On-Device AI Deployment Fit Test

• **Cactus / Needle**: runtime, models, and tool calling directly address deployment; the project can add auto-fit and benchmarks itself.
• **OnDevice.AI**: model shortlist, API, and sources already occupy “best model for a phone.”
• **PulzeMark / Apple Silicon LLM Bench / MobileAIBench**: real-device or standardized benchmarks with stronger data coverage.
• **MLC LLM, llama.cpp, MLX, Core ML, Google AI Edge documentation**: free framework-level selection and examples.
• **Internal experiments and consultants**: mature teams build harnesses, while high-value buyers may prefer services over SaaS.
• Entry wedge: combine the user’s device mix, task rubric, runtime constraints, and uploaded JSON into a model + quantization + runtime decision. Do not maintain a universal winner.
• Accessible results already include guides, papers, leaderboards, and app documentation. US-English Top 10/AIO/PAA is not measured; SEO cannot be the sole channel.

8h MVP Checklist

1. 1.Define model/runtime/result schemas, license/source fields, and public fixtures.
2. 2.Build a minimum registry of three to five models, Cactus/llama.cpp, and two device classes.
3. 3.Build Planner, shortlist, exclusion explanations, and evidence levels.
4. 4.Build benchmark-pack generator, commands/configs, and JSON validator.
5. 5.Build Results, go/no-go, next experiment, and Markdown export.
6. 6.Complete Registry, Methodology, Pricing, About, FAQ, Privacy, and the `$149` Payment Link.
7. 7.Verify reproducibility on at least three devices/public results; add runtimes/devices only after payment.