The most useful number my system produces is one I never set out to buildThe most useful number my system produces is one I never set out to build. I wrote a learning platform for my two children. To choose the next question it has to estimate where the child is, per subje5d ago · 2 min readopenred.hashnode.dev
De-overlap before you believe any walk-forward resultA worked example of the cheapest false positive in time-series backtesting, from my own evaluation. I had a result: 55.2% direction accuracy against a 51.1% majority-class benchmark, n=912. That is +421AOct 13 · 2 min readopenred.hashnode.dev
Agentic systems are a routing problem, not a model problemAgentic AI splits into two halves, and they behave completely differently. The first is model capability. It improves on its own, it arrives for free, and it asks nothing of you. Your system gets bettSep 28 · 2 min readopenred.hashnode.dev
A Forecasting Model That Correctly Refuses to ForecastIt beat the baseline by 98.7% on one series and lost to repeating the last value on another. That contrast is the most useful single fact about it. Part one covered what TimesFM 3.0 is: a stateless n1ASep 26 · 5 min readopenred.hashnode.dev
The merchant side of agentic commerce is an integration problemMost independent restaurants still route off-premise volume through marketplaces at 15% to 30% commission even though direct ordering costs roughly 3%. The reason is not inertia. Direct ordering made 3ISep 25 · 2 min readopenred.hashnode.dev
