Transport success is not delivery
2026-06-25
The green
The retrieval step printed a tidy confirmation: the document had been copied to the clipboard, all 292 lines of it. The next step, which needed that document as its starting point, was cleared to proceed.
What was actually true
The clipboard held nothing usable. The content never arrived where the next step was going to read it from. The only thing that existed was the console output, including the confirmation line itself. The writing work that depended on the document stalled until the document was fetched again by a different route.
Why the check passed
The confirmation line was accurate about what it measured. The source document really did have that many lines, and the copy command really did return without an error. What the line did not measure was the only thing the next step cared about: whether the content was now present at the destination, in a form the next step could consume.
The copy command was a transport. A transport reporting success is telling you that it accepted the payload and believes it handed it off. In some environments the destination silently discards what it is given, and the transport has no way to know. The handoff reports success either way.
The line count made it worse, not better. A specific number reads like a measurement of the result. It was a measurement of the input. A precise figure next to the word "copied" carries an authority that a bare "done" would not, and that authority belonged to the wrong end of the pipe.
Most pipelines are full of steps shaped like this one. An upload returns OK. A message queue accepts the message. A mail service returns an identifier. A sync job finishes. Each of them reports that the transport did its part. None of them, on its own, tells you that the payload is now sitting where the next consumer expects it, intact and readable. That is a separate fact, and it has to be checked at the destination.
How it surfaced
The person on the receiving end went to use the content and found nothing there. That was immediate, and the cost was small: a stalled step and a refetch.
It surfaced quickly only because a human was the destination and noticed the absence. Had the destination been another automated step, it would have received empty input, and it might well have treated that as a valid empty document and carried on. The failure would then have shown up much later, somewhere far from its cause, looking like a different problem entirely.
The rule
Verify the payload at the place it was delivered, never the transport's report that it was sent.
What it changed
Any content that another step depends on is now written to a file and read back before it is used, and the read-back is what gets checked. A success line from a copy, an upload, or a send is treated as a claim to confirm, not as a result. The operator trusts the landed artifact, re-read at the destination, and nothing upstream of it.