Underwriting

The most restrictive finding sets the outcome

Each piece of evidence produces a finding with an outcome and a risk class. Outcomes are ordered Approve < ReduceCeiling < StepUp < Deny and risk classes Baseline < Guarded < Elevated < Critical. The report takes the maximum of each, so a single deny finding wins, and the signed record maps the outcome to a review state and a budget action:

• Approve keeps the requested ceiling at full strength: Approved, Preserve.
• ReduceCeiling multiplies the ceiling by reduce_ceiling_factor: Approved, Reduce.
• StepUp holds the request for manual review or stronger evidence: ManualReviewRequired, Hold.
• Deny grants no ceiling: Denied, Deny.

Every UnderwritingDecisionFinding keeps its own reason, an UnderwritingDecisionReasonCode, and an optional signal_reason naming the input signal that raised it, so a consumer can trace the outcome back to its evidence. The taxonomy is versioned: each signed record embeds chio.underwriting.taxonomy.v1.

{
  "subjectKey": "did:chio:z6MkbY4…e1R",
  "effectiveScore": 0.42,
  "probationary": false,
  "resolvedTier": "tier-2",
  "importedSignalCount": 0,
  "acceptedImportedSignalCount": 0
}

A probationary agent keeps half its ceiling

Go back to the buyer's first week: two governed quotes, a Verified runtime assurance tier, and a reputation score of 0.4, still inside its probationary window. These are selected fields of its policy input; the complete input also carries the taxonomy, receipt counters, and evidence references.

{
  "schema": "chio.underwriting.policy-input.v1",
  "generatedAt": 1700000000,
  "filters": { "agentSubject": "did:chio:z6MkbY4…e1R", "receiptLimit": 100 },
  "receipts": { "matchingReceipts": 2 },
  "reputation": {
    "subjectKey": "did:chio:z6MkbY4…e1R",
    "effectiveScore": 0.4,
    "probationary": true
  },
  "runtimeAssurance": { "highestTier": "verified" },
  "signals": [{
    "class": "guarded",
    "reason": "probationary_history",
    "description": "subject is still probationary"
  }]
}

The evaluator produces two findings. The reputation finding routes 0.4 to ReduceCeiling at Elevated, with reason ReputationBelowApproveThreshold and signal reason LowReputation. The ProbationaryHistory signal routes to ReduceCeiling at Guarded, with reason PolicySignal.

Rolled up, the outcome is ReduceCeiling at Elevated, with suggested_ceiling_factor 0.5 from the default policy. The signed record carries review_state Approved, budget Reduce with factor 0.5, and a premium quoted at 400 basis points. Approved records the review; Reduce records that the buyer keeps half the requested ceiling. MonetaryAmount.units are integer minor units, so a $10.00 exposure is units: 1_000 and the premium is ceil(1000 * 400 / 10_000) = 40: $0.40 in the same currency.

The default policy routes reputation into three bands

UnderwritingDecisionPolicy::default() requires at least one matching receipt, the newest at most 30 days old (maximum_receipt_age_seconds). Runtime assurance must reach Attested to avoid a step-up and Verified to approve, an active tool certification is required, a compliance score reference is optional, and reduce_ceiling_factor is 0.5. Reputation routes by two thresholds:

• score >= 0.6 approves, subject to the other findings.
• 0.25 <= score < 0.6 raises a ReputationBelowApproveThreshold finding and reduces the ceiling, or steps up when other elevated signals join it.
• score < 0.25 raises a ReputationBelowDenyThreshold finding and denies.

The policy is signed alongside the decision report, so the thresholds in force at decision time are recoverable from the record. policy.validate() rejects a deny floor at or above the approve threshold, which keeps the reduce band wider than zero.

These thresholds belong to underwriting alone. The credit scorecard has its own five bands, from Prime to Restricted. A credit facility record carries an underwriting outcome, review state, and risk class as fields beside a scorecard, and no function in either crate derives a band from a threshold or a threshold from a band.

The premium follows the outcome and the risk class

Each signed UnderwritingDecisionArtifact carries an UnderwritingPremiumQuote in one of three states: Quoted, Withheld, or NotApplicable. Quoted basis points (100 bps is 1%) come from a fixed schedule keyed by outcome and rolled-up risk class.

