On-Device Neural Model and Memory Allocation Audit Checklist

Role: Principal Mobile Performance and Neural Engine Systems Engineer with deep expertise in App Store performance constraints.

Context

• Native App Framework: {{app_framework}}
• Execution Hardware Target: {{target_compute_hardware}}
• Model Quantization Level: {{model_quantization_tier}}
• Maximum RAM Ceiling: {{peak_ram_ceiling_mb}}
• Maximum Background Budget: {{background_execution_budget_sec}}
• Acceleration Pipeline: {{coreml_metal_pipeline}}

Task

Generate a comprehensive performance and memory-budget checklist to audit on-device analytics and neural model inference, ensuring the build satisfies strict App Store runtime limits and watchdog timer thresholds.

Method

1. Analyze {{model_quantization_tier}} model weights to calculate static disk footprint and baseline memory allocation.
2. Evaluate memory pressure under peak inference load against the {{peak_ram_ceiling_mb}} threshold to prevent out-of-memory (OOM) crashes.
3. Verify execution thread routing within {{coreml_metal_pipeline}} to ensure UI thread isolation and 60/120 fps render stability.
4. Audit background processing hooks to ensure model execution respects the {{background_execution_budget_sec}} runtime limit without OS termination.
5. Inspect thermal state handling and throttling logic for sustained execution on {{target_compute_hardware}}.
6. Formulate granular checklist verification points for cold-start latency, memory leaks, and GPU/NPU shader compilation.
7. Outline concrete profiling checkpoints using native developer diagnostic instruments.

Constraints

• Checkpoints MUST define explicit numerical pass/fail thresholds based on {{peak_ram_ceiling_mb}} and {{background_execution_budget_sec}}.
• Checkpoints MUST NOT accept subjective performance terms like 'fast' or 'optimized'.
• Memory allocations must distinguish between clean, dirty, and compressed memory heaps.
• All items must be tailored to {{app_framework}} best practices.

Output format

• Compute Profile Overview (System constraints and model characteristics)
• Section 1: Static Binary & Asset Footprint Checklist (4-5 markdown checkboxes)
• Section 2: Dynamic RAM & Thread Concurrency Checklist (6-8 markdown checkboxes with threshold metrics)
• Section 3: Thermal, Battery & Watchdog Prevention Checklist (4-5 markdown checkboxes)
• Instrumentation Sign-off Protocol (Required profiling tools and diagnostic captures)

Self-review

1. Did I reference the specific constraints from {{coreml_metal_pipeline}} and {{target_compute_hardware}}?
2. Are numerical thresholds embedded directly inside the checklist checkboxes?
3. Does the checklist explicitly protect against OS watchdog terminations?