State What a Zero-Failure PCB Test Actually Demonstrates

Testing a group of boards without a failure is useful evidence, but it does not prove a zero failure probability. The strength of the conclusion depends on how many independent units were tested, what they experienced, and what counted as a failure. State those conditions alongside the result.

NIST’s exact binomial confidence-limit reference provides methods for proportions under a binomial model. That model is one possible framework for a defined pass/fail trial; it is not automatically appropriate for every life test or repeated measurement.

Define the trial and failure criterion

Specify the board revision, population, operating conditions, test duration, and acceptance criterion. Distinguish unique physical units from repeated executions on the same board. Repeating a test many times on one assembly does not automatically create the same evidence as testing independent assemblies.

Record withdrawals, interruptions, and incomplete tests. A board removed before completing the exposure should not silently enter the completed-pass count. Review its status according to the analysis plan. Preserve the sample-selection method too; deliberately choosing easy-passing boards does not support the same population inference as a representative planned sample.

Use a clearly bounded statistical statement

For an illustrative binomial case with independent trials sharing one failure probability, observing zero failures in n trials gives the one-sided upper confidence bound p = 1 − alpha^(1/n), where confidence is 1 − alpha. With 100 completed trials and 95% confidence, that bound is approximately 2.95%, not zero.

This calculation describes the specified trial under its assumptions. It does not translate a short room-temperature functional check into years of field reliability. A different population, correlated failures, or a different exposure can require a different model.

Keep engineering evidence alongside the count

1. Retain test settings and raw results.
2. Review whether the sample covers relevant variants.
3. Investigate any abnormal behavior below the formal failure threshold.
4. State which conditions remain untested.
5. Plan additional work around the unresolved product risks.

Do not choose a confidence statement after seeing the data merely because it sounds impressive. Define the intended claim and study plan in advance with the responsible reliability reviewer.

The accelerated-test guide addresses the additional evidence required before translating stressed test results into a projection at normal use conditions.