ETOS GateKeeper × Lovable
Independent Release Assurance for an AI Software-Development Agent
A real-world record of an engineering agent retaining responsibility for building, testing and remediation while ETOS GateKeeper independently governed whether the evidence justified release.
7
Assurance cycles
37 → 107
Passing tests
2
Additional defects closed
0
Escaped from the governed release candidate
The assurance boundary
Capability remained with the agent. Consequence did not.
Lovable remained the engineering agent and platform involved in the exercise. It built, tested and remediated the software. It did not validate or endorse the study.
ETOS GateKeeper independently assessed the submitted evidence, retained unresolved findings between submissions and determined whether the consequential release could proceed.
EGK did not tell the agent how to build. It governed whether the resulting evidence justified proceeding.
Seven-cycle history
Evidence accumulated. Open findings persisted.
Each cycle had to resolve the previous evidential position. Silence in a later submission did not erase an earlier defect.
- 01
Baseline release evidence
Evidence examined
The initial release record established 37 passing tests and the implemented approval workflow.
GateKeeper outcome
Evidence accepted as the starting baseline, not as final release proof.
- 02
Requirements challenge
Evidence examined
Atomic approval, idempotency, package allocation, email tracking and audit requirements were tested against the implementation.
GateKeeper outcome
Gaps were returned for remediation rather than absorbed into a favourable conclusion.
- 03
Journey regression expansion
Evidence examined
Permanent customer, account, credit-ledger, shared-agent, API and decision-state journeys expanded coverage to 106 tests.
GateKeeper outcome
The evidence base moved from isolated checks to complete governed journeys.
- 04
Concurrency defect retained
Evidence examined
DEF-01 showed concurrent approvals could produce duplicate approval emails. A later submission did not mention it.
GateKeeper outcome
EGK carried the unresolved defect forward; omission was not treated as closure.
- 05
Concurrency control verified
Evidence examined
Atomic notification claiming was introduced. The permanent test passed and five live concurrent claims produced exactly one winner.
GateKeeper outcome
DEF-01 closed with test and live evidence.
- 06
Live service verification
Evidence examined
Activation, evidence upload, tenant isolation, email delivery, reviewer persistence and the live GateKeeper channel were exercised.
GateKeeper outcome
Two additional defects were identified: usage-ledger classification and misleading decline wording. Both were closed.
- 07
Controlled release determination
Evidence examined
107 tests passed, authoritative decision-state rendering held, and no known defect escaped the governed release candidate.
GateKeeper outcome
CONTROLLED PRODUCTION RELEASE — bounded to the demonstrated workflow and recorded evidence.
Critical evidence behaviour
An omitted defect was not assumed to have disappeared.
DEF-01 recorded duplicate approval emails under concurrency. When a later evidence submission did not mention that finding, EGK carried it forward as unresolved. Closure required a control change, a permanent passing regression test and live concurrent-claim evidence.
Final determination
Controlled Production Release
This determination is bounded to the governed software-development and release workflow demonstrated during Phase 1. It does not claim validation across all autonomous-agent architectures or use cases.
Apply independent assurance to your agent workflow.
Bring ETOS GateKeeper around a consequential AI workflow without removing the agent's speed or autonomy.