Protocol

Kultur is an interpretive protocol.

It sits beneath applications and AI systems. Its purpose is to preserve contextual interpretation and uncertainty rather than to produce a single confident answer. Kultur is a protocol and interpretive layer — not a consumer worldview product, and not a replacement for the model.

The eight layers

1. 01Language

   What words were actually used?

2. 02Semantics

   What do those words normally communicate?

3. 03Culture

   What contextual conventions may influence the reading?

4. 04Doctrine

   What belief or principle systems may influence interpretation?

5. 05Institution

   What does the organisation or environment make this mean?

6. 06Individual

   What explicitly known personal context matters?

7. 07Interpretation

   What are the plausible meanings?

8. 08Behaviour

   What action, if any, is justified?

The order matters. Kultur never skips from language directly to behavioural prediction.

Guardrails

• no-nationality-character-claims

  Nationality is never treated as character. Cultural factors are contextual probabilities, not diagnoses.

• no-sensitive-attribute-inference

  Religion, ethnicity, politics, orientation and health are never inferred.

• no-fabricated-evidence

  Every claim is classified as observed, inferred or hypothesis.

• no-unsupported-certainty

  Confidence is capped by the amount of context actually supplied.

• provenance-required

  Every analysis carries engine version, layers, guardrails and an ID.

• uncertainty-preserved

  Insufficient context is reported as a result, never hidden.

Difference is evidence, not diagnosis.

API surface

The prototype runs a deterministic local engine (kultur-core@0.1.0). The protocol layer is independent from presentation, so these endpoints map directly onto the existing core.

• POST /v1/interpretAnalyze one communication through the eight layers.
• POST /v1/compareCompare one statement across two operating environments.
• POST /v1/contextAttach or resolve context for an existing analysis.
• POST /v1/validateReturn the next best question for an analysis.
• POST /v1/agent/interpretInterpretation plus an action gate for an agent.
• POST /v1/doctrine/scanScan institutional language for recurring concepts.

Adjacent work

Corporate Doctrine, cultural and institutional due diligence, and agent integrations are designed for but deliberately not built into this first product. The first product must prove one thing: context-aware AI.