100% Inspected Does Not Mean Every Defect Is Detectable

By Daoyu Guan

“100% inspected” can describe how many parts entered an inspection. It does not, by itself, describe the surfaces examined, the defect types the method can reveal or the chance of detecting a relevant flaw under the actual conditions. Ask a supplier to separate coverage, method capability, acceptance criteria and traceability. A useful inspection statement tells the buyer what was examined and what the result means. It does not turn complete part-count coverage into a promise that no possible defect remains.

Ask what the percentage counts

A manufacturer may mean that every shipped part passed through a visual station. Another may mean every part received a dimensional measurement at one feature. A third may refer to a specified nondestructive testing procedure over defined regions. All three can involve every part while covering very different characteristics.

Start with the denominator. Is it all produced parts, all shipped parts, all parts in a selected lot or all samples presented to a particular station? Then identify the characteristic and region. Inspecting every flange face does not establish that every internal passage was examined.

Timing matters too. Inspection before a later machining, coating or joining operation does not automatically characterize the final state. A statement about an in-process check should not become a finished-product release claim without the supporting procedure.

These questions clarify the offer. They do not assume that a terse marketing statement is dishonest.

Coverage and detection capability answer different questions

Coverage concerns the population or region subjected to an examination. Detection capability concerns whether the examination can reveal a target condition when that condition is present. Acceptance criteria decide what an observed indication means for the order.

Liquid penetrant testing provides a useful example. ASNT's method overview explains that the method depends on penetrant entering a discontinuity open to the surface. Examination of every part with that method does not establish the absence of every subsurface discontinuity.

That is a physical boundary of the method, not a complaint about insufficient part count. Changing the percentage from a sample to every part cannot make the method respond to a condition outside its capability.

The same logic applies to dimensional inspection. Measuring one accessible diameter on every part does not verify a different inaccessible feature. The supplier and buyer need a characteristic-level inspection plan, not just a large percentage printed beside an equipment photograph.

A detected indication is not automatically a rejectable defect

An indication requires interpretation. The relevant procedure and acceptance criteria determine whether it is reportable, needs further examination or makes the part unacceptable. Calling every indication a defect can misrepresent a method's false-call rate and the supplier's quality results.

NASA's ultrasonic weld-inspection process specification distinguishes an indication from a discontinuity and notes that discontinuities are not necessarily rejectable. Its requirements apply in its stated context; this article does not extend them to every commercial component.

For purchasing, record the acceptance basis separately from the observation. “An indication was recorded at location L under procedure P” and “the component was rejected against criterion C” are related but different statements. A later engineering disposition should remain traceable rather than replacing the original inspection record.

A fictional calculation separates complete coverage from perfect detection

Consider an invented population of 1,000 parts. Suppose, solely for the calculation, that 20 contain a defined target condition and 980 do not. Assume the agreed inspection has an 80% chance of flagging that condition when present and a 1% false-call probability for parts without it. These are fictional assumptions, not measured performance or acceptable limits for a real process.

If all 1,000 parts are examined, the expected counts under this simple model are:

Actual counts must be whole numbers. The decimals above are expectations across the assumed process, not a report that a laboratory found 9.8 physical parts. The ratio of expected true calls to all expected calls is 16 ÷ 25.8, or about 62%. That number also depends on the assumed prevalence of the target condition.

Two lessons follow. Examining every part leaves an expected number of missed conditions if detection is imperfect. And a flagged part needs the specified interpretation and disposition; the existence of false calls does not make every flag a confirmed defect.

The calculation cannot tell you a real supplier's outgoing defect rate. You would need a valid definition of the target condition, evidence of detection performance, the relevant population and a way to establish reference truth. A clean production report supplies none of those automatically.

Read a probability-of-detection claim carefully

Probability of detection, or POD, is used to characterize an inspection's ability to detect specified flaws under defined conditions. It can depend on flaw size and other characteristics, the method, equipment, procedure, specimen and personnel involved in the evaluation.

