Wallet backbone segmentation, signal heuristics, and a labeled dataset
Three research-grounded capabilities, built additively on top of xrpl.network's existing transaction and token data: (1) classifying wallets into a persistent backbone vs short-term cohorts, (2) deterministic token manipulation heuristics anchored to published detection thresholds, and (3) a labeled dataset that grows every time a signal is computed. All figures below are either live computed values or clearly marked as unavailable — nothing here is fabricated.
Wallet backbone segmentation
Persistently engaged wallets — active across multiple weekly snapshots — form a stable backbone inside the network's giant strongly connected component, transacting in reciprocal, loop-dominated patterns. Short-term wallets enter and exit, and their imbalances, short transaction loops, and outward flows are early (statistically significant, not predictive) indicators studied in the literature. Method: Flow Index / bow-tie decomposition of the XRP network, arXiv:2106.10012; backbone-persistence framing from UBRI research (Sankaewtong & Ikeda, Blockchain Kaigi 2025).
Backbone segmentation — disabled on this deployment
Set INTEL_ENABLED=true to activate. Once enabled, this section shows live persistence classification computed from the same transaction pipeline used elsewhere on xrpl.network.
Token manipulation heuristic signals
Deterministic, explainable z-scores of 24h price and volume vs a 20-day EWMA reference, tiered into none / watch / elevated / high. Thresholds are anchored to published research — practical pump-event detection (arXiv:2503.08692), moving-average feature sets (La Morgia et al., arXiv:2105.00733), and graph/interaction features (PUMPWATCHER, UCL). These are heuristic risk signals, not accusations or ML predictions.
Signal heuristics — disabled on this deployment
Set INTEL_ENABLED=true to activate. Once enabled, this section shows live z-score tiers computed from the existing token data source.
Labeled dataset accumulation
XRPL has a documented labeled-data gap for on-chain research (UBRI: Dalgıç & Bahtiyar). Every backbone snapshot and token signal computed by this workstream is persisted as an observation in intel.observations, so the dataset grows automatically with usage. Analyst, community, and external labels can be attached later in intel.labels — an export path is a future licensable asset, not yet built.
Dataset — disabled on this deployment
Set INTEL_ENABLED=true to begin accumulating observations.
API endpoints
x402 payment-gated for programmatic callers (browser/session traffic on this site is always free). Full list in /.well-known/x402.json.
GET /api/intelligence/backbone — wallet backbone snapshot
GET /api/intelligence/signals — all token signals
GET /api/intelligence/signals/[token] — single token signal (e.g. /api/intelligence/signals/SOLO)
· Sankaewtong & Ikeda, wallet backbone persistence (UBRI, Blockchain Kaigi 2025) — cited in project research notes.
· Flow Index / bow-tie decomposition of the XRP network — arXiv:2106.10012
· Practical pump-event detection thresholds — arXiv:2503.08692
· Moving-average manipulation feature set, La Morgia et al. — arXiv:2105.00733
· PUMPWATCHER graph/interaction features (UBRI, UCL) — discovery.ucl.ac.uk/id/eprint/10212382
· XRPL labeled-data gap — UBRI: Dalgıç & Bahtiyar, cited in project research notes.