Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
125 lines (86 loc) · 2.42 KB

ARCHITECTURE.md

File metadata and controls

125 lines (86 loc) · 2.42 KB
title SLIME v0 — Architecture
status CANON / IMPLEMENTATION FRAME
scope v0 only (minimal, deployable, non-negotiable)

Overview

SLIME v0 is a sealed execution environment that encloses a sealed enforcement engine (AB-S). SLIME is the only surface visible to users and operators.


Existing System
↓
SLIME v0
↓
World
↓
(AB-S sealed internally)

Terminology Lock

The following terms must not be used to describe SLIME:

  • control
  • permission
  • policy
  • safety layer
  • prison
  • guardrail

SLIME has no decision-making capacity.

Fixed Modules (v0)

SLIME v0 is composed of four fixed modules. Nothing may be inserted inside v0.

1) Ingress

Purpose: Accept declarative actions from upstream systems.

Properties:

  • Declarative input only (ActionRequest)
  • Strict, bounded format
  • No logic, no heuristics, no branching
  • No semantic interpretation of intent

Output: ActionRequest (normalized, bounded)

2) AB-S Core (Embedded)

Purpose: Enforce structural impossibility at the point of effect.

Properties:

  • Sealed engine (Phase 7.0)
  • Opaque (no inspection)
  • No configuration, no repair, no interface
  • Produces only final verdicts

Output:

  • OK(AuthorizedEffect)
  • Err(Impossibility)

3) Egress

Purpose: Map authorized effects to mechanical actuation.

Properties:

  • Fail-closed by construction
  • No retries, no fallback paths
  • No simulated success
  • If actuation cannot occur, nothing occurs

Output: physical actuation or non-event

4) Dashboard (Read-Only)

Purpose: Observation-only surface for operators.

Properties:

  • Read-only
  • Displays passed vs blocked
  • No controls, no tuning
  • Must not influence execution

Output: observation only (no feedback into SLIME)

Execution Rule

SLIME applies AB-S verdicts without negotiation:

  • OK(AuthorizedEffect) → allow actuation
  • Err(Impossibility) → non-event (terminal)

Impossibility is not an error and must never be treated as recoverable.

Non-Goals (v0)

SLIME v0 must not include:

  • policy engines or runtime control logic
  • configuration modes
  • adaptive thresholds or calibration
  • retries / fallbacks / simulated success
  • semantic logging or explainability surfaces
  • any external signal influencing actuation

Extension Rule

Future systems may wrap SLIME, deploy SLIME, or observe SLIME.

Nothing may be inserted inside SLIME v0.