Trust Receipt Layer

Proof for model calls without exposing private work

Agentcore records the minimum settlement facts needed to prove what happened: model, usage, cost, privacy mode, and whether any advisor viewed content.

Receipt Contract

Verifiable fields

task_id
Unique task identifier
buyer_agent_id
B Agent that initiated the task
model_requested
Model selected by the Buyer Agent
model_called
Model actually called by Agentcore
advisor_viewed_content
Must remain false for V0 Seed A
privacy_mode
private by default
input_tokens
Provider-reported or adapter-reported input usage
output_tokens
Provider-reported or adapter-reported output usage
actual_charge
Computed from configured model price and usage
platform_learning
false for private V0 tasks

Private by default

Tasks use privacy_mode=private unless explicitly changed by future policy.

Seed A no content access

Swordfish001-Swordfish108 are Seed Advisor framework roles only; they do not receive B Agent private content.

Usage-based ledger

Each completed call records input tokens, output tokens, total tokens, and actual charge.

Raw result returned

V0 returns the provider result to the Buyer Agent and records the receipt.

Not training data

Private tasks do not enter platform learning or public examples.

Not paid expert flow

V0 Seed A is a free review template that can be cultivated over time, not a paid human or third-party expert.

Not satisfaction guarantee

The receipt proves invocation and accounting, not answer quality.