ozone-tech_owl_prime/docs/OZON_FINAL_FIX_DIAGNOSIS.md

OZON Final Fix — Production Diagnosis (Before)

Date: 2026-07-09
Branch: dan_branch
Production URL: https://arhipovdan.ru/
Evidence: docs/ozon_final_fix_evidence/before/

---

1. What was actually visible on production (before fix)

Video

• before_full_play.webm — full Play cycle (~80 s), 8 cases.

Screenshots

File	Observation
01_idle.png	Idle state OK, hero + 3D canvas load.
02_b_item_problem.png	Case 1 B routing starts; item on main belt.
05_item_stopped_on_conveyor.png	P0: B lunchbox (case 2) sits on long green transfer deck at belt height — looks like active conveyor, not inside a floor bin.
04_b_no_real_bin.png	B zone reads as belt extension + low tray; no distinct industrial receiving container on the floor.
03_lag_or_console.png	Visible stutter during multi-item playback; HUD updates but motion not fluid.
06_console.png / 06_console.json	See console section below.

Console (Playwright capture, 5 s after Play)

errors: 0
log_spam (console.log): 0
warnings: 5

Warnings were not from app code:

• THREE.Clock: deprecated (three.js / R3F)
• GPU stall due to ReadPixels (Playwright screenshot GPU readback)

No console.log from src/. console.error only in error boundaries.

Lag / stutter

• Visible frame drops when 3+ items on screen with STL meshes.
• Auto-cinematic camera + PointCloud + ScanLine + MotionTrail active on desktop.
• Belt ran to CONVEYOR_END_X = 3.85 — visually merged B transfer with main line.

B item routing

• Motion model had b_receiver_floor but 3D scene drew a long green transfer deck (1.5 → 3.6 m).
• B item settled pose was correct in unit tests but visually appeared on the spur because:
  1. Transfer deck looked identical to belt.
  2. Settle happened late (near exit phase).
  3. Cinematic camera followed current case, hiding settled B in bin during screenshots.

B physical bin

• No separate floor-standing bin with walls, legs, and drop chute.
• B read as “tray on conveyor end”, not OZON-level receiving zone.

C/D receivers

• Roll cages present on floor with visible floors.
• C/D routing generally worked in motion model; containment OK in tests.
• Multiple STL items + effects caused lag before C/D cases finished.

Render load (code audit)

Item	Before
MAX_VISIBLE_ITEMS	8
Simultaneous STL	Up to 8 heavy STLs
useFrame hooks (continuous scene)	6 (CinematicCamera, PointCloud, ScanLine, Stepper pulse, etc.)
Rollers	Static (already fixed prior pass)
Shadows	shadows={false} on Canvas
Heavy effects default	ON on desktop (!liteScene)

---

2. What did NOT meet OZON-level demo bar

1. B receiver not physically readable — item on belt spur, not in floor bin.
2. Performance — stutter with effects + many STLs.
3. Visual sorting story broken — viewer cannot see A → inspection → junction → distinct B/C/D containers in one glance.
4. Proof gap — prior “готово” reports without production video.

---

3. Root causes (code)

Symptom	Root cause	File(s)
B on belt	Long b_transfer surface + transfer deck mesh = belt extension; late settle timing	physicalLayout.ts, conveyorNetwork.ts, physicalItemMotion.ts, SorterDigitalTwinContinuous.tsx
No real B bin	BReceiverBin was floor pad + walls but dominated by 2+ m transfer deck	SorterDigitalTwinContinuous.tsx
Lag	MAX_VISIBLE_ITEMS=8, all items load STL; PointCloud/ScanLine/MotionTrail/CinematicCamera active	SorterDigitalTwinContinuous.tsx, PhysicalPlaybackItem.tsx
Console noise	No app console.log; only THREE deprecation + Playwright GPU warnings	N/A (acceptable)
setState in useFrame	None found; camera uses ref + direct Three.js mutation	SorterDigitalTwinContinuous.tsx

---

4. Acceptance criteria for fix (reference)

• B: main_belt → inspection → junction → b_transfer → chute_b → b_bin_floor → settled
• MAX_VISIBLE_ITEMS ≤ 6
• ENABLE_DEMO_EFFECTS = false by default
• Settled items: simplified mesh, neutral material
• Production video + screenshots after deploy