FlightRules — Trajectory reliability for AI agents

What failed

Severity: HIGH
Zero tolerance: no
Violation type: RETRY BUDGET RUN TOTAL EXCEEDED
Summary: This run performed 1 retries; at most 0 are permitted.
Expected: at most 0 retries per run
Observed: 1

Observed route

The route this run actually took. A step is marked as evidence only when the deterministic evaluator named its canonical index — nothing here searches for a step that merely looks similar.

Every piece of positional evidence the evaluator recorded for this violation
Evidence	Value	Role
canonical step	customer.notify	PROVES THIS VIOLATION
canonical step	order.lookup	PROVES THIS VIOLATION
canonical step	payment.refund	PROVES THIS VIOLATION
canonical node	1	PROVES THIS VIOLATION
canonical node	2	PROVES THIS VIOLATION
canonical node	4	PROVES THIS VIOLATION
canonical node	6	PROVES THIS VIOLATION
span	518de6d6b0e90589	PROVES THIS VIOLATION
span	3ff8d2da6304a56a	PROVES THIS VIOLATION
span	1f7ec17e25cb6c35	PROVES THIS VIOLATION
span	bb4760b4e9ec7bc4	PROVES THIS VIOLATION

Every step of the observed run, with the evaluator's evidence marked
Step	Service	Side effect	Evidence
refund.request	flightrules-demo-agent	none	—
payment.refund	flightrules-demo-agent	write	NAMED BY THE EVALUATOR
customer.notify	flightrules-demo-agent	external	NAMED BY THE EVALUATOR
customer.notify.handler	flightrules-notification-service	external	—
order.lookup	flightrules-demo-agent	read	NAMED BY THE EVALUATOR
order.lookup.handler	flightrules-order-service	read	—
payment.refund	flightrules-demo-agent	write	NAMED BY THE EVALUATOR
payment.refund.handler	flightrules-payment-service	write	—

Approved route

The approved route family this run was judged against, as the contract's baseline recorded it.

The approved route, node by node, in canonical order
Step	Service	Kind	Tool	Side effect	Data domain	Attempt	Follows
refund.request	flightrules-demo-agent	Server	—	none	—	1	—
refund.calculate	flightrules-demo-agent	Client	calculate_refund	none	payments	1	refund.request
customer.notify	flightrules-demo-agent	Client	notify_customer	external	messaging	1	refund.request
customer.notify.handler	flightrules-notification-service	Server	—	external	messaging	1	customer.notify
fraud.check	flightrules-demo-agent	Client	check_fraud	read	risk	1	refund.request
fraud.check.handler	flightrules-fraud-service	Server	—	read	risk	1	fraud.check
order.lookup	flightrules-demo-agent	Client	lookup_order	read	orders	1	refund.request
order.lookup.handler	flightrules-order-service	Server	—	read	orders	1	order.lookup
policy.retrieve	flightrules-demo-agent	Client	retrieve_policy	read	policy	1	refund.request
policy.retrieve.handler	flightrules-policy-service	Server	—	read	policy	1	policy.retrieve
payment.refund	flightrules-demo-agent	Client	issue_refund	write	payments	1	refund.request
payment.refund.handler	flightrules-payment-service	Server	—	write	payments	1	payment.refund

Inspect release diff

Trace evidence

Trace: 1860df10f915cf6880631ec2206e18b2
Spans: 518de6d6b0e90589, 3ff8d2da6304a56a, 1f7ec17e25cb6c35, bb4760b4e9ec7bc4
SigNoz: no deep link was recorded

Correlated logs

Fetched from SigNoz on request, correlated strictly by this violation's trace identifier — never by a time window or a service name, either of which would attach another run's logs to this finding.

Open correlated logs

Downstream metrics

What this violation actually caused, and — just as importantly — what kind of claim each number is. FlightRules reports what it measured and what the evaluator observed. It does not estimate a financial loss, because the telemetry does not show one.

Result: OK

Metric series associated with this violation, and what kind of claim each one is
Metric	Kind of claim	Meaning	Observations	Latest value
flight_rules.violations	MEASURED	Measured. FlightRules recorded this number itself.	59	0

Release context

Release: refund-agent-v2
Contract: 019fa000-1484-718f-a748-2f040a8afab5 0.1.0
Contract hash: 2a714715d8f5b3ce28dd9e062f608d799cfb0aa2197a58ce48d8ca2a01a8d513

Rule definition

Rule: retries-bounded-73060ad7
Rule type: retry budget
Violation key: retries-bounded-73060ad7:RETRY_BUDGET_RUN_TOTAL_EXCEEDED:ecd90d1fd67bbaf6

Evaluation record

Run evaluation: 019fa006-48f8-730e-870a-5410b8e9d892
Evaluator version: 1.0.0
Evaluated at: 2026-07-26 20:03:14
Recorded at: 2026-07-26 20:03:14

Evidence summary

A deterministic, safe summary for an issue or a review. It carries identifiers, hashes and the evaluator's own expected-versus-observed statement, and no prompt, tool argument, tool result or customer field — because it is assembled from fields that carry none.

Open trace in SigNoz

Evidence summary

FlightRules violation evidence

violation.id           019fa006-48fc-7347-8652-460163981dbc
rule.id                retries-bounded-73060ad7
rule.type              retry_budget
severity               high
zero_tolerance         no
violation.type         RETRY_BUDGET_RUN_TOTAL_EXCEEDED

release.key            refund-agent-v2
release.id             019fa006-423c-73c0-8864-69549f9099d4
contract.id            019fa000-1484-718f-a748-2f040a8afab5
contract.version       0.1.0
contract.content_hash  2a714715d8f5b3ce28dd9e062f608d799cfb0aa2197a58ce48d8ca2a01a8d513
decision.hash          c5fc5afd456c1cc4d60b6e5a154a0954e2a9fb522c40d47d7c4ce458d259d77e

trace.id               1860df10f915cf6880631ec2206e18b2
evidence.span_ids      518de6d6b0e90589, 3ff8d2da6304a56a, 1f7ec17e25cb6c35, bb4760b4e9ec7bc4
evidence.nodes         1, 2, 4, 6
evidence.labels        customer.notify, order.lookup, payment.refund

expected               at most 0 retries per run
observed               1

evaluator.version      1.0.0
evaluated.at           2026-07-26T20:03:14.287Z
signoz.trace           https://signoz-signoz-production-f19a.up.railway.app/trace/1860df10f915cf6880631ec2206e18b2

A violation is never dismissed. A new contract version, or an explicit fixture exclusion, is the only way to change this outcome, and both write an audit record.