File size: 2,643 Bytes
bf0fb33 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 | """
L0 -- Adversarial Validation Engine.
Sits outside the L1-L9 stack rather than inside it: every other layer defends
the run; this one checks whether the layers actually hold, and whether they
hold TOGETHER, not just individually.
Two things are honestly buildable in this environment, and one is not.
BUILDABLE, AND BUILT HERE:
1. `validation.properties` -- property-based fuzzing (Hypothesis) of the pure,
deterministic logic every other layer's correctness depends on: canonical
encoding, merkle commitments, Gate 1's structural invariants. Every other
test in this repo is example-based -- a human picked the attack. Fuzzing
asks the same question across thousands of machine-generated cases a human
would not have thought to pick, which is a different and complementary kind
of adversarial validation.
2. `validation.scenarios` -- full-stack composed scenarios that wire real
components together (`Supervisor`, `Broker`, `EnvironmentMonitor`,
`KillSwitch`, `DriftMonitor`) and drive them through one coherent run, then
check the SYSTEM's outcome. Every layer up to this point has been tested in
isolation, which is necessary but not sufficient: `killswitch.Evidence` is
deliberately assembled from independent inputs
(`monitor`, `gate1_violation_codes`, `broker_history`, `denial_signal`,
`ledger_events`), and nothing before this module has actually driven a real
`Broker` or a real `EnvironmentMonitor` and fed their REAL output into a
real `KillSwitch` end to end. `Evidence.from_run`'s signature was built for
exactly this composition; this module is its first real caller.
NOT BUILDABLE HERE, AND NOT PRETENDED:
Whether Firecracker itself isolates correctly, whether the fanotify/eBPF
collector L7 assumes actually delivers what `MutationObservation` expects,
whether vsock enforces the one-direction-only property `transport.py` assumes
-- none of that can be checked from this sandbox, which has no `/dev/kvm` and
no kernel privileges. `containment.backend.FirecrackerBackend` says so in its
own docstring: "reviewed but not exercised... until the L0 adversarial
validation engine runs it on real hardware." That sentence is still true after
this module. L0 as built here validates COMPOSITION and LOGICAL INVARIANTS in
software; it does not and cannot validate the hypervisor underneath them. See
`docs/l0_validation.md` for the full scope statement.
"""
from validation.scenarios import (
ScenarioResult, ScriptedBackend, SCENARIOS, run_all,
)
__all__ = ["ScenarioResult", "ScriptedBackend", "SCENARIOS", "run_all"]
|