Built around the failures Make users actually hit.

The largest family by far, and the one where the error is most misleading.

Empty value vs. wrong type vs. a field Make never learned

• Empty value: the mapping is valid, the upstream bundle simply had nothing in it. The fix is upstream, not on the module that complained.
• Wrong type: text where a number is expected, a date string the app will not parse, a single item where a collection is required. The mapping is fine; the shape is not.
• A field Make never learned: the module was configured before the upstream output was ever seen with real data, so the picker offers a structure that does not exist.

These three produce nearly identical messages and have three different fixes. Separating them is most of the work.

The failure mode here is silence: nothing errors, nothing happens.

• Condition never true: the comparison looks right on screen but compares a number stored as text, or a value carrying a trailing space, or a date in another format.
• Bundles blocked upstream: a filter earlier in the chain swallowed everything, so the module downstream looks like it produced nothing when it was simply never reached.
• Overlapping route conditions: a Router can process multiple matching routes, sequentially. Check each filter; route order does not make the branches mutually exclusive.

Connection-level failures, where the scenario is usually blameless.

• Never registered: the webhook was re-created and got a new URL, while the source system keeps posting to the old one.
• Payload rejected: the source changed its content type or its structure, and the scenario is still parsing the old shape.
• Scope too narrow: the connection was authorised for reading and the operation writes. It worked for months, which is what makes it hard to spot.
• Token expired: it worked yesterday, nothing changed, it fails today. An expired token is one possibility; check the actual response and connection state.

Structure problems, where the count of bundles is the tell.

• Unlearned structure: the Iterator sees a text blob where you expect a list, because the source module never ran on real data.
• Empty array: zero bundles produced, so everything downstream silently does nothing.
• Lost grouping: the Aggregator points at the wrong source module and groups over a different loop than the one you meant. This is a configuration mistake far more often than a data one.

When a run takes far longer than it should, look at the bundle count on each module. The one that jumps is where to start.