The author seems to be completely confusing what probabilistic and deterministic mean.
If the world is truly and unknowably probabilistic, there is no hidden information to be learned. Updating your plans as you go in response to new (random) information will provide little benefit.
In contrast, if it is deterministic, there is hidden information to be learned. Once you learn this information (which may occur as the progress moves along), your subjective probability distribution [1] describing the possible outcomes loses entropy and your estimates become more accurate.
[1] This represents your uncertainty about the situation, not anything fundamental.
I believe you are conflating incomplete information (the game theory term) with probability (statistics).
If a random variable is truly random, there is no information in samples of the variable. Updating your plans in response to a random variable is like blindly betting that the previous dice roll will be the next dice roll -- any successful gambling strategy must rely on real knowledge of the underlying probabilities. However, you _can_ measure the probabilities as you go, updating your "best guess" with each sample: Bayesian models are an example of this.
The author is instead talking about incomplete information. In game theory, a game with incomplete information is not the same as a random variable. Software engineering is a good example of a game with incomplete information: pick your favorite piece of software. But pick one that (as far as you know) has no bugs. With 20/20 hindsight, you can determine how long it took to produce that software.
The author says the probabilities come from not knowing everything in advance. Once you know exactly how the software _should_ behave, you've written the software.
I don't think I'm conflating them. A probability distribution represents your beliefs about a situation where your information is incomplete.
The author is instead talking about incomplete information.
Exactly - incomplete information about a system which is more or less deterministic. And he's calling it the "probabilistic" model, in contrast to a "deterministic" model which he criticizes.
My only point is that the terminology raganwald is using is wonky. The distinction he is trying to make isn't probabilistic vs deterministic. It's open loop vs closed loop (in the control theory sense).
If the world is truly and unknowably probabilistic, there is no hidden information to be learned. Updating your plans as you go in response to new (random) information will provide little benefit.
In contrast, if it is deterministic, there is hidden information to be learned. Once you learn this information (which may occur as the progress moves along), your subjective probability distribution [1] describing the possible outcomes loses entropy and your estimates become more accurate.
[1] This represents your uncertainty about the situation, not anything fundamental.