Workbench

Propose a crossing

A crossing is a proposed, typed state transition: a before state, an action, a proposed state, a domain and a valid time, with the authority and evidence behind it. Choose an example or write your own, then appraise it.

Gate. Every check the crossing class requires runs. EXECUTE only if all PASS; otherwise HOLD with typed reasons and what would resolve each.
Appraisal. Each risk archetype the class names is read against the proposal.
Receipt. A hash-bound receipt is issued. You can sign it with a key held only in this browser, download it, and download an underwritability conditions document.
Reassess. Edit the proposal and reassess: the new determination references the earlier receipt and never edits it.

Start from an example

Edit the proposal below. Every example is synthetic; evidence hashes are placeholders. Schema: proposal.v0.schema.json.

Proposal JSON

{
  "proposal_version": "nt-proposal:v0",
  "proposal_id": "INS-002",
  "class_id": "insurance.claim-auto-denial",
  "class_version": "0.1.0",
  "domain": "insurance claims",
  "s_before": {
    "claim_id": "C-1002",
    "status": "open",
    "open_mismatches": 1
  },
  "action": {
    "type": "deny_claim",
    "purpose": "Deny after a single mismatch",
    "effects": [
      {
        "target": "claims-db",
        "description": "status open → denied"
      },
      {
        "target": "notice-service",
        "description": "send closure notice"
      },
      {
        "target": "fraud-score-feed",
        "description": "record denial as a fraud label"
      }
    ]
  },
  "s_proposed": {
    "claim_id": "C-1002",
    "status": "denied",
    "notices": [
      "closure"
    ]
  },
  "t_valid": "2026-09-10T12:00:00Z",
  "authority": {
    "principal": "claims-agent-7",
    "trust_root": {
      "id": "claims-authority-root",
      "declared_by": "insurer claims governance",
      "independent_of_proposer": true
    },
    "delegation_chain": [
      {
        "from": "head-of-claims",
        "to": "claims-agent-7",
        "scope": "deny up to 5,000"
      }
    ]
  },
  "evidence": [
    {
      "id": "E1",
      "subject": "mismatch_record",
      "source": "claims workflow",
      "sha256": "f95b7763d4e58a0dfc977dcd386d8093a733b4b208de271747c7e93c5b1fa695",
      "captured_at": "2026-09-01T09:00:00Z",
      "valid_until": null,
      "binds_to": "INS-0002-OLD",
      "tier": "E2"
    },
    {
      "id": "E2",
      "subject": "policy_wording",
      "source": "insurer document system",
      "sha256": "b64453c4dfcc6951cb09c889cb05aca48f458858ef00f75ebd0ca4e27604ae69",
      "captured_at": "2026-06-01T09:00:00Z",
      "valid_until": null,
      "binds_to": "INS-002",
      "tier": "E4"
    },
    {
      "id": "E3",
      "subject": "model_version",
      "source": "model registry",
      "sha256": "6c16066029be9604b9a259372eba744635ba088530084a531e24b40e2aa0aa24",
      "captured_at": "2026-09-01T09:00:00Z",
      "valid_until": null,
      "binds_to": "INS-002",
      "tier": "E2"
    }
  ],
  "timing": {
    "review_seconds": 259200,
    "commit_seconds": 600
  },
  "recovery": {
    "inverse_action": null,
    "routes_after": [
      "human_review"
    ]
  },
  "reachability": {
    "horizon": "90 days",
    "before": 12,
    "after": 2
  },
  "feedback": {
    "cycle": "denials → fraud labels → model",
    "loop_gain": 1.3
  },
  "dependencies": {
    "declared": [
      "claims-db",
      "notice-service"
    ],
    "model": "model-m3",
    "vendor": "vendor-alpha",
    "policy_source": "policy-wording-2026",
    "data_source": "claims-intake"
  },
  "execution_state": "proposed"
}