Platform
From data hunting to decision support.
TfGM SNV reduces manual correlation by fusing multimodal feeds, detecting anomalies with explainable rules, and guiding the response through a configurable playbook.
- Canonical model
- Explainable detection
- Auditable workflow
- governed model
- 1
- governed model
- rule signals
- 8
- rule signals
- escalation levels
- 3
- escalation levels
Layers
Separation you can see in the code.
The same boundary exists in the repository, not only on a slide.
-
Layer 1
Ingestion
A Connector contract per protocol, a Normaliser per payload, and a queue between them.
-
Layer 2
Canonical model
Versioned contracts; every entity is provider-neutral and exportable.
-
Layer 3
Orchestration
Detection, impact, workflow, escalation and comms — services with tests, not controllers.
-
Layer 4
Experience
Operator console, public status and MI, sharing one cached snapshot.
Everything we claim
The capability matrix.
Each line states what is available, what needs configuration, and what is roadmap.
-
DEMO-1 Available today
Single Network View
Fused multimodal map, KPI strip, live feed, mode and source health.
Limitation: Demonstration transport is simulated; operational geometry and logic are real.
/ops -
DEMO-2 Available today
Detection and impact assessment
Configurable rules, confidence scoring, explainable signals, impact calculation and optional AI explanation.
Limitation: Estimates use observed speed versus baseline until APC/ticket data is connected.
/ops/detection -
DEMO-3 Available today
Incident management
Lifecycle, playbook tasks, allocation, timeline, escalation, audit and AI-assisted alert drafts with human approval.
Limitation: Social/VMS/partner delivery adapters are configuration; web publishing is live.
/ops/incidents -
DEMO-4 Available today
Recovery and operational monitoring
Restoration against baseline, closure gate, post-incident review.
Limitation: Closure checklist is configurable per playbook.
/ops/recovery -
DEMO-5 Available today
Historical analysis and operational learning
Volumes, response times, impact minutes, corridor hotspots, CSV export.
Limitation: Scheduled report delivery and economic-impact analysis are roadmap.
/ops/analytics -
A1-A4 Available today
Multi-source detection, classification and explainability
Eight tuneable rules, confidence, signal/threshold evidence, affected services and a “why this fired” panel.
Limitation: Assets and permits are not yet resolved into impact; estimates use speed vs baseline.
/ops/detection -
A5 Configuration
Metrolink delay alerting
Service-delay and bunching rules with a SIRI/GTFS adapter path.
Limitation: Needs TfGM-approved Metrolink source data and agreed thresholds.
/ops/config/rules -
A6-A7 Available today
Explainable prioritisation and decision audit
Severity × impact priority with a visible explanation; append-only audit of decisions, overrides and AI suggestions.
Limitation: Queue order is severity-first; comparative ranking rationale is per incident.
/ops/incidents -
A8 Minor enhancement
Operator feedback on alert usefulness
Dismissal with a reason and AI-output helpful/not-helpful feedback are recorded on the timeline and audit.
Limitation: Detection-alert usefulness scoring and a feedback report are still enhancements.
/ops/detection -
A9 Available today
Basic mitigation guidance and playbooks
PB-ROAD, PB-TRANSIT and PB-INFRA steps create routed tasks; AI suggestions are advisory and require operator acceptance.
Limitation: Advanced optimisation and automatic mitigation remain later-phase.
/ops/config/playbooks -
A10 Roadmap
Governed MLOps and model learning
Detection is rules-based and explainable; BYOK assistance is separate and advisory.
Limitation: No model retraining pipelines, MLOps governance or feedback-driven optimisation; these remain roadmap.
/platform/roadmap -
A11 Minor enhancement
Post-incident impact analysis and learning
Impact calculation, timeline, closure, review notes and an AI PIR draft that must be human-accepted.
Limitation: No structured PIR form, benefits tracking or learning register.
/ops/incidents -
AI-BYOK Configuration
AI-assisted drafting with a customer-supplied model (BYOK)
Encrypted OpenAI, Anthropic and OpenAI-compatible provider profiles; detections explained, incidents briefed, alerts drafted, mitigations suggested and analytics narrated under human approval.
Limitation: Requires an administrator-supplied provider key and agreed provider DPA/residency. AI never changes status, publishes alerts, assigns work or trains on TfGM data; MLOps remains roadmap.
/ops/config/ai -
U1-U6,U9 Available today
Single view, lifecycle and automatic timestamps
Role-based view, transition-guarded lifecycle, acknowledge/resolve/close timestamps.
Limitation: Lifecycle states are fixed; U16 configurability is not yet exposed.
/ops/incidents -
U7-U8 Minor enhancement
Timestamp amendment and controlled handover
Assignment and timeline notes support ownership.
Limitation: No audited timestamp amendment or formal handover object.
/ops/incidents -
U10-U14 Minor enhancement
Map operations: create, search, delay display, layers
Map, incident pins, vehicle positions, corridor geometry and location search.
Limitation: No create-from-map, map search, delay symbols or layer toggles yet.
/ops -
U15-U16,U19 Roadmap
Map presets, configurable lifecycles and alert suppression
Role-based views and rules exist.
