Measuring plan stability
Turn compaction-drift intuition into trackable metrics. Measure plan stability and context efficiency over time.
Define drift in measurable terms
Track when an agent revisits decisions or contradicts prior constraints. These events are proxies for plan instability and can be counted across sessions.
Context efficiency metrics
Measure the ratio of retrieval tokens to raw file tokens. As that improves, compaction should happen later in the session and recovery becomes faster.
From anecdote to signal
When drift declines and compactions are avoided, teams feel it immediately. Quantifying that shift makes the impact visible to leadership and budgets.