Research notes, engineering work, and long-form essays on systematic trading infrastructure.https://harrisoldroyd.com/https://harrisoldroyd.com/local-ai/apple-silicon-macs-for-local-llms/https://harrisoldroyd.com/local-ai/apple-silicon-macs-for-local-llms/A concise guide to the M3 Ultra Mac Studio, M4 Max Mac Studio, and M5 Max MacBook Pro for local LLMs, including real benchmarks, current and used pricing, MLX and Metal software, Thunderbolt clustering, and an estimate for a possible M5 Ultra.Sat, 11 Jul 2026 00:00:00 GMThttps://harrisoldroyd.com/local-ai/dgx-spark-gb10-for-local-llms/https://harrisoldroyd.com/local-ai/dgx-spark-gb10-for-local-llms/A practical guide to NVIDIA DGX Spark and other GB10 systems for local LLMs, covering real token speeds, memory bandwidth, MoE models, clustering, thermals, software support, pricing, Strix Halo, and Mac Studio alternatives.Sat, 11 Jul 2026 00:00:00 GMThttps://harrisoldroyd.com/local-ai/strix-halo-for-local-llms/https://harrisoldroyd.com/local-ai/strix-halo-for-local-llms/A concise guide to AMD Strix Halo for local LLMs, covering real token speeds, memory bandwidth, MoE models, ROCm and Vulkan support, clustering, and the 192 GB Gorgon Halo refresh.Sat, 11 Jul 2026 00:00:00 GMThttps://harrisoldroyd.com/local-ai/best-gpus-and-devices-for-local-ai/https://harrisoldroyd.com/local-ai/best-gpus-and-devices-for-local-ai/A comparison of GPUs, workstation cards, unified-memory devices, and CPU inference setups for local AI, focused on price, speed, memory capacity, software support, and practical tradeoffs.Thu, 09 Jul 2026 00:00:00 GMThttps://harrisoldroyd.com/local-ai/rtx-3090-for-local-ai-inference/https://harrisoldroyd.com/local-ai/rtx-3090-for-local-ai-inference/A practical look at the RTX 3090 for local AI workloads, covering specs, VRAM, bandwidth, TFLOPS, used pricing, power limiting, LLM performance, image generation, and why it remains one of the best-value GPUs for running open models at home.Thu, 09 Jul 2026 00:00:00 GMThttps://harrisoldroyd.com/quant/the-consolidation-filter-that-didnt-work/https://harrisoldroyd.com/quant/the-consolidation-filter-that-didnt-work/After the expansion-filter research, I tried to attack the drawdown problem more directly by detecting consolidation and filtering it out. The visual idea still makes sense, but the first binary versions were too thin to separate the bad loss clusters cleanly.Fri, 19 Jun 2026 00:00:00 GMThttps://harrisoldroyd.com/quant/from-trend-detection-to-expansion-detection/https://harrisoldroyd.com/quant/from-trend-detection-to-expansion-detection/A trend continuation strategy was performing well enough to justify further development, but repeated drawdowns during consolidation kept appearing. This research explores eighteen different filtering approaches and the unexpected pattern that emerged from the results.Sat, 06 Jun 2026 00:00:00 GMT