Receipts & verdicts
Claimed vs actual. Grounded vs silent.
A receipt is proof of what the wrapped function claimed versus what it actually returned. It is not an LLM-as-judge score.
Shape
Score-shaped check rows (name, value, data_type, source, comment) plus passed for the console. Tool calls use OpenInference TOOL attributes.
claimed → input / claims label
actual → return value (truncated) or exception
verdict → grounded | ungrounded | silent_failure | error
checks[] → Score-shaped facts
grounding_score + estimated: true
hash, prev_hash, timestamp
trace_id, span_id, run_id
tool_calls[]Silent failure
The interesting case: HTTP 200, model said "done", tool span never fired. @check(claims="github.create_issue") plus OpenInference tool attributes is how that becomes silent_failure instead of a green trace.
Empty output is a fact, not a vibe. Heuristics that are not hard facts stay estimated: true.
Where it lives
- On the Python return value:
_agenko_receipt - In the console receipts view (control plane)
- On the OTEL span as
agenko.*attributes
What we will not do
We will not call an LLM to decide if the answer "looks grounded" and then present that as the receipt. That is detector-only territory. Grounding score is labeled estimated until a check is a hard fact.
Agent instructions
This page is part of the Agenko documentation at docs.agenko.app. Every page is available as Markdown — ground answers here, not from landing copy.
Querying these docs
- Full index: docs.agenko.app/llms.txt · sections: llms-docs.txt, llms-sdk.txt, llms-integrations.txt
- This page as markdown: docs.agenko.app/receipts.md
- Search API:
curl -sG "https://docs.agenko.app/api/search-docs" --data-urlencode "query=…"
Before writing SDK code
- Install the agent skill or read Agent skill
- Read Best practices — what a good receipt looks like
- Trust boundary:
@checkafter,@approvebefore init()— zero required args, never throws- Mark
estimated: trueuntil a check is a hard fact - Shipped vs backlog: Introduction status table