Definition
Remediation Masking Rate (RMR) measures the percentage of automated remediations where the same failure symptom recurs within a defined window. It answers the question MTTR structurally cannot: did the fix stick?
RMR is Pilot-Ready.
The Measurement Gap
It's 2:47 AM on a Tuesday. Memory utilization on three application servers crosses 90%. Twelve months ago, this would have woken the on-call engineer, who'd SSH in, check the process list, identify the culprit, and either kill the runaway process or escalate to the application team. Resolution: 15 minutes to an hour, depending on experience and the complexity of the root cause.
That was the old world.
Your AIOps platform detects the anomaly, matches it against a remediation runbook, and automatically restarts the application services. Memory drops. Alert clears. A ticket opens, auto-resolves, and logs: "Automated remediation executed successfully. Service restored." Elapsed time: 4 minutes. No human involved.
MTTR improved by 90%. SLA compliance is perfect. Every metric on the operations dashboard says the system is working.
Except it isn't.
The memory spike came from a connection pool leak introduced in last Thursday's deployment. The leak is still there. The auto-remediation masked the symptom: the disease is untouched. The same remediation will fire tomorrow night. And the night after that. Until someone notices the pattern buried in the remediation logs, or the leak degrades enough that restarting can't keep up and a real outage occurs.
Your current metrics can't see this. MTTR measures how fast incidents are resolved. It doesn't ask whether they stay resolved. SLA compliance measures whether service targets were met. It doesn't ask whether the service is being held together by automated restarts. Incident volume measures how many events occurred. It doesn't distinguish between a genuine fix and a suppression that will recur tomorrow night.
The DORA metrics don't help either. Change failure rate measures whether deployments cause incidents, but if the connection pool leak gets auto-remediated before anyone classifies it as an incident, the deployment that introduced it never gets flagged. Deployment frequency and lead time measure how fast you ship, not whether the things you shipped are degrading quietly under automated cover.
MTTR is the metric that makes this failure invisible. When auto-remediation executes in 4 minutes and the alert clears, MTTR logs a success. When the same auto-remediation fires again seven days later for the same reason on the same service, MTTR logs another success. The system looks like it's working. You're watching a metric that improves as the underlying reliability degrades.
The failure mode has a precise shape: automated remediation restarts services and clears alerts without ever investigating or addressing the underlying cause. MTTR improves while failure recurrence increases. No standard metric connects these two facts.
The measurement gap: there is no metric that tells you whether automated remediation is suppressing symptoms without addressing causes, which means your MTTR can improve steadily while your system accumulates an invisible backlog of unresolved root causes.
Formula
Remediation Masking Rate (RMR) measures the percentage of automated remediations where the same failure symptom recurs within a defined window. It answers the question MTTR structurally cannot: did the fix stick?RMR is Pilot-Ready. The data already exists in the logs your AIOps and monitoring platforms generate. The work is the recurrence query, not new instrumentation.
In Plain Language
take every automated remediation event. Check whether the same alert signature fired again on the same target within the next 7 days. Count the ones that did. Divide by the total.
Formula
RMR = Auto_remediations_with_recurrence_within_N_days / Total_auto_remediations × 100
Where:
- Auto_remediations_with_recurrence = automated remediation events where the same alert signature fires again on the same host, service, or resource within N days
- Total_auto_remediations = all automated remediation events in the measurement period
- N = recurrence window, recommended starting value: 7 days. Adjust based on your deployment cadence. If you deploy daily, 3 days may suffice; if you deploy biweekly, extend to 14.
Worked Example
An operations team runs AIOps with automated remediation across 200 production services. Q1 generated 340 automated remediations. Checking recurrence within a 7-day window:
- 218 remediations: no recurrence on the same target within 7 days. Durable
- 87 remediations: same alert recurred once within 7 days. The fix worked temporarily
- 35 remediations: same alert recurred 3 or more times within 7 days. Chronic suppression
RMR = (87 + 35) / 340 × 100 = 35.9%
The operations dashboard showed 340 incidents auto-resolved, MTTR of 4 minutes average, and 99.97% SLA compliance for the quarter. Nothing on that dashboard indicated that 122 of those resolutions were temporary. The 35 chronic recurrences (services being restarted on a nightly or multi-daily cycle) represent the highest-risk items: unresolved root causes the automation is papering over, any one of which could escalate if the recurrence rate increases or the remediation begins to fail.
Sign Convention
Lower RMR is better. Zero percent means every automated remediation was durable: the problem didn't return. Rising RMR over time means the system is accumulating unresolved root causes faster than they're being discovered and fixed.
Edge Cases
Alert signature granularity matters. "High CPU on server X" is too coarse: different root causes can produce the same alert. "High CPU on server X from process Y" is better. If your alert taxonomy is too broad, RMR overcounts false recurrences. Start by measuring with your current granularity, then refine the signature definitions as you learn where the false positives cluster. Cascading remediations: if remediation A triggers a cascade that causes event B, which also gets auto-remediated, count only the originating alert. The cascade is a separate failure mode. Minimum sample: 50 automated remediations per measurement period for the numbers to be directionally meaningful. Cross-references: Upstream, RMR depends on ACA (AIOps Correlation Accuracy, GMI-IT-004). If the correlation engine is over-grouping alerts, masked root causes may resurface as "new" incidents rather than recurrences, making RMR understate the true masking rate. Downstream, RMR feeds OSAI (Operator Skill Atrophy Index, GMI-IT-006) directly: high RMR means engineers get fewer reps at real root-cause investigation, which accelerates the skill degradation OSAI measures. In MSP environments, RMR variability across clients drives Automation Effectiveness Variance (AEV, GMI-IT-032). Maturity: Pilot-Ready. Everything needed to compute RMR exists in the logs already generated by your AIOps platform, monitoring stack, and ITSM. The work is writing the recurrence query, not adding new instrumentation or new fields.
Proposed formula, pending validation pilots.