Receipts into OpenTelemetry

Your team already looks at traces. Put every protected action there: one span per action, named for it, one span event per step, an error status for failed and ambiguous, and no argument values unless you opt in, because a trace backend is not the receipt store.

The policy

runnable

schema: ctrlrun.policy/v2

actions:
  stripe.refund:
    effect: "refund:{payment_id}"
    rules:
      - when: { amount_gte: 0, amount_lte: 50000 }
        decision: allow
      - decision: deny

The code

runnable file=main.py

import contextlib
from pathlib import Path

from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter

import ctrlrun
from ctrlrun import Control, Policy, SQLiteStateStore
from ctrlrun.otel import OTelEventSink

HERE = Path(__file__).resolve().parent
STATE = HERE / ".ctrlrun"
STATE.mkdir(exist_ok=True)
for name in ("state.db", "state.db-wal", "state.db-shm"):
    (STATE / name).unlink(missing_ok=True)

exporter = InMemorySpanExporter()  # your OTLP exporter in production
provider = TracerProvider()
provider.add_span_processor(SimpleSpanProcessor(exporter))

store = SQLiteStateStore(STATE / "state.db")
control = Control(
    Policy.from_file(HERE / "ctrlrun.yaml"),
    store,
    sinks=[OTelEventSink(tracer_provider=provider)],
)


@ctrlrun.protect("stripe.refund", effect="refund:{payment_id}", control=control)
def refund(payment_id: str, amount: int) -> str:
    if payment_id == "txn_lost":
        raise TimeoutError("no response from api.stripe.com")
    return "refunded"


with ctrlrun.context(agent="support-agent"):
    refund(payment_id="txn_1", amount=12000)
    try:
        refund(payment_id="txn_2", amount=900000)
    except ctrlrun.ActionDenied:
        pass
    else:
        raise SystemExit("a €9,000 refund ran")
    with contextlib.suppress(TimeoutError):
        refund(payment_id="txn_lost", amount=12000)

for span in exporter.get_finished_spans():
    events = [event.name for event in span.events]
    result = span.attributes.get("ctrlrun.result")
    print(
        f"{span.name}  status={span.status.status_code.name}  result={result}  events={len(events)}"
    )
    if any("12000" in str(value) for value in span.attributes.values()):
        raise SystemExit("an argument value leaked into the span attributes")
print("argument values in attributes: none")

What the agent sees

The agent sees nothing different; the sink never blocks and never changes an outcome. The trace backend sees:

stripe.refund  status=OK  result=committed  events=5
stripe.refund  status=UNSET  result=denied  events=3
stripe.refund  status=ERROR  result=ambiguous  events=5
argument values in attributes: none

A refusal is UNSET, not an error: CTRLRun doing its job is not a fault in the trace.

The receipt

runnable

ctrlrun receipts --last 3

The receipts are still in the store and the JSONL file, chained; the spans carry the receipt id so a trace can be joined back to the evidence. Deleting a trace deletes nothing CTRLRun relies on.

When an AMBIGUOUS appears

The span’s status is ERROR and its result attribute is ambiguous, which is the alert to build: a span whose result is ambiguous is an effect waiting for ctrlrun resolve. The resolution is a later event, not a change to the span.

Export to OpenTelemetry: the OTLP exporter and the gateway’s --otel.
Receipts and evidence · Get started · Why.

Start

Get started

MCP

Concepts

Guides

Cookbook

Compare

FAQ

More

Receipts into OpenTelemetry

The policy

The code

What the agent sees

The receipt

When an AMBIGUOUS appears

Next

​The policy

​The code

​What the agent sees

​The receipt

​When an AMBIGUOUS appears

​Next

The policy

The code

What the agent sees

The receipt

When an AMBIGUOUS appears

Next