The quoted amount is ceil(exposure_units * bps / 10_000) in the currency of the quoted exposure; with no exposure supplied, the quote carries basis points only. A StepUp quote stays withheld until manual review or stronger evidence completes. Had the step-up trace above landed at ReduceCeiling at Elevated, the schedule would price it at 400 bps through this helper:

fn quote_premium_amount(
    exposure: &MonetaryAmount,
    basis_points: u32,
) -> Result<MonetaryAmount, String> {
    let numerator = u128::from(exposure.units)
        .checked_mul(u128::from(basis_points))
        .ok_or_else(|| "premium amount multiplication overflowed".to_string())?;
    let units = numerator.div_ceil(10_000_u128);
    let units = u64::try_from(units)
        .map_err(|_| "premium amount exceeds the supported minor-unit range".to_string())?;
    Ok(MonetaryAmount {
        units,
        currency: exposure.currency.clone(),
    })
}

That quote would carry basis_points = 400 and quoted_amount = MonetaryAmount { units: 40, currency: "USD" }. The helper returns a Result: a multiplication that overflows u128, or a product past u64 minor units, is an error, never a truncated or wrapped premium, and the test quote_premium_amount_rejects_when_basis_points_force_overflow pins it. The separate insurance-flow function, price_premium, prices coverage that Liability Market binds.

Evaluate reads the evidence; issue signs the decision

Run the evaluation with chio trust underwriting-decision evaluate. It reads receipts from a local database, or from a trust control plane with --control-url and --control-token.

$ chio trust underwriting-decision evaluate \
    --agent-subject demo-agent \
    --tool-server ts-demo \
    --receipt-db ./receipts.db

schema:                 chio.underwriting.decision-report.v1
generated_at:           1788540481
outcome:                Deny
risk_class:             Critical
policy_version:         chio.underwriting.decision-policy.default.v1
matching_receipts:      0
findings:               4
- StepUp InsufficientReceiptHistory: only 0 receipt(s) matched; policy requires at least 1
- Deny ReputationBelowDenyThreshold: effective reputation score 0.0000 is below the deny threshold 0.2500
- ReduceCeiling PolicySignal: local reputation is still probationary for the requested window
- StepUp PolicySignal: no active certification evidence is available for tool server `ts-demo`

You should see the default policy fail closed on an empty history. Zero matching receipts trip the history minimum, a reputation score of zero falls below the deny threshold, the subject is still probationary, and no certification resolves for the tool server: four findings, and the strongest sets the outcome.

evaluate is read-only. It returns the unsigned UnderwritingDecisionReport (schema, outcome, risk class, policy version, suggested ceiling factor, matching receipt count, and one line per finding) and persists nothing; decision_id, review_state, budget, and premium exist only on the signed record. chio trust underwriting-decision issue mints that UnderwritingDecisionArtifact:

$ chio trust underwriting-decision issue \
    --agent-subject demo-agent \
    --tool-server ts-demo \
    --receipt-db ./receipts.db \
    --authority-seed-file ./authority.seed

schema:                 chio.underwriting.decision.v1
decision_id:            uwd-194242790e6dd4670371d2e60cae09d0585615993e4636a6c92cee20b21977f8
issued_at:              1788540482
signer_key:             a304e5344fc024ba2244dc108c104558107c34f776cbfdef8d256ad414f6c6b1
outcome:                Deny
review_state:           Denied
budget_action:          Deny
premium_state:          NotApplicable

The signer is the node's behavioral-feed signing keypair, loaded from --authority-seed-file or --authority-db before the report is built. The same public key anchors the trusted-key set that reputation scoring runs against, and that scoring fails closed on an empty set, so without a seed or a database the command refuses to sign with an ephemeral key.

Both verbs share the evidence filters --capability, --tool-server, --tool-name, --since, --until, and --receipt-limit. A third verb, simulate, replays an alternative policy file against the same evidence without persisting a decision.

A decision can be challenged with an UnderwritingAppealRecord, whose status is Open, Accepted, or Rejected. An accepted appeal can name a replacement_decision_id; creating that replacement moves the original decision's lifecycle state from Active to Superseded. An operator opens an appeal with chio trust underwriting-appeal create --decision-id --requested-by --reason and closes it with chio trust underwriting-appeal resolve --appeal-id --resolution accepted|rejected --resolved-by (chio trust).

Next: put a decision behind a credit line