This is a scripted browser simulation. Synthetic actors and fixed events illustrate a voice-agent workflow and policy decisions; they do not invoke autonomous agents, external tools, customer systems, or a live system of record. The receipt hashing is real: WebCrypto computes SHA-256 hashes in your browser for the verify, tamper, and replay exercises.
“Alpine Peak Home Services,” its customer, and every event below are fictional. A fixed script illustrates an inbound request to move an appointment, settle a deposit, and prepare a confirmation message. The calendar, CRM, payment, and messaging operations are recorded scenario events, not external tool calls. The scenario lets a reviewer inspect what a governed workflow would need to record without presenting the demonstration as customer or agent execution.
Logs live in five systems, can be edited by anyone with database access, and prove nothing about authorization. When a dispute or audit lands, you reconstruct events by hand.
In this simulation, each scripted action is evaluated against policy before the scenario advances. Allowed or not, the decision is added to a cryptographically hashed demo receipt chain.
Voice-agent pilots stall when the client asks “how do I defend this?” A receipt chain their auditors can independently verify is the answer that unlocks production.
Six labeled actors play fixed roles in a scripted handoff; they are not running agent processes. Each recorded scenario event passes through nine browser-side policy checks and receives a verdict — ACCEPTED ESCALATED or BLOCKED — before the script advances. WebCrypto computes real SHA-256 receipt hashes in your browser as the synthetic run plays.
A record you have to trust is a liability; a record you can check is an asset. Everything below operates on the receipts sealed above — recomputing real hashes in your browser.
Recompute every receipt hash from its contents and confirm each link points to the one before it. Anyone with the ledger can do this — no Summit systems required.
Try to rewrite history. Change one field on a sealed receipt — the chain breaks at that exact point, and every receipt after it becomes unprovable.
Re-run the entire decision sequence from the recorded inputs alone. Same inputs, same policy engine → the same verdicts and the same hashes, or the replay fails. This is how a dispute gets settled months later.
The scripted sequence illustrates the decision points a pilot could instrument in one bounded workflow. A real pilot would require an agreed integration and evidence boundary; nothing on this page connects to a customer deployment, recording, credential, or production system.
One workflow where actions need a defensible record — bookings, CRM writes, payments. We define the decision points and receipt requirements together.
The Proof Mode gate wraps the agent's tool calls. Policy rules are tuned to the client's decision rights, autonomy caps, and data-scope boundaries.
Every action — allowed, escalated, or blocked — seals into the chain. The client watches their own ledger build, live, on their own traffic.
Day-30 review against success gates agreed up front: verification pack, control mapping, replay demonstration. Pass the gates → production rollout.