Loading data...
Nexus Glass produces signal, not dashboards. This page is the map of how that signal is built, from the four raw inputs we own, through ten engines we run, into one attributable conviction your models can consume.
Most providers ship dashboards. We ship signal. The four properties below are how we keep that promise testable, every block, every tick.
Every measurement enters through a connection we own. 24 venue WebSockets, our own Bitcoin full node, the complete Deribit options chain. No aggregator layer between the tape and the metric.
Cascade probability, GEX surface, whale clustering, UTXO cohorts. All of it runs on our infrastructure, from inputs we already own. No outsourced math, no wrappers around wrappers.
If a metric cannot be reproduced from inputs we hold, it does not ship.
Every synthesis (conviction, regime, cascade risk) ships its drivers and their weights alongside the answer. Your risk system can argue with us in the same language we use internally.
The three properties above are worthless if the answer arrives after the move. End-to-end (across venue tape, ingest, engines, synthesis, edge) we hold the budget below one second. P95 sits at 38 milliseconds.
Everything downstream is derived from the four streams below. They feed every engine and every endpoint in the platform.
Open interest, funding rates, mark prices, liquidation tape, long/short, and per-venue depth across 24 exchanges. Raw venue clock, no smoothing.
MVRV, SOPR (LTH / STH / entity-adjusted), HODL waves across 10 age bands, NUPL, Thermocap, exchange flows. Every metric computed on our own Bitcoin full node.
Gatheral SVI surface across the full Deribit chain, constrained by Fukasawa no-arbitrage bounds. Per-strike GEX, max pain, term-structure skew. Recalibrated every 300 seconds.
Entity-resolved wallets scored on 6-month Sharpe-style alpha persistence. Syndicate detection surfaces 3+ wallets moving in coordination across venues, on-chain and centralized combined.
Each engine reads a slice of the input fabric and emits a typed signal. They run in parallel; the synthesizer joins their outputs.
The synthesizer is not a dashboard. It is a regime-weighted scoring function that returns a number, its drivers, and the regime it was scored under.
Each engine carries a different weight under each market regime. Cascade reads dominate during deleveraging; GEX dominates near gamma flips; on-chain dominates during basing.
Every payload returns the contribution of every engine alongside the synthesized score. Drop the synthesizer entirely if you’d prefer to weight inputs yourself.
Same inputs, same answer. Replay a moment in market history with the same engines and get the same conviction. Versioned semver; no silent drift.
Every engine is reachable over REST and streamable over WebSocket. One auth, one key namespace, one versioned spec.
# curl https://nexusglass.live/api/v1/intel/summary?symbol=BTCUSDT
import requests
r = requests.get(
"https://nexusglass.live/api/v1/intel/summary",
params={"symbol": "BTCUSDT"},
headers={"X-API-Key": "nxg_…"},
).json()
# → conviction: -72, regime: Deleveraging, confidence: 91%
# drivers: CascadeGuard, LeviathanTrack, GenesisLens
r["conviction"], r["regime"], r["drivers"]P95 across every REST endpoint, measured against production traffic.
Three priority lanes on the WebSocket. Each is throttled independently. Your cascade alerts never queue behind a heartbeat.
Every endpoint shape is versioned. Deprecations announce months ahead; the spec never breaks underneath you.
Every payload is reproducible from inputs we still hold on disk. Same answer for the same moment, every time.
The same primary-source feed runs every endpoint at every tier. The price is what changes, and the rate limits.
Nexus Glass produces the signals. Your models. Your edge.