Performance and adversarial-regression contract

base-cli treats startup and filesystem behavior as part of its public quality contract. The checked benchmark is intentionally small and runs from the source checkout:

python scripts/benchmark_runtime.py --check

It records fresh-process import time and the cost of an isolated production invocation through base_cli.testing.invoke. The CI quality job checks the sample p95 against these budgets:

Measurement	Budget
Fresh import base_cli	750 ms
Isolated invocation and runtime filesystem setup	1,500 ms

The benchmark reports the median, p95, and maximum for seven samples. These budgets are intentionally broad enough for hosted runners while still detecting accidental quadratic startup work, unbounded metadata scans, or unexpected dependency imports. A performance improvement should preserve the same lifecycle and persistence assertions covered by the adversarial tests.

The regression suite uses deterministic Hypothesis examples (derandomize enabled), fixed multiprocessing workloads, and explicit seed values in every worker payload. Property cases cover redaction and command-protocol framing; spawned processes cover history append, private metadata replacement, logging, extension discovery caches, and run-bundle retention. Ctrl+C is tested through both the lifecycle boundary and a real POSIX subprocess signal. Windows keeps the portable lifecycle and persistence checks while skipping only assertions that require POSIX signal or descriptor semantics.