Real-target evidence when no auditor is asking

Most embedded teams never see an auditor. They still pay for intermittent faults, stale coverage maps and integration regressions - the same evidence helps.

Published 2026-08-17Last technically reviewed 2026-08-16Evidence basis: Engineering analysis

Nobody audits most embedded software

The report may never be written; the losses are real anyway: an intermittent fault that appears after days of operation, a coverage map that describes an old build, an integration regression that only shows on the real device, a timing budget that erodes release by release.

The methods on this site were developed where regulation forced someone to pay for them. The methods themselves are not safety-specific - they are expensive-specific. Wherever a failure in the field costs more than the testing would have, evidence about what the real system actually did is worth having, whether or not an auditor asks for it.

Where the same evidence pays for itself

Long-running fault capture. Passive trace observation records execution without touching the timing that triggers the fault - and because processing is live, the observation window is not limited by a capture buffer. The fault that appears after days can be observed for days. Trace-based debugging

Test impact at integration level. Which tests touch the code a change modified - measured on the integrated build, not inferred from unit-test maps. Teams use it to run the tests that matter first. Coverage for test prioritization

Timing regressions before they ship. Task runtime profiling on the real silicon, continuously - so a timing budget is a measured quantity, not a number in a spreadsheet that nobody has checked since the last port. SmartTrace WCET analysis

Acceptance feedback for AI-assisted changes. When code is generated and regenerated, the question shifts from “who wrote this” to “what accepted it”. The acceptance decision can rest on evidence measured on the real target - CEDARtools.Forge, offered as a guided pilot.

A worked public example - on code no auditor asked about

SQLite is an open-source database library, not a certification object. Our first public measurement record is measured on exactly that codebase: the SQLite -O3 comparison of native trace, CEDARtools.Anchor and GCC coverage documents configuration, method, results and limits, with machine-readable downloads. It is a measurement demonstration, not a customer deployment - the distinction our evidence taxonomy insists on. What it shows is the point of this page: the measurement works the same way when the code under observation has never seen a regulator.

What this page does not claim

No universal savings figure. What the evidence costs and saves depends on the target, the trace path and the workflow, and we would rather measure that in your configuration than project it - target support is answered per configuration.