Step 4 Β· Causal Confirmation
Governance Trace β Learner ex_junyi_graph_135350
0 interactions Β· Interaction 25 is the first division_4 encounter: T_realized share spikes from 3.3% β 25.6% (8Γ increase). Interaction 30 has no DAG edges and returns to floor (12Γ drop). Each cell below is colored by transfer activation level.
Transfer-active count is threshold-dependent. The contribution distribution is a smooth continuum (no natural gap): 0/0 exceed the descriptive 0.03 threshold; 0/0 exceed the 0.08 highlighting cutoff used below. The primary ADC classification does not rely on this count β it derives from the sealed alpha_floor (0.01) + signal_ratio (β₯0.08) criteria.
β PartialSingle-learner causal trace β strong mechanism, narrow scope
The 8Γ T_realized spike on interaction 25 is a real mechanism observation tied to a specific DAG edge activation, not noise. The page's scope is one learner β the sealed re-run widens this to a sample, and the same mechanism is what the shuffled-DAG control on /dashboard/audit tests at the population level.
Flips after:- Mechanism re-demonstrated on a sample of unseen-user traces
- Cross-trace spike incidence reported alongside the single-trace example