“We often assume that if we have good data and a good model, a decision can be made. That assumption quietly fails more often than we think,” – wrote Ron Itelman. “By decidability, I mean a boundary condition: A decision is decidable if the minimum semantic prerequisites required by the decision rule are grounded.”

“Take something trivial: a trading rule that says ‘Buy if price today is less than price yesterday.’ The logic is simple. The model works. The data exists. But the system can still halt. Why?

When a stock-market trader at a London Stock Exchange desk asks their AI agent, “What were my trades yesterday?” The system resolves ambiguity before it can answer. Which time zone? “Yesterday” from whose perspective: the headquarters in New York, U.S.A. the trader in London, UK or their client at Shenton Way, Singapore? The AI silently decides, then returns a confident answer without exposing those variables to the user.” Link