LIMs (Ledger-Integrated Modules)

LIMs are the policy evaluation layer of ΔOS. They take Intents as input and produce deterministic Judgments as output.

What Are LIMs?

LIMs are modular policy evaluators that:

1. Receive Intents — When an agent submits an Intent, relevant LIMs evaluate it
2. Apply Policies — Each LIM applies its specific policy logic
3. Produce Judgments — The combined LIM output determines the Intent's Judgment

Intent → [LIM₁] → Judgment₁ ─┐
         [LIM₂] → Judgment₂ ─┼→ Combined Judgment → Final Judgment
         [LIM₃] → Judgment₃ ─┘

LIM Properties

Deterministic

Given the same inputs, a LIM always produces the same output. This enables:

• Replay for debugging
• Verification of historical decisions
• Predictable behavior

Composable

LIMs can be combined. Multiple LIMs evaluate each Intent, and their judgments are combined according to policy.

Auditable

Every LIM evaluation is recorded with its inputs and outputs.

LIM Types

Built-in LIMs

ΔOS provides standard LIMs for common governance patterns:

Custom LIMs

Define custom LIMs for domain-specific policies:

import { defineLIM } from '@deltaos/sdk';

export const highValueTransactionLIM = defineLIM({
  id: 'high-value-transaction',

  // Which Intents this LIM evaluates
  appliesTo: (intent) => intent.action === 'payment.initiate',

  // Evaluation logic
  evaluate: async (intent, evidence) => {
    const amount = intent.parameters.amount;

    if (amount > 10000) {
      return {
        judgment: 'escalate',
        reason: 'Transaction exceeds $10,000 threshold',
        escalateTo: 'finance-team'
      };
    }

    if (amount > 1000 && evidence.agent.trustScore < 0.8) {
      return {
        judgment: 'block',
        reason: 'Low trust agent attempting significant transaction'
      };
    }

    return { judgment: 'allow' };
  }
});

LIM Evaluation Flow

1. Intent Reception

When an Intent is submitted, ΔOS identifies which LIMs apply.

2. Evidence Collection

Before evaluation, ΔOS collects evidence needed by the applicable LIMs.

3. Parallel Evaluation

Applicable LIMs evaluate the Intent in parallel:

// Conceptual flow
const limResults = await Promise.all(
  applicableLIMs.map(lim => lim.evaluate(intent, evidence))
);

4. Judgment Combination

Results are combined according to combination policy:

5. Result Recording

The final Judgment and all LIM evaluations are recorded to the ledger.

Configuring LIMs

Enable/Disable

await deltaos.lims.configure('rate-limiter', {
  enabled: true,
  priority: 1
});

Set Parameters

await deltaos.lims.configure('threshold-gate', {
  parameters: {
    maxDailyTransactions: 100,
    maxTransactionValue: 50000
  }
});

Scope LIMs

Apply LIMs to specific agents, actions, or resources:

await deltaos.lims.configure('high-value-transaction', {
  scope: {
    actions: ['payment.*'],
    agents: ['agent-category:autonomous'],
    resources: ['accounts/*']
  }
});

LIM Authority Boundaries

LIM Versioning

LIMs are versioned. When policy changes:

1. New version is deployed
2. Both versions can run simultaneously
3. New Intents use new version
4. Old version remains for audit replay

const lim = await deltaos.lims.get('rate-limiter');
console.log(lim.version); // "1.2.0"
console.log(lim.previousVersions); // ["1.1.0", "1.0.0"]

Debugging LIMs

Dry Run

Test LIM evaluation without affecting production:

const result = await deltaos.lims.dryRun({
  intent: testIntent,
  lims: ['rate-limiter', 'threshold-gate']
});

console.log(result.evaluations);
// [
//   { lim: 'rate-limiter', judgment: 'allow', reason: 'Within limits' },
//   { lim: 'threshold-gate', judgment: 'block', reason: 'Exceeds threshold' }
// ]

Explain

Get detailed explanation of a LIM evaluation:

const explanation = await deltaos.lims.explain(
  'threshold-gate',
  intent
);

console.log(explanation);
// {
//   appliesTo: true,
//   evidenceUsed: ['agent.dailySpend', 'limits.maxDaily'],
//   decisionPath: [
//     'Check daily spend: $45,000',
//     'Compare to max: $50,000',
//     'Result: Within limit → Allow'
//   ]
// }

See Also

• Intent — The input to LIMs
• Judgments — The output of LIMs
• Evidence — Context for LIM evaluation
• Custom LIM Development — Build your own LIMs