embedded-observer-effect-review

name	embedded-observer-effect-review
description	Use when target-local logging, recording, collection, tracing, profiling, or measurement could change scheduler, power, thermal, I/O, memory, wakeup, or workload behavior. Do not use for ordinary server observability with no embedded physical-footprint risk.
metadata	{"short-description":"Embedded observer-effect review"}

Purpose

Use this skill to check whether the observer changes the behavior being observed on embedded or target-local systems.

It answers one question:

Could this logging, recording, tracing, profiling, collection, or measurement path become part of the workload or hide the true physical footprint?

When to use

Use this skill when a change adds or modifies:

• target-local logger, recorder, collector, tracer, profiler, or measurement harness
• high-frequency observation
• workload-sensitive paths where extra wakeups, I/O, serialization, or allocation can perturb results
• resource claims based on instrumentation that runs on the target

Do not use it for ordinary server-side logs/metrics/traces unless the instrumentation is target-local and resource-sensitive.

How to use

1. Identify the observer:
   • component
   • cadence
   • data captured
   • storage/transmission path
   • default vs debug mode
2. Compare observer-on and observer-off behavior where possible.
3. Review perturbation vectors:
   • scheduler and wakeups
   • CPU and allocation
   • storage writes and flash wear
   • network/radio use
   • thermal behavior
   • lock contention or queue pressure
   • timing/jitter impact
4. Decide whether the observer can be enabled by default.
5. Define mitigation:
   • sampling
   • throttling
   • ring buffer with bounded drops
   • debug-only mode
   • off-target collection
   • event-driven capture
6. Write reports/resource/observer-effect-review.md using the template.
7. Route measurement scenarios to embedded-nfr-harness-design and final decision to embedded-nfr-gate.

Outputs

Produce:

• reports/resource/observer-effect-review.md

Gotchas

• Common pitfall: measuring a workload with a logger that causes the workload. Instead: compare observer-on and observer-off or mark observer overhead unknown.
• Common pitfall: adding verbose target logs to diagnose a resource issue. Instead: throttle, sample, or make the observer debug-only until measured.
• Common pitfall: hiding flash wear in diagnostic capture. Instead: include storage write rate and retention in the observer review.