v1.1.0

Code, prompts, frozen splits, per-flow predictions and the analysis behind every table of the Jev-IDS manuscript (NSL-KDD paper split, results/paper/).

The per-flow predictions (predictions.jsonl) are identical to v1.0.0. Only the analysis code, the derived CSVs and the docs changed.

• jev-ids sign-test: flow-level exact sign test with Holm adjustment (Table 4), written to results/paper/sign-test-jev-gemini.csv and sign-test-jev-random_forest.csv.
• jev-ids compare now reports one row per seed, the per-seed McNemar test of Section 6.3.
• summary.csv gains the median and 95th percentile latency of successful requests (Table 5), recall of known attacks per category (Table 3), false alarms and missed attacks per seed, and the shares of Gemini calls with reasoning and with cache reads (Section 6.4).
• make paper-summary regenerates every table from the predictions.
• README and docs/results.md report the same tests and latency summaries as the paper. docs/results-f1.svg, which showed the pooled McNemar test, is removed.

v1.0.0

Code, prompts, frozen splits and per-flow predictions behind the Jev-IDS manuscript (NSL-KDD paper split, results/paper/).