Risk OS Red Try Stoa

Underwriting evidence

Insurers pricing AI risk send a questionnaire; the applicant fills it in by hand and self-attests. Stoa can pre-fill it from evidence instead. The underwriting-evidence export renders an AI Model Risk Assessment (modeled on the Munich RE aiSure™ template) as a self-contained, offline HTML view with a print-to-PDF button.

stoa export stoa-registry.json --underwriting \
  --underwriting-config applicant.toml \
  --out submission.html

It is also reachable from the Generate underwriting evidence button in the report.

Where the fields come from

The technical sections are populated from the scan, not typed by hand:

Questionnaire fieldSourced from
Robustness / adversarial testinginjection & output-handling findings (AI001 / AI002)
Code-quality checksthe scan running in CI with stoa diff drift gating
Agent inventory & critical findingsthe registry
Post-deployment monitoringobservability evidence (a CTRL004 gap, or its absence)
Drift mitigationcapability drift tracked via stoa diff
Insurance schedule inputselevated-exposure dimensions + economic-authority findings (DECL003 / RT002)

The applicant supplies their own numbers

Identity and real model-performance figures come from a small TOML the applicant owns. When it is present, the form shows those figures as provided by the applicant; with no config, it falls back to a clearly labeled sample so placeholder numbers are never passed off as measured data.

[identity]
company       = "Acme Payments Inc"
contact_name  = "Dana Okafor"
model_name    = "Acme Transaction-Risk Agent"
# … any identity field

[[performance]]
metric  = "Ground-truth accuracy"
value   = "98.1%"
cadence = "Monthly, held-out labeled set"
# … the applicant's own measured figures

A copy-paste template ships at examples/underwriting-config.example.toml.

The workflow

enterprise runs stoa scan
   → fills applicant.toml (identity + real metrics)
   → stoa export --underwriting --underwriting-config applicant.toml
   → submission.html — their company, their numbers, scan-sourced technical fields
   → applicant confirms and signs the PDF
   → forwards it to the underwriter

The form carries a template attribution and an applicant-to-confirm note — standard pre-filled-form hygiene, not a claim of audited data. Like every Stoa output it renders fully offline and makes no network request.