Cohort flow Sankey
every transfer aggregated by sender → receiver display cohort, full chain historyLoading flow graph…
Flow matrix
pairwise volume between display cohorts (heat = AGRS volume)Loading matrix…
Top flow corridors
largest sender-cohort → receiver-cohort pairs (≥ 0.05% of total volume)Loading…
Net cohort flow over time
positive = cohort accumulating · negative = cohort distributing · monthly bucketsLoading…
Supply velocity
monthly volume / circulating supply — high values = active turnoverLoading…
Top flow hubs
wallets with the largest aggregate flow — network centrality leadersLoading…
GPU graph analytics
Wallet behavioural map
Node2Vec embedding (16-dim, GPU) → UMAP 2D projection. Proximity = similar token-flow patterns, not proven coordination.Wallet embedding not yet computed
Network influence (PageRank)
Weighted directed PageRank on the full transfer graph. Higher score = more value flow passes through this wallet.Loading…
Personalised propagation
Personalised PageRank from each anchor cohort. Cohort shares = where the recursive influence concentrates, NOT where direct transfers settled.Loading…
Discovered wallet communities
Louvain modularity clustering on the undirected weighted transfer graph. High internal volume = probable co-control, but not definitive.Loading…
methodologyNode2Vec walks the transfer graph 5 times per node (12-step biased random walks), trains 16-d SkipGram embeddings on GPU, and projects to 2D with UMAP. Louvain optimises modularity over the symmetrised weighted graph. Personalised PageRank uses α = 0.85 with the anchor cohort as the teleportation distribution. All three are heuristics that surface hypotheses about co-flow patterns; none of them prove ownership, coordination, or direct token movement.