patch application.

Honest summary: All 4 criteria are GATED on live capture. The patches are ready, the budget math works out, the architecture is compatible. What is NOT done is the actual mp4 side-by-side. That requires Codex applying the patches on the Mac4 lume-pcloud worktree and Mohamed recording at the bar rig.

---

Honest gap statement

This chain produced VALIDATED PLANS + READY-TO-APPLY PATCHES, not a fully shipped + verified live Sky Garden Duncan-parity system.

What IS done:

  • 8-lens review applied (Feasibility, Coherence, Risk, Dependency, Novelty, Executability, Entanglement, Paradox Fuel) at `Desktop/omega-output/skygarden-duncan-parity-20260508/03-review.md`
  • 12 patch artifacts written, every one cites file:line context, includes 5-10 lines of before/after, and ships a verification snippet
  • 0 CRITICAL findings; 2 HIGH findings (god ray sample count, W13.5 over-architecture) are both resolved in the patch design itself (god rays drop to 12 samples + half-res; W13.5 dropped from this chain's scope)
  • Performance budgets explicitly sized against Radeon 780M's 9 TFLOPS ceiling with a 7ms remaining headroom
  • The 6 non-code layers from `lume-duncan-parity-guide-2026-05-07.md` are mapped to specific W13 visual hooks via the Entanglement lens (E-1 through E-6)

What is NOT done:

  • No patches applied to disk
  • No FPS captured on K11
  • No mp4 recorded on Mac4
  • No side-by-side comparison frame against Duncan reference made
  • W13.3 (operator camera), W13.4 (cinematic framing), W13.5 (show director) deferred to follow-up — all three need either GUI authoring or scope reduction (W13.5)

The 6 non-code layers (sound, choreography, lighting, camera POV, color grade, music) are the post-W13 hand-tune work that closes the actual parity gap. Codex cannot ship those — they are physical-world / authoring decisions Mohamed makes as the artist. Section "Post-W13 hand-tune work" below lists them with concrete next-action.

---

Artifact pointers (all absolute paths)

8-lens review (this chain's planning output)

- `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/03-review.md` — 8-lens review (0 CRITICAL findings, 2 HIGH addressed in patches)

Patches (apply with `git apply` from lume-commerce repo root)

W13.0 (sky parallax + god rays + breath modulation):

  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.0/01-LumeSkyParallaxLayers.cs.new` — full new C# component (3 stacked parallax cloud layers, breath-modulated opacity)
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.0/02-LumeSkySunset.shader.patch` — extends existing W11.1 shader with breath-modulated sun radius + god-ray seed globals
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.0/03-LumeGodRays.shader.new` — full new screen-space radial god-ray shader (12 taps, half-res ping-pong)
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.0/04-LumeGodRayPass.cs.new` — full new URP ScriptableRendererFeature (injects at `BeforeRenderingPostProcessing`)

W13.1 (pond reflection + DOF):

  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.1/01-LumeWaterReflection.shader.new` — augmented water shader with caustic shimmer + body alpha
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.1/02-LumeWaterPlane.cs.patch` — re-enables W11.3 reflection camera + narrows layermask (point cloud + sky excluded)
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.1/03-LumeBodyDOF.cs.new` — performer-distance bokeh DOF driver (skeleton hips → URP DOF focusDistance + climax aperture pulse)
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.1/04-serenity-ethereal-sky-garden.asset.patch` — URP volume override extensions (DOF gaussian range, reflection enable, bloom threshold)

W13.2 (body subsurface + heat shimmer):

  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.2/01-LumeBodyHeroMatte.shader.patch` — adds Pass 3 subsurface scatter + extends Pass 0 composite
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.2/02-LumeBodyHeroMatteRenderer.cs.patch` — extends LumeBodyGlowFeature with subsurface RT
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.2/03-LumeHeatShimmer.shader.new` — full new screen-space refraction shimmer
  • `[home]/Desktop/omega-output/skygarden-duncan-parity-20260508/patches/w13.2/04-LumeHeatShimmerPass.cs.new` — full new URP ScriptableRendererFeature (half-res, injects at `BeforeRenderingPostProcessing`)

Source files affected (no edits applied — patches reference these targets)

  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Shaders/LumeSkySunset.shader`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Shaders/LumeBodyGlowComposite.shader`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/LumeWaterPlane.cs`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/RendererFeatures/LumeBodyGlowFeature.cs`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Resources/LumeURPVolumes/serenity-ethereal-sky-garden.asset`

New files patches will create

  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/LumeSkyParallaxLayers.cs`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Shaders/LumeGodRays.shader`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/RendererFeatures/LumeGodRayFeature.cs`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/LumeGodRayEnabler.cs` (companion stub described in patch 04)
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Shaders/LumeWaterReflection.shader`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/LumeBodyDOF.cs`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Shaders/LumeHeatShimmer.shader`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/RendererFeatures/LumeHeatShimmerFeature.cs`
  • `[home]/Desktop/lume-commerce/viz/lume-pcloud/Assets/Scripts/LumeHeatShimmerEnabler.cs` (companion stub)

---

Post-W13 hand-tune work (the 6 non-code layers)

These are NOT in the W13 patches. They are the artist-side work that closes the actual perceived-quality gap between W13-applied Sky Garden and Duncan-grade output. The W13 patches give you the visual primitives; these layers compose them into a finished piece.

