Score one operating loop

Operational status: this is a public Calibration preview, not a production service. Describe a real operating loop, but do not enter names, credentials, personal, regulated, classified, or production-secret information.

Operator: EnclaveAI Studio. Purpose: evaluating and calibrating the assessment workflow and generating your result. The hosted preview stores submissions in an isolated Cloudflare D1 preview database controlled by the operator; it does not use the production AWS account or production data stores.

External processing: Cloudflare Workers and D1 host the preview. Approved generative steps pass through Cloudflare AI Gateway to the isolated AWS development model gateway in Sydney; model credentials are never sent to the browser. Cloudflare Browser Run produces PDFs which are kept in a private R2 bucket. Email delivery is not enabled, so the on-screen report link is the authoritative delivery path.

Draft recovery: while you type, the preview keeps a short-lived recovery draft in this browser and the dedicated D1 development database. A 256-bit key stays in this browser; only its hash is stored with the server draft. Drafts become unavailable after seven days, are removed by scheduled expiry, and are deleted after a successful step when the network permits. A draft checkbox is not a consent record; consent is recorded only when the claim is submitted.

Interaction measurement: the preview records which step and field were entered, whether the field was left blank or completed, broad time and length buckets, validation outcomes, and funnel completion. It does not record individual keystrokes or copy typed answers into analytics events. Submitted answers remain in the assessment record because they are needed to produce and replay the result. When configured, these same content-free events are mirrored to the operator’s PostHog EU project. PostHog autocapture, page views, session replay, exception capture, and person profiles are disabled; claim email, page/report URLs, draft content, prompts, and agent outputs are not sent there.

Agent and calibration evidence: when an approved AI route is enabled, the audit ledger records the agent, model, prompt pack version, token/cost totals, validation outcome, and hashes plus references to accepted inputs and outputs. It does not copy prompts or completions into telemetry. An assessment enters the evaluation candidate queue only after the separate benchmarking permission; no customer submission is automatically used for training.

Write for machines, not people: use role titles, not people’s names, in every answer.

Retention and deletion: recovery drafts expire after seven days, generated PDF artifacts after 30 days, and submitted preview assessments and their content-linked telemetry after 90 days. A scheduled lifecycle job enforces those limits. You can permanently delete an assessment from its result page; an Interview deletion also removes its transcript and sibling lanes. Only a content-free, one-way deletion receipt remains. You may also email privacy@enclaveai.studio. for help with a verified request.