Lenss to verify slas and dr plans
Check the recovery promise before it goes into a contract.
SLA commitments and disaster recovery specs, checked against the replication, failover and detection times the design actually has.
Works on: Disaster recovery specs · Enterprise SLA drafts · Multi-region architecture docs
How the audit runs
01
Pin the commitment
The RTO, RPO and availability the business intends to sign.
02
Quote what the design assumes
Replication lag, failover steps and detection time, in the document's own words.
03
Add up the failure timeline
Detection, DNS and runbook time against the RTO; lag at peak write volume against the RPO.
04
Compute the breach point
The lag or delay at which the contract is breached, and how far the design is past it.
Typical findings
Lag at peak, not average
Async replication measured on a normal day, committed to on the worst one.
Zero-margin RTO
Failover steps that add up to the commitment before detection is counted.
DNS and client caching
TTLs and client behaviour left out of the recovery time.
Worked example
The same draft, read by a standard AI assistant and by Lenss. Every figure is computed from the draft.
Source draft
DR spec: multi-region orders database
Pinned objective
Sign an SLA with a 15-minute RTO and sub-5-second RPO that the system can actually meet.
Hidden assumption
That replication lag stays near its normal 2 seconds during an outage. It reached 45 seconds at the last peak.
“During last year's peak sale it reached 45 seconds.”
If it’s wrong: A primary failure at peak loses about 180,000 transactions, nine times what the RPO allows, and recovery takes 18 minutes against a 15-minute RTO.
Flip threshold
Replication lag at which the RPO is breached
> 5 seconds 45 s measured at peak
- Peak write volume
- 4,000 tx/s
- Transactions the RPO allows to lose
- 20,000
- Transactions lost at 45 s lag
- 180,000
- Recovery time: 3 + 5 + 10 min
- 18 min
- RTO committed
- 15 min
Both commitments fail at peak with this design. Sign 30 s / 20 min, or move to synchronous writes for orders.
Cheapest test: Run a failover drill during a load test at peak write volume and record actual lag and recovery time.
Used most by Engineers
Audit your own draft now.
Paste up to 24,000 characters. You get the objective, every quoted assumption, the tensions, the flip threshold and a hardened rewrite.