IntentSpider - The Fluid Language Web for Textual Tension and Prediction

IntentSpider is an on-device, privacy-first predictive text engine built in C++ STL, modeled as a human's intent graph. Prediction and graph updates operate on-device. Each edge carries a coefficient and transmission fidelity value deciding whether it is TF or capture type, sending signal toward the human's next selection. Roles are reversible. IntentSpider uses personalized PageRank through local push, initialized from recent words using fan dispersion rather than equal initialization values. IntentSpider addresses blindness to the valency by Valence Gated Signed Diffusion using typing interval patterns. The Necessity Gating Algorithm controls whether the ranked word is displayed. Hexarousal measures how much typing rate differs from that human's ordinary rate. Others include shared reinforcement, increasing diffusion radius, sudden suppression, settling, typing substate search, and combined behavioral values, motivated by the Orb Spider, Wandering Spider, and struggling isopod prey. A 2D process converts sentence diffusion outputs into paths. Across 500 events, IntentSpider achieved 14.40% first rank and 27.80% Top 3 accuracy. Every shared event set had greater measured first rank and Top 3 accuracy, outperforming Google Gboard, Samsung Keyboard, Microsoft SwiftKey, and OpenBoard. Against Gboard on 187 events, IntentSpider reached 17.65% vs 7.49% first rank and 33.16% vs 14.44% Top 3 accuracy. Prediction produced 2.4 ms median time and 3.66 ms 95th percentile. Largest changes appeared when the recent word seed was reduced to 1 and the Necessity Gating algorithm removed. This depicts uncertainty in text prediction, potentially altering, simplifying, or negatively influencing human communication. IntentSpider Webnet is one approach to this issue.