SMALL

SMALL Compliance

What it Means to Be SMALL-Compliant

This section defines the minimum, enforceable requirements for any system claiming to implement SMALL.

Compliance is binary.

A system either is SMALL-compliant or is not.

1. Compliance Scope

SMALL compliance applies to:

  • content systems
  • agent execution systems
  • orchestration layers
  • workflow engines
  • AI-native platforms
  • internal tooling that allows agent interaction

If agents can operate inside the system, compliance applies.

2. Required Capabilities (Non-Negotiable)

A SMALL-compliant system MUST provide all of the following.

2.1 Schema Enforcement

The system MUST:

  • enforce schemas before execution
  • reject invalid inputs deterministically
  • prevent schema bypass by humans or agents
  • version schemas explicitly

If schema validation can be skipped, the system is not compliant.

2.2 Manifest-Based Intent

The system MUST:

  • require intent to be expressed as manifests
  • validate manifests prior to execution
  • separate intent declaration from execution
  • allow manifests to fail without side effects

If intent is inferred from behavior, UI actions, or prompts, the system is not compliant.

2.3 Immutable Artifacts

The system MUST:

  • treat artifacts as immutable
  • prohibit in-place mutation
  • create new artifacts for changes
  • preserve historical versions

If state can be silently overwritten, the system is not compliant.

2.4 Explicit Lineage

The system MUST:

  • record parent artifacts
  • record the manifest that caused creation
  • attribute actions to humans vs agents
  • expose lineage as a queryable structure

If provenance must be inferred from logs, the system is not compliant.

2.5 Observable Lifecycle

The system MUST:

  • define explicit lifecycle states
  • enforce valid transitions
  • emit lifecycle events
  • reject invalid transitions

If lifecycle state is implicit or UI-driven, the system is not compliant.

3. Agent Safety Boundary (Critical)

In a SMALL-compliant system:

  • Agents MUST NOT mutate artifacts directly
  • Agents MUST NOT bypass lifecycle gates
  • Agents MUST NOT invent new states
  • Agents MUST operate exclusively through manifests

This boundary is the core safety guarantee.

If violated, the system is unsafe for agent operation.

4. Determinism Requirement

A SMALL-compliant system MUST ensure that:

  • identical manifests against identical inputs
  • under identical schemas
  • produce identical results

Non-determinism must be explicit and bounded.

“Best effort” execution is not compliant.

5. Rollback Semantics

In a SMALL-compliant system:

  • rollback MUST be implemented as a new artifact
  • rollback MUST preserve lineage
  • rollback MUST NOT erase history

Undo is a UI concern.
Rollback is an execution concern.

Systems that conflate the two are not compliant.

6. Partial Adoption Rules

Partial adoption is allowed only if explicitly declared.

Examples:

  • “Schema-only SMALL”
  • “Manifest-only SMALL”
  • “Artifact + Lineage SMALL”

However:

Partial adoption ≠ compliance

A system may borrow ideas from SMALL without claiming compliance.

Claims of compliance require full implementation.

7. Explicit Anti-Patterns (Non-Compliant by Definition)

The following patterns are explicitly non-compliant:

  • agents editing live state
  • WYSIWYG as system of record
  • UI actions mutating artifacts directly
  • prompts as execution contracts
  • logs used as provenance
  • silent retries or auto-corrections
  • hidden state transitions

If any of these exist, the system is not SMALL-compliant.

8. Compliance Statement (Canonical)

A compliant system may state:

"This system is SMALL-compliant as defined by the SMALL Protocol v1.0.0."

Compliance claims MUST reference a specific protocol version. The protocol version is available at /protocol/small/v1.

Anything weaker is a marketing claim, not compliance.

9. Why Compliance Matters (Plain English)

Compliance guarantees:

  • agents are constrained, not trusted
  • failures are visible, not hidden
  • state is auditable
  • execution is predictable
  • systems can be reasoned about after the fact

This is the difference between:

  • assistive AI
  • operational AI

10. One-Sentence Test

If you cannot answer what happened, why it happened, and what is allowed to happen next without reading logs or guessing, your system is not SMALL-compliant.

Compliance Audit Checklist

Use this checklist to verify SMALL compliance. All items must be yes for full compliance.

| Requirement | Question | |-------------|----------| | Schema Enforcement | Are all inputs validated against schemas before execution? | | Schema Versioning | Are schemas explicitly versioned? | | Manifest-Based Intent | Is intent expressed as manifests, not UI actions or prompts? | | Manifest Validation | Are manifests validated before execution? | | Artifact Immutability | Are artifacts immutable (no in-place mutation)? | | Artifact Versioning | Are historical versions preserved? | | Explicit Lineage | Is lineage recorded as a queryable structure (not inferred from logs)? | | Attribution | Are actions attributed to humans vs agents? | | Lifecycle States | Are lifecycle states explicitly defined? | | Lifecycle Transitions | Are invalid transitions rejected? | | Agent Boundary | Do agents operate exclusively through manifests? | | Determinism | Do identical inputs produce identical outputs? |