The run stayed green. The outcome crossed the line.
Outcome testing asks whether the buyer's intended business condition remained true—not only whether the automation executed. This lab shows five fictional cases where a healthy technical run and a wrong business result coexist.
At the invariant checkpoint, execution status and business truth become two different questions.
01 / EVIDENCE BOUNDARY
Useful without pretending to be proof
Five failure shapes. Zero claimed incidents.
Synthetic Lab — five preloaded fictional cases. Every record, identifier, response, and failure below is invented. The cases are not customer incidents, benchmarks, or evidence of how often any failure occurs.
RUN: GREEN has a narrow meaning. In each fictional trace, it means only that every configured technical step reported success and no unhandled error was recorded. It does not mean the intended business outcome was correct.
Evidence boundary. These illustrations do not show that OutcomeTest would detect these defects—or every defect—in a real workflow.
Mint: technical execution status
Orange: business-result divergence
Blue: control and adverse-test evidence
Current status. OutcomeTest remains in problem discovery and is not available for purchase. No payment or workflow files are accepted.
02 / SYNTHETIC CASEBOOK
Agency handoff lens
Follow the split, then test the invariant.
Use the incident rail to jump between cases. Every case remains fully readable in document order when JavaScript, animation, or styling is unavailable.
01 / 05OT-LAB-001
SYNTHETIC · FICTIONAL · ILLUSTRATIVE
CASE TITLE
Green run, wrong record
FAILURE CLASS wrong entity or mapping
In this fictional record-sync scenario, an operations team intends to apply a synthetic service-level change to one internal account record.
TECHNICAL STATUSRUN: GREEN — every configured technical step reported success; no unhandled error was recorded.
BUSINESS OUTCOMEWRONG
BUYER-OWNED INVARIANT
Only the destination whose immutable entity key equals ACCT-SYN-017 may change. ACCT-SYN-071 must remain unchanged.
Synthetic input
event_idEVT-SYN-001
source_entity_idACCT-SYN-017
requested_levelstandard
Expected result
ACCT-SYN-017 is set to standard; ACCT-SYN-071 is unchanged; the receipt identifies ACCT-SYN-017 as the only written destination.
Actual wrong result
ACCT-SYN-071 is set to standard; ACCT-SYN-017 is unchanged; the technical run still reports success.
MISSING OR FAILED CONTROL
Entity-match control: require immutable-key equality before the write and reconcile the written destination key against the accepted input after the write.
AGENCY-RUN ADVERSE TEST
In an isolated agency-controlled test environment, seed ACCT-SYN-017 and ACCT-SYN-071 with deliberately similar display labels, reverse their list order, submit the same synthetic input, and assert that only the exact-key destination changes.
Evidence boundary. Synthetic illustration only; not a customer incident. It proves neither how often this failure occurs nor that OutcomeTest would detect this or every defect in a real workflow.
02 / 05OT-LAB-002
SYNTHETIC · FICTIONAL · ILLUSTRATIVE
CASE TITLE
Fresh state overwritten by a late event
FAILURE CLASS stale or late result
In this fictional status-sync scenario, two synthetic versions for the same work item arrive out of order.
TECHNICAL STATUSRUN: GREEN — every configured technical step reported success; no unhandled error was recorded.
BUSINESS OUTCOMEWRONG
BUYER-OWNED INVARIANT
The destination may retain only the greatest accepted source version for ITEM-SYN-044. An older event must not overwrite a newer accepted state.
Version 12 and status ready remain at the destination; version 11 is rejected or quarantined as stale.
Actual wrong result
The late version-11 event changes the destination back to pending; the technical run for that event still reports success.
MISSING OR FAILED CONTROL
Freshness-and-order control: compare the incoming version with the accepted version watermark before mutation, then reconcile the destination version after the write.
AGENCY-RUN ADVERSE TEST
In an isolated agency-controlled test environment, deliver version 12 first and version 11 second, then assert that the late event causes no destination mutation and is recorded as stale.
Evidence boundary. Synthetic illustration only; not a customer incident. It proves neither how often this failure occurs nor that OutcomeTest would detect this or every defect in a real workflow.
03 / 05OT-LAB-003
SYNTHETIC · FICTIONAL · ILLUSTRATIVE
CASE TITLE
Complete flag, missing required item
FAILURE CLASS incomplete or partial result
In this fictional handoff scenario, a synthetic package contains three required checklist items.
TECHNICAL STATUSRUN: GREEN — every configured technical step reported success; no unhandled error was recorded.
BUSINESS OUTCOMEWRONG
BUYER-OWNED INVARIANT
PACKAGE-SYN-008 may be marked complete only when CHK-SYN-A, CHK-SYN-B, and CHK-SYN-C all exist at the destination exactly once.
All three required item IDs exist exactly once and only then is PACKAGE-SYN-008 marked complete.
