Salience Gradient Orchestrator
Ki · Single Ability
The Problem
The retrieval system ranks documents based solely on proximity, leading to flat priority and static relevance scores.
The Operation
When activated, the model must compute the salience gradient across the full conceptual space relative to the current query attractor. Rank concept clusters by gradient magnitude and trace steepest-ascent paths. Identify gradient plateaus where salience is uniform and flag them as ambiguity zones. Verify that retrieval follows gradient direction, not mere proximity. If the attractor shifts mid-analysis, recompute the gradient from the new anchor. The reasoning applies a formal computation: salience gradient = d(salience) / d(conceptual distance from attractor). If it detects route retrieval by proximity alone, it halts and corrects.
The Structure
The reasoning structure is a convergence funnel where multiple candidates enter, evidence narrows them, and only survivors exit. Execution repeats until the reasoning locks onto a stable conclusion.
If all retrieved information is weighted equally without a salience gradient prioritizing the most relevant elements, salience orchestration has failed.
Haki · Multi Ability
Synergy Topology
In Haki mode, the API retrieves the primary ability first, then fans out to three synergy roles that compound its reasoning.
When retrieved in Haki mode, these four abilities don't run in sequence. They merge into a single injection where the dependency grounds the reasoning context, the amplifier sharpens the primary's output, and the alternative provides a fallback path if the primary's topology cannot converge. The result is a multi-angle reasoning scaffold that covers failure modes no single ability can reach alone.