Statistical Model

Why statistics matter here

Time-based inference depends on response latency differences, but network jitter and server variance introduce noise. A fixed timing threshold can misclassify conditions under unstable latency.

Decision strategy in StatSQLi

TimingAnalyzer applies a one-tailed Welch t-test:

• compare baseline sample set vs delayed-condition sample set,
• use unequal-variance assumptions (equal_var=False),
• infer delayed condition as true when p_value < alpha.

With default confidence 0.99, alpha is approximately 0.01.

Baseline estimation

Before condition tests, the extractor collects baseline timings with a false condition (1=0) to model normal latency.

The analyzer computes:

• mean baseline latency,
• sample standard deviation,
• optional adaptive threshold estimate (mean + 3 * std style heuristic).

Sample-size behavior

The implementation enforces minimum sample counts (min_samples) to avoid overconfident decisions from tiny datasets. It also includes a helper for rough sample-size estimation based on expected effect size and variance.

Practical interpretation

• Lower p-values increase confidence in true delay effects.
• High variance demands larger sample sets or stronger delays.
• Statistical detection improves reliability but still depends on transport stability and endpoint behavior.