Zeigarnik Task Tracking: The Math of Pending Cognitive Tension
In 1927, Soviet psychologist Bluma Zeigarnik observed that waiters could remember complex orders that had not yet been paid for, but immediately forgot those orders once the bill was settled. This phenomenon—the Zeigarnik Effect—reveals that the human brain retains incomplete or interrupted tasks more vividly than completed ones.
Traditional vector search engines have no concept of task completion; they treat resolved logs the same as outstanding blockers. Spector builds the Zeigarnik Effect directly into its scoring funnel.
The Zeigarnik Score Multiplier
Spector tracks task state using metadata flags. Any memory node representing an open task, pending question, or unfulfilled goal is given a dynamic priority multiplier.
The Zeigarnik priority boost \(Z_i\) is formulated as:
\[Z_i = 1.0 + \gamma_{zeigarnik} \times e^{-\alpha \Delta t}\]
Where:
- \(\gamma_{zeigarnik}\) is the maximum Zeigarnik boost factor.
- \(\Delta t\) is the time elapsed since the task was created or last updated.
- \(\alpha\) is a decay coefficient ensuring that ancient, abandoned tasks slowly lose priority.
- If the task is completed (resolved), the flag is toggled, and \(Z_i\) drops instantly to \(1.0\).
Maintaining Goal Alignment
By multiplying the final retrieval score of a memory by \(Z_i\), Spector ensures that pending goals always surface during context retrieval.
Open Task: [Fix Build Error] (Score = Raw * Z_i = 0.75 * 1.5 = 1.125) -> High Priority
Closed Task: [Fix Build Error] (Score = Raw * 1.0 = 0.75) -> Low Priority
This math prevents agents from losing context of what they were trying to do when navigating complex multi-step workflows.