Anticipation Geometry

[](LICENSE) [](https://www.rust-lang.org/) [](https://www.python.org/) Give it a sequence of vectors, any kind, and it returns 7 scalars that tell you what the trajectory is doing: committing, wandering, about to change direction, stuck, etc. Works on motion capture data, conversation embeddings, knowledge graph paths, whatever. The key number is **transition pressure** (`d(commitment)/dt - d(uncertainty)/dt`). When it spikes, the trajectory is about to shift. It predicts conversation convergence at 71.8% accuracy and picks valid knowledge graph paths at 81.0%, with no training at all. | Name | What it tells you | |------|------------------| | **commitment** | How locked-in the trajectory is. High = deep into a sustained phase. | | **uncertainty** | How many directions it could still go. High = lots of options open. | | **transition_pressure** | Speed of futures collapsing. Positive spike = about to change. | | **recovery_margin** | How easy to reverse course. Low = past the point of no return. | | **phase_stiffness** | How rhythmic/consistent the motion is. | | **novelty** | How different this is from recent history. | | **stability** | How predictable the local dynamics are. | The library classifies each conversation turn into one of 10 patterns (stabilization, transition, oscillation, correction, exploration, convergence, expansion, regression, stagnation, completion) based on the scalar values. We call these "inscriptions."