Terminals that back out: checking they are fully seated

Your continuity test passed because it only asks whether current can flow while the probe is touching. A terminal can sit short of its lock, touch the probe and pass, then move back when the connector is mated or shaken.¹ To prove a terminal is fully seated you have to load it or measure where it sits: push, click and pull each one as it goes in, close the secondary lock, and where a lost contact matters, put a push-back or step probe on every cavity at end of line.

A sample cannot do that job. At an illustrative 100 ppm per cavity, about 5 in every 1,000 harnesses of 50 cavities carry an unseated terminal, and checking 5 whole harnesses from that 1,000 has a 2.5% chance of finding even one.

Why the test passed

Cirris, a tester maker, describes a contact "sufficiently inserted to pass initial visual inspection and standard continuity testing, but which isn't fully seated".¹ The lock has not engaged, so the contact can be pushed back later, and the result is an open circuit, high resistance or an intermittent fault. The primary lock may be a lance on the housing (JAM), a lock finger (Molex) or a lance on the contact (TE).²³⁴ The figure shows the first.

The fixture can make it worse. Cirris warns that with heavy spring probes "pushed back pins may still be in contact with the spring probes" unless the fixture backs the probes off before measuring.¹ So a pass tells you about the conductor path at that moment and little about position. Continuity and hipot each have their own blind spots.

First work out which part moved

"Backed out" covers two faults, and one can cause the other. A terminal can leave its seat in the housing, or a housing can leave its mate. On some connectors an unlocked terminal stops the housing latching: Molex says the MX150 Unsealed "connector system will not latch if terminal is not locked".⁵ So do not assign an owner until you have looked.

UK recall text cannot tell you how often terminals back out. It is written at vehicle level, a recall count is not a defect rate, and a backed-out terminal could be reported as a poor connection. Ford R/2019/421 reports a pin defect that "may potentially cause an intermittent poor connection on a number of terminals", without saying how. A word search of the 863 recalls we read finds one that says terminals left their seat: Kawasaki R/2025/107 (39 Z e-1 and 36 Ninja e-1 on separate rows), where traction battery pack terminals "may become dislodged during plugging or unplugging" because of a manufacturing error.⁶

The record does show 15 of the 422 wiring recalls in our count where a connector or its lock may not have been secure, in vehicles already on the road. We read 10 as happening at fitting, but only 4 say so in words: Porsche, Volvo Trucks and Honda say assembly or assembled, and Audi says installed. Three say a lock might not meet specification and 2 give no stage.

Nine of the 15 name a lock, latch, retainer or clip. McLaren R/2022/060 (141 vehicles) has a driver airbag connection that may not be fully engaged, and the remedy checks resistance and then "that the secondary locking tab is correctly engaged". Mercedes-Benz Vans R/2021/100 has an additional lock on a seat belt tensioner connector that "may not be locked". In Mercedes-Benz R/2021/284, R/2021/286 and R/2022/033 the crash sensor connector's locking mechanism itself "might not meet specifications". The largest of the 15, Audi R/2023/254 (104,162 vehicles), names no lock: a connector feeding the interior fuse carrier "may have not been installed correctly".

So when a harness comes back, look before anyone unplugs it, because unplugging changes the evidence. We would photograph first: whether the connector position assurance (CPA) is closed, whether the housings are fully together, whether the terminal position assurance (TPA) is closed, and, once apart, whether any terminal sits further back than its neighbours when seen from the mating face.

Which check finds which fault

This table is our reading of the published documents and the recall text, and nobody here has run it on a harness. Try each row on your own connector. Build a master with one terminal deliberately short and one TPA left open, and run it through your check at the start of each shift. Check the short terminal's position before each use, rebuild it when it moves, and keep the master locked away from production.

How hard to pull: three of these documents give no number

The pull figures that do exist run from about 3 N to 31.1 N, on different connectors and methods. So a number from one connector tells you nothing about another, and "light" is not an instruction anyone can audit.

The pull must also be harder than the drag on a short terminal. MX150 Unsealed allows up to 12 N to insert a contact,⁵ so a 3 N tug on that connector might not move a short one (our inference, not a Molex statement). Ask the maker for a pull that moves an unlocked terminal and stays below the lance and crimp limits, then let operators feel it on a force gauge. Pull straight out from the back of the connector: NASA warns that above 5 degrees the contact may bind and give a false reading, and limits the load to 80% of the minimum crimp tensile requirement.⁸

During insertion, Molex says to hold the wire at least 25 mm from the crimp, to avoid rotating the terminal, and to stop at the click.³ JAM says unusual resistance may mean the terminal is tilted or deformed, so check it before pushing harder.²

NASA limits seating tests to one per insertion and requires the result to be recorded.⁸ A pull followed by a push-back probe exceeds that, so ask the maker whether repeated loads are allowed, and choose which test is the recorded one.

