First pass yield: measure it so the number means something

Count a unit as a pass only if it went through the step once with nothing repaired, retested or touched again after the operator declared it done, and divide by every unit that entered that step for the first time. Reported that way, first pass yield (FPY) shows where you are paying for rework, which final yield hides. Compare harnesses by defects per opportunity and not by FPY: a 500-crimp harness at 100 ppm passes 95.12% of the time and a 20-crimp harness at 500 ppm passes 99.00%, yet the first has crimps five times better.

The rules to write down before anyone counts

The words are not settled, so your report has to carry its own definition. Wikipedia opens by calling first pass yield "throughput yield" and defining it as units out over units in, which would include rework, then its worked example takes reworked units out.¹ iSixSigma uses "total throughput yield" for the figure after rework.² So a supplier's "98% yield" can mean either, and you have to ask which. Fix four things in writing.

1. Define the unit: one finished harness by serial number, one sub-assembly or one lead. A batch is not a unit, and a batch of 50 with one bad harness has not failed by 100%. Cutting, stripping and crimping make leads, not harnesses, so count those steps per lead and convert before you multiply, keeping the unit the same throughout.³ With 200 leads at 99.99% first pass, the harness-level figure is 98.0%.
2. Say where each step ends. An operator who re-pins a wire before handing the harness on is either correcting within the step or doing rework, and two shifts must make the same call. A workable rule is that anything found after the operator has declared the unit done counts as a fail of that step.
3. Let the first result stand. A harness that fails, is reseated and then passes is a first-time fail with the cause "no fault found". It becomes a false fail only when the fixture is shown to be at fault, for example when the failure follows the fixture to a known good harness and does not follow the harness to another fixture.⁴ Until then it stays in the defect count, because a contact not fully seated or a stray strand can pass on the second try.⁵⁶
4. Count each unit once per step. Defects are counted separately, because one harness can carry three.

What a crimp rate does to a harness

A harness passes only if every crimp is good, so yield is (1 − rate)^n. These figures give the share of harnesses with no bad crimp, assuming independent defects.

Read it backwards to set a target. For 99% on a 200-crimp harness the crimps have to run at 50.3 ppm, and on a 500-crimp harness at 20.1 ppm. Crimp height and pull checks alone will not show rates that low, as the sampling article works out.

Real defects cluster, because a bad setup or reel makes a run of them. At an average of 100 ppm on 200 crimps, independent defects give 98.02%, but if all of them come from 1% of harnesses built at 10,000 ppm the yield is 99.13%. So do not convert FPY back to a process ppm without looking at whether failures bunch by reel, tool or shift. A bunch points to a setup problem, and an even spread points to the process itself.

Rolled yield across the line

Rolled throughput yield (RTY) is the product of the first pass yields of every step, and rework and scrap do not enter into it.² This flow is an illustration with invented round numbers, with the two lead steps already converted to harness level.

Every step looks fine, and 78 harnesses in 1,000 still get touched again. The arithmetic also suggests a priority: raising electrical test from 97% to 99% lifts RTY to 94.14%, raising crimp from 98% to 99.5% lifts it to 93.65%, and perfecting cut and strip lifts it only to 92.70%.

Be careful with that first figure. Electrical test creates almost none of the defects it finds. A wrong pin or a bad crimp was made earlier, so "improve the test step" means tracing each failure to the step that made it. That is why the log at the end records origin as well as where the fault was found.

Why rework must count as a fail

This illustration takes 1,000 harnesses at electrical test: 970 pass first time and 30 fail. Of those 30, 10 are traced to the fixture and 20 are real, and 18 of the real ones are repaired and 2 are scrapped.

Final yield of 99.8% is true and tells you that the rework bench works. It says nothing about 30 harnesses, each pulled out, diagnosed and retested. RTY exists to show exactly this gap: a high throughput figure next to a low rolled one means a lot of rework.²

Report the false fails as well. They are not product defects, but they eat the same test time, and DIT-MCO, a tester supplier, points to worn adapter cables and contaminated fixtures as common causes on hipot.⁴ Without their own line in the log, a worn fixture looks like a process problem.

Comparing harnesses: use defects per opportunity

FPY punishes complexity. Convert it with defects per unit, DPU = −ln(FPY), which assumes independent defects (a Poisson model), then divide by the opportunities in the unit to get defects per million opportunities (DPMO).

On this count B looks the worst, provided the test sees the same share of faults on each harness and the volumes are large enough to trust the FPY. An opportunity on a fine wire and one on a heavy lug are not equally hard, so keep crimps, insertions and routings on separate lines if you can. B's rate on a harness the size of A would give 77.7%. If you log actual defects, the count form is simpler: 52 defects on 1,000 units with 500 opportunities each is a DPU of 0.052 and a DPMO of 104.

The weak point is the opportunity count. Count the same harness as 1,500 opportunities, with every wire end, label and tie included, and the same 52 defects become a DPMO of 35. Nothing got better. Choose one counting rule (for example, one per crimp, one per insertion, one per wire routed) and hold it fixed, or your comparison across products means nothing.

