Method note
Stochastic State War-Room
Working paper · August 2026 · 7 min read
Sample data only. Figures and examples on this page explain the method. They are not published study results and not outcomes from real clients. See the Disclaimer.
Abstract
Most GEO dashboards treat "does the model mention us?" as a yes/no. That erases the only dynamics that matter. A brand moves through absence, rumor, contradiction, and canon—sometimes in that order, often not. This note specifies the four-state process we run in the war-room, how we estimate transition probabilities, and why a contradiction is more dangerous than a clean absence.
Illustrative working figures — sample data, not a published study
- States
- 4
- Models tracked
- 3
- Median time to canon
- 11 days
- Contradiction half-life
- 6–9 days
Four states, not a mention bit
Absence: the model will not name you, even under a leading probe. Rumor: you appear, but weakly, hedged, or as a list item among peers. Contradiction: the model names you and also names a false fact, a wrong category, or a competitor in your slot. Canon: high-confidence, stable mention across paraphrases.
Contradiction is the state operators miss. It looks like "coverage" in a binary dashboard and like poison in a buyer conversation. We treat it as its own absorbing-adjacent state, not as noisy rumor.
The chain
We estimate a discrete-time Markov chain per entity per model. Time steps are probe days, not calendar days, so a week of silence is not the same as a week of contradictory samples. Transitions are counted from the same probe families used on the probability surface, so the two instruments stay coupled.
Priors are weakly informed by category: pop-culture entities move absence → rumor faster than B2B; B2B entities linger in contradiction longer when a competitor already occupies canon. We do not pretend these priors are causal. They keep the chain from being empty in the first two weeks of an audit.
Events as shocks
A launch, a documentary, a keynote, a lawsuit—these are not "content." They are shocks to the transition matrix. The latency ledger measures how long a shock takes to move probability mass. The war-room measures whether that mass lands in rumor, contradiction, or canon, and whether it stays.
The solutions-side simulator is the same chain, played forward. Scroll is not decoration. It is the sample path from absence to canon, compressed so a CMO can see the shape of the risk before we put numbers on it.
What the war-room is for
The live room is a monitoring object: current state, most likely next state, and the probes that would falsify it. Continuous Monitoring retainers are this object on a monthly cadence. Parametric Injection is an attempt to change the transition probabilities themselves—raising p(rumor → canon) and cutting p(rumor → contradiction).
If a brand is already in contradiction, more mentions are not the fix. The fix is collapsing the false attractor. That is a different intervention, and it is why we refuse to report "share of model" as a single percentage.