Network Intelligence & Analytics
Real-time CDR processing, quality intelligence and consensus-verified network benchmarks
This paper describes the VPX Analytics Node architecture - the data processing layer that ingests raw CDR data, enriches it with network metadata, computes quality and fraud signals in real time and commits consensus-verified intelligence aggregates to VPX Protocol. It covers the streaming ingestion pipeline (<100ms ack, <2s enriched availability), the 14 quality dimensions tracked per route, the anomaly detection models, the on-chain benchmark commit cycle and the Analytics Node staking and rewards economics that align operator incentives with data accuracy.
CONTENTS · 10 SECTIONS
1. Introduction
In traditional wholesale telecom, network intelligence is produced after the fact. CDRs are exported in batches, reconciled overnight and presented to NOC and finance teams as reports that are 12-48 hours out of date by the time they are read. Quality degradation, fraud signatures and traffic anomalies are detected retrospectively, often after thousands of customer-affecting events have already occurred.
VPX Analytics Nodes replace this batch-and-report model with a real-time streaming intelligence layer. Every CDR is ingested, enriched and scored within seconds of call completion. Quality benchmarks, fraud signals and traffic intelligence are produced continuously and made available via API for any downstream system - Routing Nodes consuming quality scores, NOC dashboards consuming anomaly alerts, billing systems consuming enriched CDRs, the VPX Protocol consuming consensus-derived network benchmarks.
This paper specifies the Analytics Node architecture, the data pipeline, the quality and anomaly models, the on-chain commit cycle and the staking and rewards economics that align Analytics Node operators with data accuracy and timeliness.
2. Role in the VPX Ecosystem
Analytics Nodes sit between the raw event sources (Gateway Nodes, connected carriers) and every downstream consumer of network intelligence. They are the data processing backbone that turns raw CDR streams into the structured, scored, consensus-verified intelligence that other nodes and applications depend on.
- Ingest CDRs from Gateway Nodes and connected carriers in real time.
- Enrich CDRs with network metadata: route path, node hops, codec, fraud risk.
- Compute per-route, per-destination and per-carrier quality scores continuously.
- Detect traffic anomalies, quality degradation and fraud-correlated signatures.
- Commit consensus-derived benchmark aggregates to VPX Protocol hourly.
- Serve enriched CDRs and quality data via REST API, WebSocket and SFTP export.
Routing Nodes consume Analytics Node quality scores to make quality-weighted routing decisions (see VPX Protocol §5.1 cache architecture in the AI-Powered Routing Intelligence whitepaper). Fraud Nodes consume Analytics Node anomaly signals as one input to fraud consensus. Carrier NOC and billing teams consume Analytics Node APIs directly. Every dashboard, quality SLA report and billing reconciliation in the VPX Ecosystem ultimately derives from Analytics Node output.
3. Streaming Ingestion Pipeline
Analytics Nodes are designed for continuous high-throughput ingestion - not batch processing. The ingestion and enrichment pipeline is built on the proven streaming stack the telecoms industry already runs at scale (Apache Kafka for durable, ordered event transport; Apache Flink for stateful stream processing) rather than a bespoke engine, so the novel work is the telecom enrichment and scoring, not reinventing stream processing. Each node processes CDRs as a stream, with target acknowledgement latency under 100ms and full enrichment availability under 2 seconds from call completion (design targets ahead of mainnet, not measured production figures).
3.1 CDR Submission
CDRs are submitted via authenticated HTTPS POST or via persistent WebSocket connection. Each CDR is signed by the submitting Gateway Node or carrier using its on-chain identity key. The Analytics Node verifies the signature, validates the CDR schema and acknowledges receipt within 100ms.
Duplicate detection is performed at ingestion using a content-addressed hash of the call leg identifiers. Duplicate CDRs (e.g. from retry logic on the submitter side) are detected and merged rather than double-counted.
3.2 Enrichment
Each accepted CDR is enriched with network metadata derived from the call leg context: the full route path through the VPX Network, the sequence of node hops, the negotiated codec, the originating and terminating carrier identities, the MNP (Mobile Number Portability) status of the destination and the current fraud risk score for the originating and terminating numbers.
Enrichment completes within 2 seconds of CDR receipt. The fully enriched CDR is then available via API, WebSocket stream and is queued for downstream scoring and aggregation.
3.3 Throughput
Each Analytics Node is engineered for sustained throughput of 100,000 CDRs/second with burst capacity to 500,000 CDRs/second per node. At target network scale (multiple Analytics Nodes operating in parallel), the aggregate ingestion capacity exceeds the projected peak traffic of the full VPX Network with substantial headroom.
4. Quality Dimensions
Analytics Nodes track 14 quality dimensions per route, per carrier and per destination. The five primary dimensions (ASR, ACD, PDD, NER, MOS) are exposed to Routing Nodes as inputs to routing decisions. The remaining nine dimensions are tracked for analytics, benchmarking and anomaly detection. The full quality dimension list is shared with the AI-Powered Routing Intelligence specification (§4):
- ASR (Answer-Seizure Ratio) - percentage of call attempts that are answered.
