TENSOR Framework

TENSOR Documentation for Implementers

Start Here: Implement Core v1.0

This page is the implementation contract. Start with required fields and deterministic execution semantics, then layer optional metadata after you pass conformance.

  1. Pin a schema version and graph version from the release channel.
  2. Validate graph packages against the pinned schema before execution.
  3. Traverse branches using explicit yes, no, and unknown decisions.
  4. Record traversal output with stable node and edge IDs for replay.

Required Package Contract

A compliant graph package includes required metadata (namespace, product, version, schemaVersion) and required collections (nodes, edges).

Required: top-level metadata keys, stable IDs, valid decisions, and schema-valid structure.
Optional: localized labels, operator hints, and namespaced extension metadata.

Node Contract

Each node includes required id, text, and category fields. Optional fields include label, translations, and extensions.

Required: unique ID matching Q<number>, non-empty question text, and valid category.
Optional: display labels and localized copies that do not alter semantics.

Edge Contract

Each edge includes required source, target, and decision. Edge IDs must encode the same relationship (example: Q12-yes-Q1).

Required: source/target node IDs that exist, decision in allowed enum, and encoded edge ID consistency.
Optional: extension data for confidence, control mapping, or provenance.

Execution Semantics

TENSOR supports loops and reconvergence so investigations can revisit shared questions when new evidence arrives. Deterministic branch semantics are mandatory even when model output is probabilistic.

  • Branching: every transition is explicit, never implied.
  • Re-entry: cycles are valid for iterative investigations.
  • Unknown handling: uncertainty remains first-class through unknown edges.
  • Traceability: traversals are reconstructable from node and edge IDs.

Compatibility and Breaking Changes

Minor versions can add categories, nodes, and edges without invalidating existing consumers. Major versions may introduce breaking changes and must ship migration guidance and deprecation windows.

Required compatibility behavior for implementers:

  • Pin schemaVersion in CI and block releases on validation drift.
  • Do not repurpose existing IDs to new semantics.
  • Reject artifacts that violate encoded edge identity rules.
  • Document supported version range in your integration README.

Implementation References

Use Implement Core for the 30-minute setup path, Conformance Suite for pass criteria, and Standards for compatibility policy details.