dbl-core 0.3.6

Deterministic Boundary Layers core substrate on KL Kernel Logic

DBL Core

DBL Core is a deterministic event substrate for the Deterministic Boundary Layer (DBL). It records intent, decisions, and executions as a single ordered stream.

Why DBL Core exists

DBL Core exists to provide a deterministic, audit-stable event substrate for systems that need to separate:

• intent from decision
• decision from execution
• normative history from observational artifacts

It is designed for systems where replayability, auditability, and governance correctness matter more than convenience or performance.

Mental Model

DBL Core maintains a single append-only event stream V:

INTENT → DECISION → (optional) EXECUTION → (optional) PROOF

Only DECISION events are normative. All other data is treated as observational and excluded from digests.

Scope

• Single-stream event model with deterministic t_index.
• Canonical serialization and digest for events and behavior logs.
• Gate decision events (ALLOW or DENY) as explicit Deltas.
• Embeds kernel traces as observational artifacts with canonical integrity digests.

Non-Goals

• No policy engine or templates.
• No execution of user tasks.
• No orchestration, UX flows, or intelligence.
• No time, randomness, or I/O side effects.

What DBL Core is not

DBL Core is intentionally minimal. It is not:

• a workflow engine
• a policy engine
• a domain framework
• an execution orchestrator
• an LLM wrapper

If you need domain semantics, validation, or verdict logic, implement a domainrunner on top of DBL Core.

Contract-first design

DBL Core behavior is defined by a stable, normative contract.

• Code must conform to the contract.
• Tests enforce contract invariants.
• Domain-specific semantics are explicitly out of scope.

See:

• DBL Core Contract
• Domainrunner Contract

Contract

• DBL Core Contract

Install

pip install dbl-core

Requires kl-kernel-logic>=0.5.0 and Python 3.11+.

Public API

• DblEvent, DblEventKind
• BehaviorV
• GateDecision
• normalize_trace is a canonicalization adapter only

Ordering

Ordering is derived from t_index (position in V). Timestamps and runtime fields are observational only.