- ACD (Average Call Duration) - mean duration of answered calls.
- PDD (Post-Dial Delay) - time from call initiation to first ringback or answer.
- NER (Network Effectiveness Ratio) - percentage of calls that reach the destination network successfully.
- MOS (Mean Opinion Score) - estimated voice quality score (1.0-5.0).
- SER, ABR, SRTT, Jitter, Packet Loss, Codec Distribution, CLI Pass-Through Rate, Fax Success Rate, DTMF Reliability.
Each dimension is computed continuously at multiple time windows: 1-minute rolling, 1-hour rolling, 24-hour, 7-day and 30-day. Short-window scores feed real-time routing decisions and anomaly detection; longer-window scores feed benchmarking, SLA reporting and trend analysis.
5. Anomaly Detection
Anomaly detection runs continuously on the enriched CDR stream. The detection layer combines statistical baselines (per-route, per-destination, per-carrier and per-time-of-day) with learned models that capture multi-dimensional anomaly signatures.
5.1 Statistical Baselines
For each tracked entity (route, destination, carrier), Analytics Nodes maintain a rolling baseline distribution for each quality dimension. Deviations from baseline are scored using a robust z-score with outlier-resistant variance estimation. Alerts are raised when a quality dimension exceeds threshold deviation for sustained windows (typically 3+ standard deviations sustained for 5+ minutes).
5.2 Multi-Dimensional Patterns
Single-dimension thresholds catch obvious degradations but miss subtle fraud signatures and quality issues that manifest only across multiple dimensions simultaneously. A learned model (gradient boosted decision tree) scores the joint distribution across all 14 quality dimensions plus traffic volume features, surfacing anomalies that are not visible in any single dimension.
Multi-dimensional anomalies are particularly useful for early fraud detection. A traffic pattern that combines a sudden volume spike, an unusual destination concentration, a reduced ACD and an elevated PDD is far more diagnostic of an attempted fraud event than any single feature alone.
5.3 Anomaly Output
Confirmed anomalies are published via webhook to subscribed consumers (carrier NOC teams, Fraud Nodes, internal monitoring) and recorded as on-chain events on VPX Protocol. Each anomaly record carries the full feature vector and the explanatory features driving the alert - eliminating the "black box alert" problem that characterises proprietary anomaly systems.
6. On-Chain Benchmark Commits
A core function of Analytics Nodes is to produce consensus-verified network intelligence the entire VPX Ecosystem can rely on as authoritative. Two constraints shape how this is done: the raw CDR volume is far too large to place on-chain, and per-node scores would be vulnerable to single-operator manipulation. The commit cycle solves both by publishing a compact, cryptographically committed aggregate (a hash-rooted summary, never the raw records - which also keeps personal data off-chain), agreed across independent nodes so that no single operator can move the number. The trimmed-median-of-signed-candidates described below is the trust mechanism at launch; a cryptographic verifiable-compute proof - proving the aggregate was computed correctly over tamper-evident data - is the forward-looking upgrade that takes the benchmark from trust-minimised to trustless.
6.1 Hourly Commit Cycle
Once per hour, every active Analytics Node computes a benchmark aggregate covering all routes, destinations and carriers it observed during the preceding hour. Each Analytics Node signs and submits its aggregate to VPX Protocol as a benchmark candidate.
6.2 Consensus and Median Selection
VPX Protocol collects benchmark candidates from all participating Analytics Nodes for each metric. The consensus benchmark is derived from the candidate set using a manipulation-resistant aggregation function: a trimmed median that excludes the top and bottom 10% of candidate values before taking the median of the remaining range. This neutralises both honest outliers (faulty pipelines, partial outages) and adversarial outliers (attempted benchmark manipulation). The aggregated value is committed on-chain as the canonical network benchmark for that hour. The specific trim percentage is a network parameter governable by DAO vote post-mainnet.
Analytics Node operators whose candidate values fall outside the trimmed band repeatedly are flagged for review. Persistent deviation - whether from faulty pipelines, misconfigured nodes or attempted manipulation - results in slashing under the Analytics Node staking model (see §8).
6.3 Auditable History
Consensus benchmarks are immutable once committed. Every routing decision, fraud consensus event and SLA report in the VPX Ecosystem can be traced back to the specific hourly benchmarks that informed it. This is the auditability foundation that distinguishes VPX intelligence from proprietary analytics platforms where the underlying data is unauditable.
7. API and Data Delivery
Analytics Node outputs are available to authorised consumers via multiple channels designed for different integration patterns:
- REST API - per-route, per-carrier, per-destination quality and benchmark queries with sub-second response.
