01Someone complains.
The customer noticed before you did.
Hours to days. It costs a ticket and the relationship.
What happens today
Four ways to find out a flow stopped. All of them late.
02The scheduled check.
A reconciliation job, a nightly audit, a consistency sweep.
Accurate, and it tells you at the end of the cycle, not during it.
03The one flow you watch by hand.
You built a step-by-step check for your most important flow. It works.
It took a project, and nobody had time to build one for the rest.
04The dashboards.
Every service reported healthy the whole time.
They never alerted, because nothing broke. Green dashboards are why nobody looked.
How it works
Learn the flow. Catch the stop. Keep the proof.
No replacement stack and no new query language required.
01 / ANALYZEStart with one flow.
Bring one day of evidence. Get back the steps Clutta found, the id that ties them together, and where the flow stopped, with the records cited.
One-shot investigation02 / SCANKeep watching.
Clutta learns how each flow normally runs from repetition, then names the run that stops partway, against the deadline that flow usually meets.
Continuous detection03 / CASE FILESKeep the proof.
One Case File per stop, with every affected run and the records behind it, not a thousand alerts. The investigation starts where the search would have ended.
Operational memoryWho this is for, and how to judge it
Five questions worth asking every vendor. Including us.
For teams that built one completion check by hand and have far more flows than checks.
How long between a flow stalling and a person knowing?
Measure it against your batch run, not a dashboard refresh.
Does catching a new kind of failure need someone to write a rule first?
If yes, you only catch what you already predicted.
Does the alert arrive with the evidence, or does it start a search?
The search is most of the cost.
Can you prove the fix worked, or do you assume it?
A retry that quietly fails again is the same incident twice.
Does it cover flows that cross systems you do not own?
That boundary is where most of these failures happen.
Proof, not posture
Every claim on this page opens into its evidence.
Independent research shows the problem is far larger than one incident.
Case 01 · A valid key was rejectedHappened in our own systems
401
rejected, while every service reported success.
Every service returned 200
Authentication answered in 3.6 ms
The key was valid
Nothing crashed, nothing timed out
what the sending service returned
{"isValid": true}
what the receiving service expected
{"data": {"isValid": true}}
one field in the wrong place -> read as false
false -> a valid key rejected
the evidence
Every service was healthy.
The message between them was not.
Clutta tied the healthy services, the mismatched message and the rejected request into one traceable chain.
Case 02 · Checkout stopped before confirmationValidated case study
3 of 4
steps happened. The confirmation never did.
Storefront healthy
API healthy
Payment authorised
No error, no timeout, no alert
how this flow normally runs, learned from watching it
order accepted -> payment authorised -> order recorded -> confirmation sent
what happened this time
order accepted -> payment authorised -> order recorded -> (nothing)
the finding
Step four never arrived.
Clutta opened a Case File with the records of the three that did.
Nobody told Clutta what this flow should look like. It learned that from watching, and the dashboards stayed green the whole time.
A measurable gap in modern operations.
At scale the failure hides inside the average. A thousand customers stuck out of a million is 0.1 percent on every dashboard and a thousand tickets by Monday. Every new integration adds a flow where that can happen, and a proper check for one still takes a quarter.
name complexity and overhead as their largest observability obstacle.
Grafana 2025still learn about interruptions through complaints, tickets, or manual checks.
New Relic 2025silent semantic failures studied across nine distributed systems.
USENIX OSDIOne system. One day of evidence. Read only.
Nothing installed in your services. Point Scan at one system for a day. We come back with the flows it found and which of them are ready to check. Nothing is activated until you approve it.
What Clutta will not do
- It is not a dashboard. You already have those.
- It is not a chat. It answers with records, not opinions.
- It never changes anything in production.
Get Clutta
curl -fsSL https://clutta.io/install | sh
Start with the evidence you already have.