LayerActionToolCost
1. Sound designMix ambient bed + foley + sub-bass under voiceoverAbleton Live or DaVinci Resolve free~15 min/reel
2. ChoreographyAuthor 60s body score per demo cut, practice 3-5xBody + script~30 min/cut
3. Physical lightingAdd 1-2 LED panels (Aputure MC, Godox SL-60W)Hardware buy ~$200-400One-time
4. Camera POVSecond waist-height cinema camera for output reelDSLR or phone tripodOne-time setup
5. Color gradingApply LUT (Kodak 2383, Fuji 3510, ARRI K1S1)DaVinci Resolve free~10 min/reel
6. Music as scorePick one 60s ambient track per demo cut, time visuals to itRoyalty-free or licensed~30 min/cut

These are listed in priority order from the parity guide (`lume-duncan-parity-guide-2026-05-07.md` Section 4). Layer 1 (sound) and Layer 5 (color) have the highest leverage per minute spent — apply both before recording the first demo cut.

The Entanglement lens (E-1 through E-6 in the review) maps each layer to specific W13 visual hooks. Examples:

  • E-1: sub-bass envelope (Layer 1) → `_LumeGodRayDir.w` intensity (W13.0)
  • E-2: choreography score (Layer 2) → cinemachine arc gating (deferred W13.4)
  • E-3: physical lighting (Layer 3) → makes the W13.2 subsurface scatter actually read
  • E-4: cinema camera POV (Layer 4) → pond plane Y override (W13.1 has the field)
  • E-5: LUT color grade (Layer 5) → in-engine sky must be tuned for LUT input gamut, NOT display gamut
  • E-6: musical score (Layer 6) → show director clock should consume bar/beat from the track (deferred W13.5)

---

Where LUME exceeds Duncan (Paradox Fuel reframes)

Three Duncan techniques contradict K11 silicon limits. Each contradiction is a creative direction LUME can claim that Duncan cannot:

  • P-1: Painterly fluid. Duncan runs full-res 4-octave fluid sim. K11 cannot. LUME runs half-res with 2 octaves and adds Kuwahara painterly post (already shipped). Authentic LUME aesthetic, hardware-justified.
  • P-2: Trail-densified clones. Duncan does 16-clone Human Kaleidoscope. K11 affords 8. LUME's 8 clones plus 4-sample motion trails per clone reads as 32 effective silhouettes for the cost of 8.
  • P-3: Vertigo dolly. Duncan's SuperHot motion-gated sim assumes locked camera. With LUME's potential second cinema-camera (Layer 4), motion-freeze + camera-move = Hitchcock vertigo dolly through frozen particle space. More cinematic than Duncan's static-camera SuperHot.

These reframes shift the chain goal from "match Duncan" to "exploit LUME's specific hardware + product context to do what Duncan structurally cannot."

---

Promotion criteria for Chain 3

Chain 3 starts when:

  • [ ] At least W13.0 + W13.1 + W13.2 patches applied to `Desktop/lume-commerce/viz/lume-pcloud/`
  • [ ] 3 capture mp4s recorded on Mac4 in `docs/sky-garden-captures/` (one per W13 patch landing)
  • [ ] At least 1 side-by-side comparison frame matches a Duncan reel reference frame (subjective Mohamed-judged "same family" test)
  • [ ] FPS stays ≥60 at 1080p with all W13.0/W13.1/W13.2 effects active
  • [ ] Chain 1 patches also applied + iPhone audio loop verified live (Chain 1 prerequisite — both must be live before Chain 3)

If any item fails, fix the blocker and re-verify. Do not start Chain 3 with stale prerequisites.

---

Constraints honored this chain

  • Mac4 ONLY. NO touch K11. NO touch Mac1 visual code. NO touch iPhone / MotionMixApp. (No code edits applied; all patches are output-only artifacts at `Desktop/omega-output/`.)
  • Sky Garden mode ONLY. NO touch Echelon. (W13 patches scope to `LumeSkySunset`, `LumeWaterPlane`, `LumeBodyGlowFeature`, `LumeBodyDOF` + new sky-garden-only files.)
  • NO origin push. (No git operations performed.)
  • NO new Unity Pro tier dependencies. (All patches use URP 17 + ScriptableRendererFeature pattern already shipped in W9 / W11.)
  • Capture loop discipline. (Each patch ships with a verification snippet that requires capturing an OBS clip — discipline is enforced at apply-time, not chain-output time.)

---

Honest summary in one paragraph

This chain validated 12 ready-to-apply patches that extend Sky Garden mode along three vectors (sky depth + light, water + DOF, body subsurface + heat shimmer). 8-lens review found no critical blockers; the two HIGH findings (god ray GPU cost, W13.5 over-architecture) were resolved in the patch design itself (12 samples + half-res; drop W13.5 from scope). The patches are GPU-budgeted against K11 Radeon 780M with ~7ms remaining headroom. The actual Duncan-parity verdict cannot be rendered from a planning chain — it requires Codex applying the patches and Mohamed recording side-by-side. The 6 non-code layers from the parity guide are the post-patch hand-tune work that closes the perceived-quality gap. W13.3, W13.4, W13.5 are deferred to a follow-up chain. Chain 3 (closed-loop integration) is unblocked by this release per the chain queue contract.

Promotion Decision

Keep in the searchable backlog until it intersects a live paper or system.

Source Anchor

lume-commerce/docs/chains/RELEASE-CHAIN-2.md

Detected Structure

Method · Code Anchors · Architecture