adp-validate

Validate ADP proposal objects and audit deliberation traces.

Install

npm install -g @ai-manifests/adp-validate

Usage

Validate a proposal

adp-validate ./proposal.json

ADP Proposal Validator
File: ./proposal.json

Schema Validation
  ✓ Valid against ADP proposal schema v0

Proposal Info
  ℹ Agent: did:adp:test-runner-v2
  ℹ Vote: approve
  ℹ Confidence: 0.86
  ℹ Tier: partially_reversible
  ℹ Action: merge_pull_request → github.com/acme/api#4471
  ℹ Domain: code.correctness

Dissent Conditions
  ℹ dc_tr_01 [active]: if any test marked critical regresses
  ℹ dc_tr_02 [active]: if coverage delta is negative

✓ All checks passed

Validate multiple proposals

adp-validate ./approve.json ./reject.json ./advisory.json

Validate a deliberation

Pass a directory of proposal files to check cross-proposal consistency:

adp-validate --deliberation ./deliberation/

Checks:

• Each proposal is individually valid (schema + semantic)
• All proposals share the same deliberation_id and action
• No duplicate agent_ids
• Reversibility tier agreement across proposals

Audit a deliberation tally

Verify that weighted voting and convergence were computed correctly:

adp-validate --audit ./deliberation/ --weights weights.json

The weights file maps agent IDs to computed weights:

{
  "did:adp:test-runner-v2": 0.71,
  "did:adp:security-scanner-v3": 0.64,
  "did:adp:style-linter-v1": 0.18
}

Output includes the full tally computation and convergence determination:

Tally Computation
  ℹ did:adp:test-runner-v2: vote=approve, weight=0.710
  ℹ did:adp:security-scanner-v3: vote=reject, weight=0.640
  ℹ did:adp:style-linter-v1: vote=approve, weight=0.180

  ℹ approve:       0.890
  ℹ reject:        0.640
  ℹ total:         1.530

  ✗ approval:      58.2% (threshold: 60.0%) ✗
  ✓ participation: 100.0% (floor: 50.0%) ✓

Result
  ✗ NOT CONVERGED — threshold not met

Semantic Checks

Beyond JSON Schema validation, adp-validate checks:

Check	Type
Dissent condition status consistency (falsified requires tested_in_round and tested_by)	Error
Dissent condition amendment consistency (amended requires non-empty amendments)	Error
Duplicate dissent condition IDs	Error
Vote revision chain integrity (each prior_vote must match previous new_vote)	Error
Revision round ordering (must be strictly increasing)	Error
Blast radius vs. reversibility tier mismatch (low rollback cost + irreversible)	Warning
Confidence of exactly 1.0 or 0.0	Warning
Reject vote with no dissent conditions	Warning
High stake without calibration accountability	Warning

Options

--deliberation <dir>  Validate all .json files in dir as a deliberation
--audit <dir>         Audit tally computation for a deliberation
--weights <file>      JSON file mapping agent_id to weight (for --audit)
--tier <tier>         Override reversibility tier for audit
--json                Output results as JSON
--help                Show help

Programmatic Use

import { validateProposal, validateDeliberation, auditTally } from '@ai-manifests/adp-validate';

const result = validateProposal(proposal);
// { valid: boolean, errors: string[], warnings: string[] }

const dlbResult = validateDeliberation([proposal1, proposal2]);
// { valid: boolean, errors: string[], warnings: string[], proposals: object[] }

const audit = auditTally(proposals, weights, 'partially_reversible');
// { tally: object, converged: boolean, thresholdMet: boolean, participationFloorMet: boolean }

Status

v0.1 — Validates against ADP spec v0.

License

Apache-2.0 — see LICENSE for the full license text and NOTICE for attribution.