- WebSocket Stream - real-time enriched CDR feed and anomaly alerts.
- SFTP CDR Export - billing-grade per-second CDR delivery in CSV, JSON or XML.
- Webhook Alerts - push delivery of anomaly events and SLA breach notifications.
- On-Chain Benchmark Access - direct query of consensus benchmarks via VPX Protocol.
API authentication uses OAuth 2.0 with on-chain identity binding. Rate limits and quota are enforced per authenticated consumer and metered against the consumer's VPXN balance under the Token Economy consumer anti-abuse framework (see Token Economy v3.5 §11.5 Anti-Abuse Measures).
8. Analytics Node Staking and Rewards
Analytics Node operators participate in the VPX Token Economy under the same staking and rewards framework that governs Routing Nodes, Fraud Nodes and Gateway Nodes. Analytics-specific parameters and the rationale for them are summarised here; the full economic specification is in the VPX Token Economy whitepaper (v3.5).
8.1 Minimum Stake
Analytics Node operators stake VPXN tokens to participate in CDR processing and benchmark commits. Stake amount is set in the VPX staking portal and may be adjusted by governance vote post-mainnet. Stake is the operator's economic skin in the game: it backs the operator's commitment to accurate, timely data processing.
8.2 Reward Sources
Analytics Node operators earn rewards from two sources: (a) a share of the CDR processing fees paid by carriers and Gateway Nodes for ingestion, enrichment and API access, and (b) a share of the Node Operator Rewards allocation of VPXN emissions (Token Economy v3.5 §2 - Node Operator Rewards: 35% / 350M VPXN, distributed across all node operator classes per the epoch reward mechanics in §6 and the Analytics-specific performance scoring in §7.3).
8.3 Reward Weighting
Reward distribution within the Analytics Node class is weighted by verified work: CDRs ingested, enriched and delivered on schedule. Nodes whose benchmark candidates consistently fall within the trimmed consensus band earn a quality bonus. Nodes whose candidates fall outside the band are not eligible for the quality bonus and, if deviation persists, are slashed.
8.4 Slashing
The Analytics Node slashing tier aligns with the canonical three-tier slashing schedule (25% / 75% / 100% of stake, see Token Economy v3.5 §10 Slashing Economics; the canonical three-tier schedule is published in §10.2 Slashable Offences with the comparative analysis and trigger summary in §10.2.1). 25% applies to first-offence sustained baseline deviation or repeated downtime; 75% applies to severe or repeated offences (deliberate benchmark manipulation, double-signing); 100% applies to terminal offences such as coordinated benchmark manipulation across multiple Operators. Slashed stake is burned (see Token Economy v3.5 §10.4 Slash Destination), removing supply from circulation. The 3× documented-damages multiplier referenced in earlier protocol literature is NOT active at publication date pending damages-oracle specification (see TE §16.6).
9. Operational Specifications
The following operational parameters are the engineered design targets for Analytics Nodes in the production VPX Network. They are scheduled targets ahead of mainnet, not measured production guarantees:
The 99.9% figure is the published operational target for Analytics Node operators competing for the quality bonus in §8.3. It is not the eligibility threshold. The canonical eligibility floor is the 90% of the epoch defined in Token Economy v3.5 §6.1, and any conflict between the two figures resolves in favour of TE.
10. Conclusion
Analytics Nodes provide the streaming intelligence layer that the entire VPX Ecosystem depends on. They turn raw CDRs into real-time enriched data, scored quality benchmarks and consensus-verified network intelligence - replacing the 24-48 hour reporting lag of traditional telecom analytics with sub-2-second freshness.
For Analytics Node operators, the economics are aligned with the work: stake VPXN, process CDRs accurately and on time, contribute benchmark candidates that track the consensus and earn proportional rewards from both processing fees and Network Operations emissions. Persistent deviation or malicious behaviour is slashed under the canonical schedule.
For consumers of network intelligence - carriers, NOC teams, billing systems and downstream nodes - the value is auditability: every quality score, every benchmark, every anomaly alert is traceable to consensus-verified source data, eliminating the black-box problem of proprietary analytics platforms.
This document is published by VoicePro Plus Ltd for informational purposes only. It does not constitute investment advice, a prospectus, or an offer of securities. The VPX Ecosystem is under active development; specifications described herein are subject to change. VoicePro Plus Ltd is registered in England and Wales (Company No. 14520016). Registered office: 128 City Road, London, EC1V 2NX.
This document is published by VoicePro Plus Ltd for informational purposes only. It does not constitute investment advice, a prospectus, or an offer of securities. The VPX Ecosystem is under active development; specifications described herein are subject to change. VoicePro Plus Ltd is registered in England and Wales (Company No. 14520016). Registered office: 128 City Road, London, EC1V 2NX.
Ready to learn more?
Speak to our team about the VPX Ecosystem and how it integrates with your existing infrastructure.
Contact Our Team