An Andon cord makes a problem visible early so it can be corrected before it spreads. In software, divergence between code and its specification is an Andon moment: pause long enough to resolve the mismatch using the specification’s decision boundaries. Escalate only when the intent itself is ambiguous. section 4.1.4
In production, pulling the cord does not mean stopping the whole system. Each failure calls for a deliberate action, such as rolling back a bad deploy, pausing a resource-heavy job, or disabling a problematic feature flag. Define these actions in advance: the wrong intervention can make the problem worse. section 12.3.2
Listing 4.4 Spec file
Accept: GET /api/orders/export (authenticated user) Return: CSV file with columns: order_id, date, total, status Filter: Only orders belonging to the requesting user Format: Standard CSV, UTF-8 encoding Never: Include other users' data, build the CSV in the browser, load all orders into memory, allow unauthenticated access Prefer: Stream CSV rows from the server using bounded database reads
A concise export specification makes expected behavior explicit, giving the team a basis for spotting divergence. section 4.1.4
Table 12.3 What "pull the cord" means in practice
|
Failure mode
|
Andon action (the cord)
|
What it stops
|
What keeps running
|
Implication if wrong
|
|
Bad deploy causing errors
|
Roll back to previous version
|
The defective code path
|
All traffic on the known good version
|
Rollback introduces its own change; must be tested
|
|
Dependency failure (API, DB)
|
Trip circuit breaker
|
Calls to the failing dependency
|
Degraded service with fallback or cached data
|
Users see partial functionality, not full service
|
|
Traffic spike overwhelming capacity
|
Shed load via rate limiting
|
Excess requests beyond capacity
|
Priority traffic within safe limits
|
Some legitimate users get rejected
|
|
Feature flag causing latency
|
Disable the flag
|
The new code path
|
All traffic on the old code path
|
Assumes old path still works; must not have been removed
|
|
Batch job exhausting shared resource
|
Pause the batch job
|
Resource consumption by the job
|
All user-facing traffic unaffected
|
Batch data arrives late; downstream reports may run
against incomplete data
|
|
Data corruption from migration
|
Halt pipeline, freeze writes
|
Further data modification
|
Read traffic continues; writes queue or fail
|
Extended read-only mode affects transactional workflows
|
The table maps failure modes to specific actions, what each action stops, and the risk if it is wrong. section 12.3.2
