I have sat through more than one bake-off between Nautobot, NetBox, and ServiceNow, and the team always leaves with a decision. Six months later the tool is populated, the API is wired up, and nobody trusts what is in it. The tool was never the gap.

A source of truth stays authoritative only as long as someone keeps answering three questions: who can write to it, what triggers a write, and what happens when two writes disagree. None of those are settings you configure once during implementation. Most organizations stop answering them right after the rollout project closes, which is the same pattern that sinks patch management projects: the automation works, and the process it depends on was never defined.

A field technician swaps a switch, the ticket closes, and the source of truth still lists the serial number of the device that left the rack. Nothing in the change process required the swap to produce a write back to the record. That is staleness: a record that survives every event that should have invalidated it.

The network team’s automation owns the IP assignment field through the API. An engineer under pressure at 2am edits the same field directly in the GUI to unblock an incident. Neither system knows the other write happened, so the field holds whichever value landed last with nothing recording which one that was. That is an authority gap: two writers with an equal claim to the same field and no rule for which one wins.

A tool can enforce field-level permissions. It cannot decide whether a manual override should outrank the next automated sync, or whether the sync should overwrite the override and erase the reason someone made it. Most organizations never write that rule down, so the next scheduled sync silently reverts the override, and the incident fix disappears with no error and no alert. That is a conflict resolution gap: no rule for which write wins, so the last one always does.

None of that shows up on an RFP scoring sheet. It shows up the first time an auditor pulls the source of truth against a discovery scan and the two do not match, months after the team that picked the tool has moved on to the next project.