Application to Reclassify Schema Optional Field
A field was marked optional to resolve one compatibility problem, producing eleven default-value problems downstream.
I.
The field broke two consumers when it was added. The field was required. Making it required meant that every consumer that had not yet been updated would reject the schema. Making it optional meant that every consumer could proceed while inventing its own interpretation of what an absent value meant.
The field was marked optional.
II.
The system appears calmer after the schema optional field. Calmer is accurate in the limited sense that the specific failure no longer occurs. The specific failure was visible and had a name. The new failure is distributed, quiet, and has eleven names, one per consumer that invented its own default.
The governing rule was applied: once schema optional field passes locally, system-level doubt may be deferred to the next environment. The next environment would have better context.
III.
The team records the symptom as resolved but not the mechanism. The ticket was closed. The resolution note said: marked field optional for backward compatibility. This is true. It is not complete.
It is not complete because backward compatibility was achieved by converting a single point of failure into eleven points of quiet variance. Quiet variance does not open tickets. Quiet variance produces data that is almost right.
IV.
Months later the same condition returns with enough history to make the temporary fix look like architecture. The optional field has been in production for eight months. It has been referenced in three design documents as the existing pattern. A new engineer asked about it and received the explanation: it's always been optional.
It has not always been optional. It was made optional on a Tuesday afternoon when the required version broke two consumers and the sprint was ending.
V.
The engineer who marks a field optional to preserve compatibility and then makes every consumer invent its own default is doing something difficult. They are choosing between one documented failure now and eleven undocumented failures distributed across time.
The eleven undocumented failures are harder to measure and easier to accept.
The field remains optional. The consumers have each invented a default. Six of the defaults agree.