Data is scattered across many sources
Devices, interfaces, addressing, VLAN/VRF, routing and topology usually live in separate formats.

NETWORK SOURCE OF TRUTH
4SO Network turns scattered network information into a machine-readable model: inventory, interfaces, L2/L3, routing, topology and history are kept with clear provenance, and inference never silently becomes fact.
4SO Network components: from source evidence to derived views with provenance
Product problem
Inventory, diagrams, configuration, routing state and operator knowledge can disagree with one another. A good single source of truth does not hide that disagreement; it keeps the provenance and confidence of each piece of data attached to the data itself.
Devices, interfaces, addressing, VLAN/VRF, routing and topology usually live in separate formats.
A relationship derived only from a diagram or naming must stay separate from an observed fact.
An inferred link candidate or site/role is not promoted to a definitive authority until it is corroborated.
Knowing where data came from, and when, is essential for change and incident analysis.
Capability overview
The goal is to turn config dumps into an operational model that supports queries, correlation and forensics.
A structured model for devices and operational metadata, independent of the source format.
Interfaces, descriptions, bundles and their relationships to other domains.
An L2 model for segments, membership and aggregation.
Addressing and routing context in a separate, queryable structure.
Static, BGP and OSPF data for path and policy analysis.
A node/edge model carrying the source and confidence of every relationship.
Network service dependencies such as DNS, time, AAA and logging in a common model.
Timestamps, sources and change history for analysis and forensics.
In this product, authority is not a simple boolean. Every record must keep its own source, timestamp and confidence so that correlation and automation can tell fact from inference.
Authority model
Instead of a flat table that treats everything as definitive truth, data is classified by its origin and extraction method.
| Domain | Example records | Authority / Confidence | Use |
|---|---|---|---|
| Inventory | Device identity, platform/role metadata, source mapping | Fact within the scope of the source snapshot | Inventory, ownership and normalization |
| Interfaces / L2 | Interface, description, VLAN, port-channel | Configuration-derived + source timestamp | Link-layer analysis and aggregation |
| L3 / Routing | IP/subnet, VRF, static routes, BGP, OSPF | Observed/config-derived record | Path, routing context and policy analysis |
| Topology | Node, edge, device matching, link candidate | Fact / Candidate / Derived with confidence | Correlation without turning inference into truth |
| Dependencies | DNS, time, AAA, logging relationships | Source-owned relation or bounded inference | Service dependency analysis |
| History | Snapshot identity, source history, change context | Timestamp + provenance preserved | Forensics and change comparison |
Architecture
Sources are ingested separately, parsers normalize them, and the single source of truth keeps records with their provenance and confidence. Analytical views are only projections.
Inference Guardrails. 4SO Network keeps an explicit distinction between what a source actually stated, what has been correlated across several sources, and what has only been inferred.
| Evidence type | Classification | Usage rule |
|---|---|---|
| Configuration snapshot | Fact within the scope of that snapshot | Interface, addressing, VLAN/VRF and routing facts are valid only with their own source and timestamp. |
| Topology/diagram source | Source-scoped fact | A relationship shown in a diagram does not prove the live state of the network. |
| Naming / Description correlation | Candidate | Link/role/site inference is not promoted to definitive authority before corroboration. |
| Multi-source corroboration | Confidence-bearing relation | Supporting evidence and sources stay attached to the relation; confidence is never separated from provenance. |
| Derived view | Projection | Summaries, path views or analytics are built from source-owned data and do not create a second SoT. |
| Historical comparison | Change context | The difference between two snapshots shows change; it does not claim the cause of an incident without additional evidence. |
Lifecycle
Until a snapshot has been parsed, classified and correlated, it is only raw input; a derived result does not take the place of a fact until it has been reviewed.
Data profiles
Separating domains keeps analysis from depending on one large, ambiguous CSV.
Device identity and metadata.
Interface and aggregation detail.
Segment and routing context.
Routing data for path analysis.
Nodes and edges with explicit authority.
Change tracking and confidence.
Data and evidence flow
Rather than treating everything as truth, the pipeline keeps source-owned facts, correlation and inference in separate stages.
Every relation is stored with its own evidence type.
Correlation does not remove source ownership.
Confidence and provenance remain throughout the review loop.
Source identity is preserved before parsing.
Different formats are converted into a canonical model.
Relations are built without removing provenance.
Inference is not fact until independent evidence exists.
A change view is a projection, not a new authority.
Path analysis and conflict resolution are carried out without turning a projection into an authority.
A probable path is computed from structured evidence, and the source of every hop is preserved.
Source conflicts are kept explicit before any promotion to fact.
| Action / Flow | Admission / Preconditions | Execution / Lock | Success Criterion | Failure / Recovery |
|---|---|---|---|---|
| Ingest Snapshot | Source identity/hash/timestamp must be known | The parser materializes only that source's data | Records are traceable back to their snapshot/source | A parse failure does not overwrite the original source and does not record partial success |
| Normalize | Schema/domain parser for Device/Interface/L2/L3/Routing | Different formats are converted into the canonical domain model | Identity and source mapping are preserved in the record | Unknown fields are not fabricated; missing data stays missing |
| Correlate Sources | Match keys and provenance available on both sides | Correlation builds the relation but does not remove source ownership | Every relation can show its supporting evidence/sources | Conflicts between sources are not hidden and are not automatically turned into a single truth |
| Infer Candidate | Naming/description/topology hints used only for bounded inference | The candidate is recorded with its confidence and inference method | Candidates are distinguishable from facts and reviewable | Inference does not enter definitive authority without corroboration |
| Review / Corroborate | Independent evidence or operator confirmation required | The candidate is evaluated against additional evidence | Classification/confidence are preserved with a trace | No evidence = unresolved; no fabricated rejection/confirmation |
| Compare / Export Safe View | Specific snapshots and a sensitivity policy selected | Derived view/history/report is built from existing records | Reproducible projection without creating a second SoT | Public export removes real IP/ASN/device/site/topology data; root cause is not claimed without evidence |
Core rule: Projection ≠ Authority. Summaries, path views, analytics or forensics may be built from several sources, but the original fact stays bound to its own source and timestamp.
Operations catalog
4SO Network is built for decisions on network data, with Provenance and Confidence as part of the model itself.