Eight views of micro-drama Issue No. 9. The composite tells you where each brand sits. The Stability layer tells you whether the ground holds.
Each brand carries a Structural Brand Power Index (SBPI) score from 0 to 100, weighted across five dimensions:
Borrowed from control theory by way of the lyapunov-research project. Picture the category as a landscape with valleys. A brand’s composite score is its altitude. The Stability layer asks a different question: how steep is the slope under that brand’s feet? A high-altitude brand on flat ground is in a different position than a high-altitude brand on a 30-degree slope. The second one falls faster when something shocks the system.
We measure how far each brand sits from the cohort’s center of gravity, weighted by the dynamics we’ve observed across W10 through W17. That distance is V. Lower V means the dynamics support the brand’s current position. Higher V means the position depends on conditions that may not hold.
The model was fit on eight weeks of W10–W17 panel data. Validation 0.714 reads like a B− on a 0–1 scale of how well the model’s predictions track real movement. Two checks must clear at once: V must decrease along observed transitions, and lower V must predict which brands recover after a shock. Acceptance threshold is 0.65.
The penalty formula:
adjusted = composite × (1 − 0.05 × V / basin_radius) with basin_radius = 4.41
Carried forward from W17. The 5% per unit haircut is intentionally small, it surfaces rank inversions only where stability really diverges from composite.
Twenty-one brands across the global micro-drama vertical. Public signals only: trade press (Variety, Hollywood Reporter, Music Ally, Deadline), regional press (BigGo Asia, Asia Financial, IWMBuzz), platform-disclosed metrics, regulatory filings, and verifiable announcements within the April 27–May 3 window. No proprietary or unverifiable data.
The Stability layer is an additional reading on the existing scoreboard, not a replacement for it. It surfaces brands whose composite reads stronger than their dynamics support, and brands whose composite undersells how firmly they’re holding their ground. We aren’t predicting failure. We’re measuring how much the dynamics have to be wrong before this week’s score changes.
Validation 0.714 is comfortably above the 0.65 threshold but below 1.0, the model misses some movement. The single-V fit treats the whole 21-brand panel as one valley; the per-cluster fit (which lifts coverage to 0.923) reveals two valleys, one for active competitors and one for laggards. Both readings are reported.
Full Stability methodology: lyapunov-research.pages.dev
W17 v2 stability edition (the prior week with this layer applied): microco-w17-stability.pages.dev