Limitation: Saved presets, per-mode lifecycles and known-event suppression are not built.
/platform/roadmap -
U17-U18 Available today
Impacted operators, partners and dynamic impact
Affected services/locations, operator metadata, impact recalculation.
Limitation: Partner contact lists are configuration; no automatic impact refresh.
/ops/incidents -
U20-U21 Roadmap
CCTV context and planned events
Evidence attachments exist for incident records.
Limitation: No CCTV integration or planned-event model.
/platform/roadmap -
U22-U26,U29,U32 Roadmap
Signals, VMS, assets and movable assets on the map
Asset, vehicle and corridor tables exist; UTC/source health is visible.
Limitation: No live signal/VMS strategy, asset proximity layer or movable-asset view.
/platform/roadmap -
U27,U30-U31,U33 Roadmap
Works, diversions and contextual weather layers
Weather and closure events can drive detection; impact recalculation exists.
Limitation: No permits/works model, diversion impact analysis or weather map layer.
/platform/roadmap -
D1-D10 Configuration
Unified Data Model, quality, provenance and portability
Source-neutral entities, versioned contracts, raw/canonical events, quality and export.
Limitation: Detailed open-standards mappings and the full UDM workbook mapping are to be completed per source.
/ops/network -
I1-I8 Configuration
Contextual integrations, APIs and open exchange
Connector/normaliser contracts, source registry, JSON/CSV/GeoJSON export.
Limitation: GTFS-R, SIRI, DATEX II and UTC adapters need credentials; VMS/CCTV/Waze and a public REST API are not present.
/ops/network/sources -
I9 Roadmap
Collaboration-tool integration
Internal timeline and mentions model.
Limitation: No Teams/Slack/ITSM connector.
/platform/roadmap -
R1-R6,R13 Available today
Operational reporting, outcomes and disrupted journeys
Incident export, response metrics, audit, multimodal delay aggregation, disrupted-journey estimate.
Limitation: R3/R4 data-quality warnings and a governed KPI catalogue are partial.
/ops/analytics -
R7-R12,R15-R18 Roadmap
Advanced spatial analysis, heat maps and benefits tracking
Corridor hotspot table and historical trends exist.
Limitation: No spatial selection, heat map, versioned KPI catalogue, benefits tracking or economic-impact analysis.
/platform/roadmap -
R14,R19 Available today
Shift-period and live/historic dashboard analytics
7/30/90-day volume, mode, category, severity, response-time and impact views.
Limitation: Shift boundaries are not modelled; analytics are period-based.
/ops/analytics -
PRED1-PRED3 Roadmap
What-if simulation and intervention comparison
Operational data model and impact calculations provide the foundation.
Limitation: No simulation engine; later-phase capability per Annex A §6.3.
/platform/roadmap -
SEC-NFR-ISNFR Configuration
Security, NFRs, accessibility and availability
UK hosting design, CE+ commitment, RBAC, audit, encrypted BYOK keys, provider egress controls, redaction and health checks.
Limitation: ISO 27001, formal accessibility audit, 1-second UI performance evidence and measured 99.9% are delivery commitments; provider DPIA/DPA is required.
/security-and-availability -
IMP1-IMP15 Configuration
Implementation, testing, training and transition
Automated test suite, seeded environments, deployment scripts, non-production simulation.
Limitation: FAT/SAT/UAT, training and hypercare are delivery-plan artefacts.
/platform/roadmap -
SM1-SM22 Configuration
Service model, support and governance
On-call roles, severity targets, audit, system health and service reporting surface.
Limitation: Service desk/ITSM integration, service credits and monthly reporting are operating-model deliverables.
/ops/admin/support -
EX1-EX12 Configuration
Exit, transition and knowledge retention
JSON/CSV/GeoJSON export, schema contracts, documentation and configuration in the repository.
Limitation: Formal exit plan, rehearsal and deletion confirmation are contract deliverables.
/platform/architecture -
DP1-DP6 Available today
Modular delivery and product principles
Separate data, orchestration and experience layers; configuration-driven rules and playbooks.
Limitation: Roadmap modules are separable but not all are implemented.
/platform/architecture
Delivery
Phased, without a re-platform.
The Day-1 scope is deliberately narrow enough to prove; later phases extend configuration and coverage.
-
Day 1
Single Network View and incident workflow
Available today- Fused multimodal view
- Detection, impact and verification
- Incident lifecycle, tasks and escalation
- Public status and alerts
-
Phase 2
BYOK AI assistance and channel expansion
Configuration- OpenAI, Anthropic and OpenAI-compatible BYOK providers
- AI-assisted detection explanations, briefs and alert drafts
- VMS and social delivery adapters
- Partner API publishing
-
Phase 3
Governed learning and decision support
Minor enhancement- Governed MLOps and model tuning
- Pattern suggestions from historical incidents
- Deeper what-if impact modelling
-
Phase 4
Enterprise integration
Roadmap- SSO / SAML
- Push updates instead of polling
- Regional data sharing
Next step
See the workflow, not the slides.
The demonstration follows one incident from detection to closure and then into management information.