Libro

A manager writes to her team: "We are moving forward with a restructuring, effective the first of next month." No one says the word layoffs. By lunchtime, everyone in the building is talking about who is going to lose their job. This book asks whether that unstated word is, in some measurable sense, already present the moment the softer one is chosen. Complementary Semantics, a construct of Learnable Theory, holds that a speaker's unselected alternative does not vanish: it recedes into the Umbra Cone, a zone of reduced activation that keeps shaping judgment without ever reaching explicit discourse. For the first time, that claim can be tested not only through what people say but also through what a language model's own internals do with the word it never writes down. Across eleven articles written as a single, self-correcting research program, this volume proposes three falsifiable hypotheses, builds and rebuilds the instruments needed to test them, and follows the evidence wherever it leads, including away from its own earlier conclusions. A broadened claim is proposed, then retracted once a properly powered test comes back against it. A borrowed mechanism from thought-suppression research is tested, found wanting, and replaced with one native to the theory. The resulting account, affect-gated rather than universal, is pushed to the most extreme content imaginable, tested across model families and clause wordings, and finally asked what it might be worth to AI safety research trying to catch a coercive construct before it reaches the surface.