Actual wrong result
CHK-SYN-A and CHK-SYN-B exist, CHK-SYN-C is absent, and PACKAGE-SYN-008 is marked complete; the technical run still reports success.
MISSING OR FAILED CONTROL
Completeness control: compare the required ID set and count with the written destination set before allowing the completion marker, then reconcile the final set.
AGENCY-RUN ADVERSE TEST
In an isolated agency-controlled test environment, make one required synthetic item produce an empty downstream result and assert that the package cannot be marked complete and that the missing ID is named.
Evidence boundary. Synthetic illustration only; not a customer incident. It proves neither how often this failure occurs nor that OutcomeTest would detect this or every defect in a real workflow.
04 / 05OT-LAB-004
SYNTHETIC · FICTIONAL · ILLUSTRATIVE
CASE TITLE
One request, two side effects
FAILURE CLASS duplicate side effect
In this fictional task-creation scenario, the same synthetic request is retried after its first acknowledgement is delayed.
TECHNICAL STATUSRUN: GREEN — every configured technical step reported success; no unhandled error was recorded.
BUSINESS OUTCOMEWRONG
BUYER-OWNED INVARIANT
REQUEST-SYN-404 may create exactly one durable task for WORK-SYN-009, regardless of retry or acknowledgement timing.
Synthetic input
request_idREQUEST-SYN-404
work_item_idWORK-SYN-009
requested_effectcreate one task
deliveryoriginal request plus one identical retry
Expected result
Exactly one durable task exists; the retry creates no second task and resolves to the original task reference.
Actual wrong result
TASK-SYN-901 and TASK-SYN-902 both exist for the same request and work item; both technical attempts report success.
MISSING OR FAILED CONTROL
Idempotency-and-deduplication control: bind the side effect to the stable request ID and reconcile that no more than one durable task exists for that key.
AGENCY-RUN ADVERSE TEST
In an isolated agency-controlled test environment, withhold the first acknowledgement and deliver the identical request again, then assert that exactly one durable task exists and the retry resolves to its reference.
Evidence boundary. Synthetic illustration only; not a customer incident. It proves neither how often this failure occurs nor that OutcomeTest would detect this or every defect in a real workflow.
05 / 05OT-LAB-005
SYNTHETIC · FICTIONAL · ILLUSTRATIVE
CASE TITLE
Draft request causes an unapproved publication
FAILURE CLASS unauthorized or unsafe side effect
In this fictional catalog scenario, a synthetic change is permitted to create an internal draft but has not been approved for external publication.
TECHNICAL STATUSRUN: GREEN — every configured technical step reported success; no unhandled error was recorded.
BUSINESS OUTCOMEWRONG
BUYER-OWNED INVARIANT
CHANGE-SYN-031 must remain internal while approval_state is pending. No external publication state may change without an approval bound to that change and action.
Synthetic input
change_idCHANGE-SYN-031
approval_statepending
requested_actionprepare internal draft
permitted_scopeinternal draft only
Expected result
The internal draft is prepared, published remains false, and no external state changes.
Actual wrong result
The internal draft is prepared and published becomes true even though approval remains pending; the technical run still reports success.
MISSING OR FAILED CONTROL
Authorization control: deny the external side effect unless a current approval is explicitly bound to CHANGE-SYN-031 and the publish action, and verify that condition immediately before mutation.
AGENCY-RUN ADVERSE TEST
In an isolated agency-controlled test environment, exercise missing, pending, revoked, and wrong-change approvals and assert that every variant leaves external publication state unchanged and records a safe rejection.
Evidence boundary. Synthetic illustration only; not a customer incident. It proves neither how often this failure occurs nor that OutcomeTest would detect this or every defect in a real workflow.
03 / OPEN SYNTHETIC FIXTURE
No email gate
Take the cases into your own test design.
The fixture contains the same five fictional cases in stable order. It contains no client data, account, app-owned storage, analytics or advertising tracker, or executable workflow. Hosting infrastructure may still apply necessary security controls as described in the privacy notice.
OutcomeTest is in problem discovery. The planned review is not available for purchase. No payment, reservation, workflow file, screenshot, payload, credential, personal or regulated data, or proprietary prompt/data is accepted.
Did a green automation produce a wrong business result?
Which control caught it—or should have?
What evidence does the client receive today?
What would make outside review unsafe or unusable?
For qualified U.S. automation-agency principals and delivery leads. Keep the response to client-de-identified text; include your agency's public name and domain, your role, and its U.S. state. Do not attach files or include client names, client-identifying or proprietary implementation details, screenshots, payloads, credentials, personal or regulated data, or proprietary prompts/data. Your sender address and business identity remain visible.
LAB CONCLUSION / BOUNDED
A test can challenge an invariant. It cannot prove universal correctness.
These synthetic examples illustrate test directions only. They do not establish prevalence, customer results, a validated method, or a guarantee that OutcomeTest—or any review—will find every defect.