Blog
Articles on statistics, supply chain optimization, and high-performance computing.
Get new articles in your inbox
Practical articles on supply chain forecasting, statistical modeling, and high-performance Rust systems. No spam, no marketing — just new posts when they're published. Unsubscribe anytime.
Powered by Buttondown. Double opt-in confirmation required.
Forecast Value Added: Where in Your Process Does the Forecast Actually Get Better?
Your S&OP process has six steps. You measure the final forecast accuracy at the end. You do not measure which steps in between made it better and which made it worse. The result: planning effort is spread evenly across steps that add value and steps that destroy it. Here is how Forecast Value Added (FVA) makes the difference visible — and which conversation to have once it is.
Your 95% Prediction Interval Probably Isn't 95%
When your forecast says "100 ± 30 with 95% confidence," your reorder points, service-level commitments, and working capital are all sized against that band. If the band is miscalibrated, every downstream decision is wrong. Here's how to make intervals that mean what they say.
When the Forecast Pyramid Lies — and How MinT Stops It at 30,000 Leaves
Forecasts at the SKU-store level rarely add up to your category total. That gap is where money goes missing. We benchmarked four hierarchical reconciliation methods on the full M5 retail dataset (30,490 series, 12 levels, 28-day horizon). Here's what each does and which one to pick.
Magpie-RS — Semantic Search That Survives Real Workloads
Vector search libraries are easy to write and hard to ship. Magpie-RS is a Rust vector database designed for production from the start: bounded memory, atomic checkpoints, and durable resume — proven by indexing the entire CRAN.
Wielding Occam's Razor
A landmark study proves that reducing your forecasting model pool from 19 to 8 models improves accuracy while cutting compute costs by 41%. For large retailers, that's over $1M in annual savings — and 108,000 tonnes less CO₂.
Stop the Inventory Death Spiral
Traditional demand classification systems rely on rigid, decades-old mathematical thresholds that create inventory bloat and revenue loss. Replace them with probabilistic model-based classification.