InfraGuard Engineering Devlog, part 3 of 6.
A security score should improve when defenses work. On September 4, ours did the opposite.
The formula deducted points for observed high-severity alerts, anomaly-engine blocks and newly blocked sources. The recorded dataset contained 889 anomaly-engine blocks and 789 newly blocked reputation-based sources that day. Those successes pushed components toward maximum penalties.
Of 282 high-severity IDS alerts, 227, about 80.5%, came from already blocked ranges. They still counted against the score.
Count the unresolved remainder
We replaced volume penalties with measures of what remained unaddressed:
- IDS: high-severity alerts from sources not yet blocked, relative to observed high-severity alerts.
- Anomaly detection: above-threshold findings not blocked by the engine or current blocklist.
- Reputation: the share of observed attacking source ranges still unblocked, counted by source rather than event volume.
“The sensor observed something bad” and “we still have actionable exposure” are not interchangeable.
The formula also uses smoothing. A tiny denominator should not make one unresolved finding send a quiet environment straight to its worst score. At displayed precision, recorded examples were 100 for one unresolved item, 98 for five, and 55 for 50,000 wholly unaddressed items. These are implementation examples, not thresholds to copy without examining the formula and context.
One dataset, two answers
Two overview screens implemented different formulas: one displayed 62 and another 55 for the same data. Both now call a shared scoring service.
We also found an “unaddressed threats” metric that displayed successful blocks. The label described open work; the value counted completed work.
Recalculating the recorded dataset changed the overall score from 62 to 80. Even complete blocking would have left that snapshot at 94, because separate penalties remained for exposed critical ports and weak TLS settings.
Expose the evidence
Each component shows its unresolved and observed counts. If data cannot be read, the application should not invent a score from traffic volume.
A blocked source also does not prove the absence of compromise. This is an operational prioritization signal, not a security certification.
Takeaway: Does a metric's label match what it counts? When the number improves, does the operational outcome improve too? Ask both before putting a score at the center of a dashboard.
About this series: Adapted from InfraGuard engineering records. Measurements describe the recorded environment, not product-wide benchmarks. AI was used to draft and translate this article.
Disclosure: Solution Market develops and sells InfraGuard. This is our engineering devlog, not an independent review. InfraGuard is a self-hosted infrastructure operations and security platform. Project · Documentation · Solution Market
Top comments (0)