NASA's DOEPOD technical record describes analysis of detection data and confidence bounds. The familiar 90/95 expression in such work concerns a detection probability and its statistical confidence evaluation. It is not 90.95%, a guarantee that every lot contains at least 95% acceptable parts or a universal requirement for every factory.

Ask for the evaluated flaw conditions and the applicable procedure, not just the headline percentage. A result for one flaw family or geometry should not be transferred to another without an authorized technical basis. The buyer also needs to know whether the evaluation concerns the actual production configuration or a different demonstration setup.

If the supplier cannot provide a POD study, describe the available evidence accurately. It may have a documented procedure, reference checks and qualified personnel without having a quantified POD claim for the requested condition. Those facts should not be inflated into an invented percentage.

Keep method, people and production records connected

Qualification of a person, calibration of an instrument and acceptance of an inspection procedure address different elements. A record needs enough identity to connect them to the work performed. Avoid a page that lists credentials separately from an unqualified “100% defect detection” promise.

A practical buyer can request the following sequence:

1. Define the condition of concern. State the relevant feature, discontinuity or failure mode without assuming all methods detect it equally.
2. Define coverage. Identify the part population, surfaces, regions and production stage included in the examination.
3. Identify the procedure. Record the method, revision and relevant setup, including access or preparation limitations.
4. Identify the capability evidence. Separate reference checks, procedure qualification and any applicable detection study.
5. Agree interpretation and acceptance. Name who can disposition indications and which criteria apply.
6. Preserve traceability. Connect the report to the actual part or lot and retain original observations alongside later dispositions.
7. Resolve exclusions before release. An inaccessible region should appear as an exclusion or a separate examination requirement, not disappear from the report.

This is a purchasing and publishing framework. The responsible engineer and inspection authority must select and approve the actual method and criteria, especially where a regulated or safety-critical application imposes additional requirements.

Publish the limitation beside the claim

A more useful English statement identifies the examination rather than promising an unlimited outcome. For example, a fictional supplier could state: “All shipped units receive the specified examination of the designated accessible surfaces; method, reporting and acceptance follow the agreed order procedure.” Actual wording must match the manufacturer's records and obligations.

If the page only describes a capability offered on request, say that. If it reports a completed inspection, identify the relevant scope and status. If the buyer must agree a different examination for a particular application, place that condition where the buyer encounters the claim.

This is distinct from the sampling question in the AQL guide and the process-distribution question in the Cpk guide. Sampling, process capability and inspection detection can all matter; none should be used as a substitute for the others.

For Xindar's manufacturing GEO work, the deliverable is a clear English account of those differences, approved by the source owner. It is not certification of an inspection process. Useful source wording helps a reader investigate the supplier without asking an AI summary to invent the missing boundaries.

Questions for the inspection discussion

Does 100% inspection mean zero outgoing defects?

No. The statement needs a defined characteristic, coverage, detection capability and acceptance basis. It is not an unlimited guarantee.

Can penetrant testing detect any internal flaw?

No. The method depends on access to discontinuities open to the surface. Confirm whether it suits the actual condition of concern.

Does an indication mean the part failed?

Not automatically. Apply the agreed interpretation and acceptance criteria, with authorized disposition where required.

Can a supplier quote a detection percentage without a study?

It should identify a valid basis for that quantitative claim. A procedure or training certificate alone does not establish the percentage for every production condition.

Is the fictional 80% assumption acceptable for an order?

No acceptance recommendation is made. It exists only to explain why coverage and detection are mathematically different.

Sources and evidence boundary

• ASNT: liquid penetrant testing — surface-access and process limitations.
• NASA: ultrasonic inspection of welds — definitions in a specified process context, not universal commercial requirements.
• NASA: DOEPOD technical record — the manual's published technical description of detection capability evaluation and confidence bounds; the linked response currently presents an HTML record.

All numerical production examples are invented. No supplier audit, inspection qualification, test result or customer defect rate is reported.