Documentation
Platform Principles
These are the cross-cutting rules that make rFabric one connected robotics platform rather than a set of adjacent products. They shape every workflow, API surface, and operational decision in the platform.
1. One entity graph across the lifecycle
Every major object in the platform exists in the Unified Data Model with explicit relationships:
- raw sessions and processed episodes
- annotations and curation rules
- dataset snapshots and training runs
- models, artifacts, and deployments
- updates, telemetry, and incidents
- maintenance cases and intervention sessions
None of these live in disconnected side systems. They connect in one queryable graph.
2. Every action is traceable
Every significant lifecycle action answers the same questions:
- who or what triggered it
- on which entity or scope
- through which interface or workflow
- under which policy or approval path
- with what resulting state
Traceability is what makes reproducibility, rollback, incident analysis, and audit-ready operations credible.
3. Automation with human gates
The platform automates aggressively, but robotics operations require structured human decision points:
- data review before finalization
- model promotion approval
- rollout sign-off
- maintenance escalation
- teleoperation and supervised autonomy
Human involvement is encoded into the platform itself, not handled outside it through ad hoc coordination.
4. Incremental adoption is first-class
Teams adopt the platform in stages:
- start with ingestion and replay while keeping the current training stack
- adopt curation and dataset finalization next
- centralize release and rollout later
- bring fleet operations into the same control plane over time
The Platform API, integrations, and canonical entity model exist so that partial adoption preserves continuity instead of creating new silos.
5. Coexistence over forced replacement
Robotics teams already use specialist tools, and the platform is built to work alongside them:
- keep visualization tools like Foxglove or Rerun where they are useful
- keep existing training code where migration is not yet worthwhile
- keep customer-specific operational systems where needed
rFabric standardizes lineage, workflows, approvals, and handoffs across those tools rather than demanding they be replaced.
6. Operations data feeds development
Field evidence is one of the highest-value data sources in robotics:
- intervention spikes reveal autonomy gaps
- rollout regressions reveal missing evaluation coverage
- maintenance cases reveal hardware or configuration sensitivity
- site-specific anomalies reveal distribution mismatch
That evidence flows back into review, annotation, curation, evaluation, and training instead of remaining trapped in monitoring dashboards.
7. Regional compliance is architectural
Residency, transfer restrictions, retention policy, support-access boundaries, and regulated AI evidence cannot be bolted on afterwards. These controls attach directly to entities, workflows, APIs, rollout policy, and operations access.
8. Robustness is a product feature
Dependable robotics infrastructure is defined by more than advanced ML workflows:
- explicit lifecycle contracts
- reliable provenance
- rollout safety
- operational recoverability
- auditability
- clear separation of concerns
The platform is built to feel dependable before it feels clever.
End-To-End Lineage
Put together, these principles produce end-to-end lineage across the full lifecycle:
robot → raw session → processed episode → annotation and quality score → dataset snapshot → training run → checkpoint → candidate model → approved model → artifact → deployment → telemetry / intervention / maintenance
This chain is what lets teams answer questions like:
- Which source data produced the model now running in production?
- Which evaluation evidence approved that rollout?
- Which field regressions appeared after deployment?
- Which incidents should now become curation and evaluation inputs?
That is the practical meaning of a full-lifecycle platform.