GUID-shaped target systems
What you're seeing
A Medium data-health finding:
guid_target_system — N record(s) in the last 24 hours name a per-run id as the target system (e.g.
dbd32a6a-471a-4f33-a69e-69a6629afccd,0d5ea8d9-9d1f-4a1e-8832-526732c0ce42); every run appears as a new system, which breaks reach, coverage and cost grouping.
What it means
The emitter is putting a unique-per-run identifier in action.targetSystem. The platform treats each new value as a new system, so one real system explodes into thousands of fake ones. Everything keyed by system degrades: the agent's reach list becomes noise, coverage has thousands of unattested pairs instead of three, and cost per system is meaningless.
How it shows up on the Brief
Besides the finding, the Brief reports N systems touched for the first time and the number is inflated: most of the "new systems" are run ids. If the count of first-time systems jumps overnight on an agent that has not changed, this is the likely cause.
What to do
- Confirm the shape. Open one record in Investigate and look at the action's target: if it is a UUID/ULID that changes every run, this recipe applies.
- Fix the mapping at the emitter. The contract is:
targetSystemis the stable name of the system being touched (e.g.cruzclaw-backend); the per-run or per-object id belongs intargetEntityId. Change the emitter's mapping so the stable name goes intargetSystem. - Decide what to do with history. Existing records keep their stored values; changing an alias does not rewrite them. Two options:
- Let the old GUID-named pairs age out of the 30-day windows. Acceptable when the volume is low and the windows will roll over before any audit.
- Use the alias backfill (Settings → system aliases, or
POST /v1/system-aliases) to map the known GUID shapes to their stable names, so history groups correctly immediately. Add an alias per GUID shape you recognise.
- Resolve the finding with the cause and the fix date once new records arrive with stable names. Verify by opening the agent's Reach tab: it should list a handful of named systems, not thousands of GUIDs.
What good looks like
The agent's reach list reads like an inventory (cruzclaw-backend, Cruz runs), coverage pairs are countable, and the data-health finding stays resolved because the emitter keeps sending stable names.
If you cannot change the emitter
Some vendor agents hard-code their payload shape. In that case the alias layer is the standing fix: map each observed GUID to its stable name, and re-check monthly for new shapes. A systemic alias rule (by prefix or pattern) can be requested via support — it is on the roadmap as a first-class mapping feature.