An advisor that learns what normal looks like for your deployment and surfaces drift before it becomes failure — locally, with no cloud AI. When it's unsure, it says so. The restraint is the point.
Every operator has muted an alerting tool that fired too often. Once you stop trusting the alarm, the alarm is worthless — no matter how clever the model behind it.
The advisor is built the other way around: it pauses before it speaks, runs a set of self-checks, and only surfaces a signal it can stand behind. We'd rather miss a marginal blip than train you to ignore it. An advisor you keep listening to beats one that's technically more sensitive and practically muted.
Most tooling shows you state — graphs, current values. The advisor learns the signature of how your system behaves over time, so it can tell when today doesn't look like a normal day.
This is what aura-pce level prints on a fresh install, because nothing on
your box has been graded yet. There is no code path that prints a reliability
percentage without an n. The number is yours to earn by grading, from zero,
as emissions accumulate — a vendor cannot ship you a precision figure for your own
infrastructure.
aura-pce evidence — which captures what the box looked like at the moment of
each gated emission (processes, loadavg, journal tail) so you grade from evidence, not from
memory. Full pattern composition — the engine's namesake — is the research edge we're still
walking, and we say so instead of shipping the name without the verb.
The engine's self-test gate is already public and runnable. Install it, run one command, and read every check — you receive the mechanism, open-source, never anyone else's accumulated data.
# the advisor's "I care" self-test gate — Apache-2.0
pip install aura-pce
aura-pce proveRun the gate today, or get a note the moment a hosted advisor is ready to point at your stack.