When a change in the number is real

A weekly FPY on low volume carries a lot of noise. These are 95% Wilson intervals for an observed 97%.

To tell 97% from 98% with the usual 5% significance and 80% power takes about 3,800 units in each period, and 95% from 97% takes about 1,500. So plot it on a p-chart.⁷ At 97% and 200 units a week the three-sigma limit sits 3.6 points below the average, at 93.4%. Investigate a week below that line, or a run of points on one side of the average, and ignore a one-point wobble inside it. Read the cause codes too, where a new failure type shows up in single counts.

FPY also moves when the test changes. Suppose the true defect rate is 0.10 per unit, a true yield of 90.5%. A test that sees 30% of those defects reports 97.0%, and one that sees 60% reports 94.2%. Add a hipot or a stricter limit and FPY falls while quality stands still, so log test changes alongside the number. A low figure can mean a good test, because each test finds different faults.

What a point of FPY is worth

Cost of rework = repairs × (diagnosis, repair and retest time × labour rate + materials) + scrap × (materials + labour already in the unit). Here it is with invented round numbers: £30 an hour, 1,000 harnesses a week, 48 weeks a year.

The customer line counts only labour and parts, and is the same £50 used in the sampling plan example. Add your own figure for sorting, line stoppage and chargebacks, which can far exceed it.

One point of FPY at test is 10 repairs a week: £145 a week, £6,960 a year and 4.2 hours of labour a week. Two scrapped units a week add £140, at £40 of materials and an hour of labour already built in to each. The model leaves out the delay when the test station is the bottleneck, which may matter more than the labour.

That gives the rule for where to spend effort. Rank causes by defects × cost at the point of detection. The lowest-yield step is not automatically first. A 2% loss at crimp found at test costs nearly ten times a 2% loss found at the crimp bench (£14.50 against £1.50), so better detection close to the source can pay even when yield does not move. Put your own figures in. For crimps, the bench-side control is crimp height.

One-page log

Print one sheet per part number per week. The table counts first-time units only.

PART NO. ______  REV ___  WEEK ____  LINE ____
UNIT: lead steps = one lead. Other steps = one harness, by serial.  LEADS PER HARNESS = ____
OPPORTUNITIES PER HARNESS = ____  (crimps ___ + insertions ___ + routings ___ )  RULE VERSION ___
TEST PROGRAM / LIMITS VERSION ____   CHANGED THIS WEEK? Y / N  WHAT: ________

STEP (UNIT)          |ENTERED |PASSED  |FAILED|OF WHICH |OF WHICH|REWORKED|SCRAP-|REWORK|MATERIALS|SCRAP
                     |1ST TIME|1ST TIME|      |NO FAULT |FIXTURE |OK      |PED   |MINS  |COST     |COST
---------------------+--------+--------+------+---------+--------+--------+------+------+---------+-----
Cut and strip (lead) |        |        |      |         |        |        |      |      |         |
Crimp (lead)         |        |        |      |         |        |        |      |      |         |
Insert (harness)     |        |        |      |         |        |        |      |      |         |
Lay up, tape (harn.) |        |        |      |         |        |        |      |      |         |
Electrical (harness) |        |        |      |         |        |        |      |      |         |
Final visual (harn.) |        |        |      |         |        |        |      |      |         |

DEFECT TALLY (one mark per defect, not per unit)
CODE | DEFECT                      | MADE AT (STEP) | FOUND AT (STEP) | COUNT | REEL / TOOL / SHIFT
C1   | Crimp height out, bad crimp |                |                 |       |
P1   | Wrong pin or cavity         |                |                 |       |
S1   | Terminal not seated         |                |                 |       |
I1   | Insulation or strand damage |                |                 |       |
L1   | Length or label wrong       |                |                 |       |
N1   | No fault found on retest    |                |                 |       |
F1   | Fixture shown to be at fault|                |                 |       |
X1   | Other: ________             |                |                 |       |

WEEKLY SUMS
Step FPY          = passed first time / entered first time
Lead step, harness level = lead FPY ^ leads per harness
RTY               = product of harness-level FPYs
Final yield       = (harnesses started - scrapped) / harnesses started
DPU               = defects (including N1, excluding F1) / harnesses built
DPMO              = DPU / opportunities per harness x 1,000,000
Rework cost       = rework mins / 60 x rate + materials + scrap cost
Rule check: a retest pass is still a first-time fail. A batch is not a unit.

Notes

1. Wikipedia, First-pass yield. Accessed November 2026. Its definition cites ASQ and others, which were not opened. ↩

2. iSixSigma, Rolled throughput yield. Trade site. ↩ ↩² ↩³

3. Six Sigma Material, Throughput yield. Training site. ↩

4. DIT-MCO, How to fix inconsistent hipot test results in 7 steps. Supplier guide. ↩ ↩²

5. Cirris, Connector contact retention. Supplier guide. ↩

6. Wiring Harness News, Electrically checking crimps. Trade article built around a CableEye tester. ↩

7. NIST/SEMATECH e-Handbook of Statistical Methods, Proportions control charts. ↩