Test Coverage: How to Calculate What Your ICT Program Actually Catches

2026-11-12 · Joe Elsenburg
Test Coverage: How to Calculate What Your ICT Program Actually Catches

"95% coverage" can mean almost anything. Here is how to count what your ICT program actually catches — and where the number hides its gaps.

Test Coverage: How to Calculate What Your ICT Program Actually Catches

"Ninety-five percent coverage" is one of the least meaningful claims in electronics manufacturing. Two programs can both report the same number and catch entirely different sets of defects. The figure only means something once you know what was counted, and how.

This is a method for calculating coverage in a way that tells you what will actually escape to your customer.

Start with what is being counted

Coverage figures are usually expressed as a percentage of something. That something is where the ambiguity lives:

  • Nets covered — the net appears in the test program at all.
  • Nodes probed — a probe physically contacts the net.
  • Components covered — at least one test exercises the component.
  • Pins covered — individual device pins have a test.

A program can show 98% net coverage and still miss every solder joint under a BGA, because the net was tested at a via three centimetres from the joint. The number was never wrong. It was measuring the wrong thing.

The basic calculation

At its simplest:

Coverage % = (structures with at least one effective test) / (total structures) × 100

Worked example on a hypothetical board:

| Metric | Count | |---|---| | Total nets | 1,240 | | Nets with a probe | 1,180 | | Nets with no probe access | 60 | | Nets probed but tested only for shorts | 40 | | Effective structural coverage | (1,180 − 40) / 1,240 = 91.9% |

That last line matters. A net where you only check for a short to a neighbour is not tested for an open. Counting it as covered inflates the figure by roughly 3 points before you have even discussed BGA joints.

The framework that actually works: PCOLA-SOQ

Rather than counting nets, count the defect types you can detect. The widely used PCOLA-SOQ framework breaks structural coverage into categories:

| Category | Defect detected | |---|---| | Presence | Component is fitted | | Correctness | Right part value/type fitted | | Orientation | Part is rotated correctly | | Liveness | Part is functional (not internally open) | | Alignment | Part is positioned correctly on its pads | | Short | Unintended connection between nets | | Open | Intended connection is broken | | Quality | Solder joint or connection quality |

For each component on the board, ask which of these eight your program detects. The output is not a single percentage but a matrix — and the gaps in that matrix are your real exposure.

This is more work than counting nets. It is also the only version that predicts field failures.

Where the number hides its gaps

BGA and fine-pitch joints. As covered in our boundary scan comparison, a fixture cannot reach interior balls. Nets are still counted as covered because they are probed elsewhere on the board. This is the single largest source of inflated coverage on modern assemblies.

Polarised and multi-pin components. Orientation coverage for a two-pin capacitor is largely theoretical unless the test is designed specifically to detect it. Reversed electrolytics often pass basic analog tests.

Nets with no test points. These are sometimes simply excluded from the denominator rather than counted as zero. Always ask what was excluded.

Powered versus unpowered. A powered test can detect far more than an unpowered one, but requires the board to be safely powered in the fixture. Coverage claims frequently mix the two without saying so.

False coverage from adjacent probing. A probe on a net can mask an open if the test only measures to a nearby node rather than the far end of the net.

A practical method

  1. Get the netlist and the test-point report. These two files are the ground truth.
  2. Classify every net. Probed and tested, probed and short-only, not probed.
  3. Build the PCOLA-SOQ matrix per component. Even a rough version on the top twenty components by failure impact will change your decisions.
  4. Flag the inaccessible joints separately. Do not let them disappear into a net-coverage figure.
  5. State your denominator explicitly in any coverage claim you make to a customer or internally.

What to do with the gaps

Coverage gaps are not automatically failures — every program trades coverage against cost. The point is to know which trades you made and why.

  • Inaccessible BGA joints → add boundary scan where devices are compliant.
  • Orientation coverage on polarised parts → design a targeted test rather than assuming.
  • Nets with no probe access → consider adding test points at the next layout revision.
  • Analog nets outside practical test limits → accept, document, and cover by functional test.

The engineers who get this right do not chase a coverage number. They can name, without checking, which defect classes their program does not catch and why that was acceptable.

If you need a program built or a coverage assessment on an existing one, start with the ICT service providers in the directory.

Related in the directory: ICT program and software specialists, Keysight 3070 test houses, used and refurbished ICT equipment and the full ICT vendor directory.

test coveragePCOLA-SOQICT programDFTdefect coveragetest strategy