Command centerPredictive risk floorSystem 02 / 09

QUANTPROP // VERIFIED CASE STUDY

Real-Time Predictive Risk Engine

Predictive edge. Controlled risk.An event-driven sports analytics platform with probabilistic modeling, automated settlement, CLV measurement, and regression watchdogs.

AI CAPABILITY PLANEWhere intelligence operates—and where authority stops

Monte Carlo simulationNegative-binomial modelingRegression watchdogs

Models may score and simulate opportunities; circuit breakers and human controls retain authority over promotion and financial execution.

01 // SYSTEM TOPOLOGY

Connected architecture

Event-driven decision platform spanning source acquisition, feature computation, probabilistic scoring, governed recommendation delivery, settlement, and closed-loop model monitoring.

PlatformData planeEdge / externalAuthority / control

01 // Source acquisition

Market feedsSportsbook lines + price movement

WHYEstablishes trusted analytical inputs

MLB / StatcastPlayer, game + pitch features

WHYEstablishes trusted analytical inputs

Weather contextVenue + game conditions

WHYEstablishes trusted analytical inputs

02 // Orchestration

Airflow control plane24 scheduled and event-driven DAGs

WHYMakes pipeline execution repeatable

Quality contractsFreshness, schema + cohort checks

WHYMakes pipeline execution repeatable

03 // Operational data

TimescaleDB27-table analytical system of record

WHYStores current and historical decision state

RedisHot features + execution state

WHYStores current and historical decision state

04 // Model intelligence

Feature blendingForm, matchup, park + market signals

WHYTurns features into probability estimates

Probability enginesNegative binomial + Monte Carlo

WHYTurns features into probability estimates

05 // Decision governance

Edge + tier policyGold, Silver + Bronze contracts

WHYConstrains recommendations and authority

Circuit breakersRegression and hit-rate gates

WHYConstrains recommendations and authority

Recommendation recordImmutable tracked cohort

WHYConstrains recommendations and authority

06 // Measurement loop

Settlement serviceWin, loss, push + void

WHYMeasures outcomes against recommendations

CLV analyticsPrice-quality measurement

WHYMeasures outcomes against recommendations

07 // Observability

Grafana20-panel operational evidence

WHYMakes system health and outcomes visible

PrometheusPipeline and service health

WHYMakes system health and outcomes visible

SupersetAnalytical exploration

WHYMakes system health and outcomes visible

08 // Feedback + authority

Model watchdogsCalibration, cohort + promotion gates

WHYKeeps final authority with operators

Operator controlAlerting + manual execution authority

WHYKeeps final authority with operators

GOVERNED OUTCOMEA narrow proof loop aligns the alert, stored recommendation, watchdog cohort, CLV record, settlement, and dashboard.Ingest → Model → Alert → Settle → Verify

02 // PROOF LIBRARY

Evidence gallery

Verified // reviewed 2026-07-27Live model track record

Read-only Grafana evidence showing 3,145 graded recommendations, realized win rate, outcome distribution, and performance segmented by tier and market over the selected 30-day view.

QuantProp private operational Grafana dashboard; captured through a temporary least-privilege Viewer account and sanitized of infrastructure identifiers

Verified // reviewed 2026-07-27Daily performance telemetry

Operational trend view exposing day-level variance instead of presenting only aggregate outcomes; adjacent tables retain market and tier context.

QuantProp private operational Grafana dashboard; read-only Viewer capture with infrastructure and account details excluded

Verified // reviewed 2026-07-27Airflow orchestration inventory

Live read-only inventory of the QuantProp DAG family, including feature computation, calibration, CLV, contract-cohort monitoring, and daily decision workflows.

Private Airflow control plane; temporary Viewer account with account identity redacted

Verified // reviewed 2026-07-27MLflow promotion-gate history

Forty tracked model-validation and promotion-gate runs demonstrate an auditable evaluation history rather than an unversioned model claim.

Private MLflow experiment tracker; host and infrastructure identifiers excluded

Verified // reviewed 2026-07-27Orchestration control plane

Sanitized repository facts spanning 24 DAG files, ingestion, model controls, settlement, CLV, and watchdogs.

Local QuantProp DAG source and SYSTEM.md; not live telemetry

Verified // reviewed 2026-07-27Model promotion gates

Public rendering of verified shadow, calibration, cohort-health, warning, alert, and manual-authority logic.

Local QuantProp watchdog source; synthetic presentation of verified logic

Architecture mapAvailable now

Sanitized logical topology with evidence states.

Dashboard / DAGAvailable now

Sanitized repository-derived dashboard or orchestration evidence is included above.

Video / demo clipCapture pending

A short captioned product proof is still required.

Product screenshotAvailable now

Reviewed product-interface evidence is included above.

PRIVATE REVIEW ROOMAuthentication boundary prepared

Production endpoints and sensitive artifacts stay disconnected until an identity provider and access policy are explicitly configured.