ADVFAB WIKI · FIELD EVIDENCE

A fabric you can
reason about.

Concepts, operational semantics, and a public projection of the evidence inspected in the Advanced Fabric Headlamp plugin.

01 / MODEL

Network reality becomes evidence before it becomes a score.

Advanced Fabric does not score the network directly. Collectors create versioned measurements; those measurements become scoped evidence; only compatible evidence contributes to an assessment.

RealityRoutes · peers · probes
→
MeasurementsVersioned definitions
→
EvidenceScoped and time-bound
→
O / S / IOptimality · stability · impact
O

Relative optimality

Compare the current path only with feasible alternatives measured in the same scope and window.

S

Temporal invariance

Track variance, loss bursts, BGP churn, and route churn across an explicit window.

I

Intervention impact

Keep observed outcomes separate from assumptions about a proposed or completed change.

02 / FIELD SNAPSHOT

What Headlamp can see.

This static public projection contains counts and hashed identifiers only. It is not a control surface and cannot query the cluster.

READ ONLY

Loading sanitized evidence…

Path assessment states

Observation episodes

03 / PATH ASSESSMENTS

Evidence, without topology disclosure.

Each row corresponds to a Headlamp-visible assessment. Source names are replaced before publication by stable, domain-separated hashes.

Private IDStateNodePathsEvidenceObserved

04 / PRIVACY CONTRACT

Public by derivation, not by redaction.

Published

  • Aggregate resource and state counts
  • Readiness booleans and evidence counts
  • Observation timestamps
  • 12-character SHA-256 fingerprints

Never published

  • Node, peer, namespace, or resource names
  • IP addresses, prefixes, interfaces, or AS paths
  • UIDs, resource versions, labels, or annotations
  • Secrets, credentials, logs, or Kubernetes object bodies

Hash construction: SHA-256("advfab.org/public-v1\0" + identifier)[0:12]. Domain separation prevents these fingerprints from matching hashes made for another publication. Short hashes are pseudonyms, not anonymity; low-entropy identifiers are therefore omitted rather than hashed.