QORE is built as an event-driven system, multitenant by design, with container isolation. Each component does one thing and does it well. Every decision is logged, every execution can be reconstructed.
The architectural choices favor production stability and measured latency on critical paths. No experimental components in hot paths, no trend-of-the-day — only what has proven to hold up in real, high-intensity scenarios. The result is a system that does not sacrifice performance for reliability, or vice versa.
Sub-millisecond event bus for inter-service communication. Native acknowledgement, minimal operational footprint. Each component talks via a dedicated per-client subject pattern.
Institutional-grade tick-by-tick market data feed, with automatic re-subscription at futures rollover. Latency measured from the real trade to the decision engine.
Execution layer decoupled from the strategy engine. Atomic sequence with pre-order what-if margin and stop attached from the first phase. Replaceable, testable, isolated.
Tick-by-tick persistence for backtesting and deterministic replay. Every decision can be reconstructed after the fact, independent of the live feed.
Every client is a dedicated service with external configuration. Adding a tenant means spinning up a new service: zero changes to the shared code.
Responsive dashboard with real-time updates. Authentication with refresh, multitenant middleware to auto-filter per-client information.
Each strategy produces a 0–100 score on the next tick. When multiple readings agree on direction, the signal gains weight. Thresholds configurable per symbol and per tenant.
The footprint chart shows aggressor volume per price level. The strategy computes bid/ask imbalance and identifies institutional absorption zones — where volume hits but price does not move, a sign of level defense.
// Footprint snapshot — ES Z6 @ 14:32:11 level bid_vol ask_vol imbal 5418.50 141 62 2.27× 5418.75 2,847 12 237× ABSORPT 5419.00 28 186 0.15× 5419.25 14 94 0.15× ───────────────────────────────────────── signal: LONG · score: 87 edge: absorption on bid 5418.75
VWAP is a static measure of institutional consensus on the session. When price loses it then reclaims it with rising volume, that is the market declaring: "the right level is up here, not down there".
// VWAP reclaim setup vwap: 5417.42 last_loss: 5417.00 (8 min ago) reclaim_bar: 5418.50 → 5419.25 reclaim_volume: +34% vs avg confirm_bar: pending ───────────────────────────────────────── action: arm LONG on confirmation stop: 5416.75 (rejection low) target: 3-phase trailing
Iceberg orders are large institutional orders fragmented to hide their real size. The strategy looks for the typical signatures: repeated refills on the same level, constant size, independence from the aggressor flow.
// Iceberg detection level: 5418.75 refills: 7 in 42s refill_size: ~150 contracts (σ 12) aggressor: SELL net −1,847 persistence: level active for 3.4 min ───────────────────────────────────────── read: institution defending 5418.75 action: LONG with stop below the level
Genuine momentum has three simultaneous markers: price acceleration, bar range expansion, volume peak. When the three indicators converge, it is rare. When they do converge, it should be followed.
// Momentum burst trigger price_accel: +2.8 σ (threshold ≥ 2.0) range_expand: +167% vs prior 5 volume_peak: 3,412 (p99 = 2,847) all_triggers: YES ✓ ───────────────────────────────────────── signal: LONG burst · score: 93 trail: phase 3 tight from 1.5 → 1.0 ATR
One pipeline serves N clients. Each tenant has its own orchestrator container,
its own authenticated middleware, its own automatic filter for tenant_id.
Nothing is hardcoded. Everything comes from external configuration.
No client-specific value is ever present in the code. Everything comes from
.env or YAML files with tenant_id as routing key.
Adding a tenant means spinning up a new container — without modifying a single line.
super_admin · infra_admin · strategy_admin · support · finance · trader · demo_viewer. Capabilities tracked section by section.
Every action leaves a trail: actual_user_id, effective_user_id, IP, timestamp. Login impersonation tracked in a dedicated audit log.
Four supported models: shared TWS, dedicated, paper-only, simulator. The choice is the tenant's, the infrastructure does not change.
Every execution passes through 17 verified steps: none is optional. If a step fails, execution is halted with a precise state.
Local state is compared with the broker state every 4 seconds. Discrepancies are detected immediately and handled with explicit rules.
The margin value always comes from the broker what-if. If the what-if fails for more than 3 minutes, the system enters "block" mode and refuses new orders.
Emergency button in the dashboard. All open positions are closed at market and all strategies disabled, logged in audit.
Automatic futures contract switch at 02:00 Italian time on volume crossover, with 1-day protection: no positions open during rollover.
Every release first goes through an identical staging environment (separate subnet). Only after passing E2E tests is the release promoted to production.
Full architectural documentation is available under NDA for technical partners and institutional clients. Get in touch to request it.