A light pull shows that something is holding the terminal, not that the lock would hold its rated load, such as Molex's 90 N minimum for MX150 Unsealed.⁵

What a secondary lock proves

The part has several names. Molex and Cirris say TPA. JAM and Sumitomo Wiring Systems say retainer.²⁹ Yazaki says front holder.¹⁰ Where it is designed to detect, the idea is the same: a short terminal stops it closing. Molex says the MX150 Unsealed TPA "will not seat into final locked position" if a terminal is not locked,⁵ and its guide tells the operator that resistance "may be detecting a partially installed terminal".³

A press can defeat this. The same Molex flyer gives TPA closure at 30 N maximum and closure with terminal detection at 65 N minimum.⁵ We read that as a window: a press set well above 65 N may drive the TPA past a short terminal. If you close TPAs with a press or pliers, find these two figures for your connector and set the limit between them.

Not every design detected reliably. A Yazaki patent describes an earlier arrangement in which the front holder "may be secured to the housing 3 with the metal terminal 11 half-inserted".¹⁰ Ask your supplier whether your connector's secondary lock detects a short terminal, or only blocks one that is already seated.

The part also has to be closed in the first place. Cirris calls TPA designs "100% effective" against pushback, then adds that the design "requires inspection for the presence of the TPA".¹ That is a supplier's claim, and the second half matters more. TPAs ship in pre-lock,³ so an open one looks normal in the bin. Molex says the TPA must be seated before any tape goes on the harness,³ which is a useful place to put the check.

The CPA has its own trap. Molex notes that a CPA can move to final lock during shipping and must be pulled back to pre-lock.³ So a closed CPA seen before mating proves nothing. Look for it in pre-lock before the housings go together, then look for it closed afterwards.

After any extraction, treat the cavity as suspect. JAM says not to reuse the housing, and if you must, to check the lance for position, cracks and whitening.² A damaged metal lance on a contact shows no whitening, so replace the contact. Molex warns that excess force on the service tool can damage the lock finger.³

Sample or every cavity: the arithmetic

We found no published rate for unseated terminals, so the rates here are illustrations. Your own can come from counting pull failures and TPA refusals. We assume each cavity fails independently.

Harnesses with at least one unseated terminal, per 1,000 harnesses. 1 - (1 - p)^n.

Now compare the options on one illustrative case: 1,000 harnesses of 50 cavities at 100 ppm, which is about 5 unseated terminals.

The audit and 10-cavity rows assume every check finds the fault. The 90% and 99% are assumptions, since we found no published figure for how often a pull finds an unseated terminal. The last row also assumes the two checks fail independently, which will not hold if one fault fools both.

NASA reaches the same answer by rule: each contact with clips or tines "shall be push or pull tested 100 percent for seating".⁸ That document is public domain and may not be the standard your customer specifies.

A sample still has a job. It finds a process that has gone wrong, such as a bad lot of housings or a method taught badly. For a 95% chance of finding one fault, you need 59 cavity checks if 5% of cavities are affected, 299 at 1%, and 29,956 at 100 ppm. That is 2 harnesses, 6 harnesses and 600 harnesses of 50 cavities. So use a sampling plan to audit the every-cavity check, not to replace it.

A clean sample also proves less than it seems to. At 95% confidence, and if every check finds the fault, 300 clean pull-backs only rule out rates above about 1%, and at 1% some 39% of 50-cavity harnesses would carry a fault. To rule out anything above 100 ppm takes 30,000 clean checks.

When a sample does fail, everything since the last good check is suspect, the same logic as a failed crimp pull. At 30 harnesses an hour and an 8-hour gap that is 240 harnesses and 12,000 cavities, or 10 hours of rechecks at an assumed 3 seconds each. That is 150 s per harness against a 120 s cycle, so a manual end-of-line check needs a second person or a fixture. On taped or TPA-closed stock you can push from the mating face. That can find a short terminal but not a weak lance, so a bad lot of housings needs its own containment.

Adding a push-back probe to the tester

Peak Test, a supplier, says push-back probes use "a very high spring force" and test whether terminals can withstand a certain pressure. It also lists step probes, which make no contact if a terminal is missing or improperly installed.⁷

Cirris compares two push-back builds. Heavy spring probes need a two-stage fixture that presses, then backs off before measuring. Switch probes cost more, but a pushed-back pin fails as an open circuit with no second stage.¹

Before you buy, check these.

• Fixture force. Cirris gives 3 lb, or 13.3 N, per contact, so a 12-way connector needs 160 N. Plan on a lever or pneumatics.
• Probe force against the lock. 13.3 N is well under Molex's 90 N for MX150 Unsealed, but that says nothing about a smaller connector. Get the lance-only figure for each one. Once the TPA is closed, a push loads lance and TPA together, so it cannot show a weak lance.
• Damage from the test itself. Molex warns that an oversized probe "will overstress the beam in the terminal" and create intermittent connections, and that pogo pins should be checked for damage or sticking several times a shift.³

For low volumes, a hand retention tool tests one contact at a time from the mating face. Cirris lists labour as the cost.¹ NASA asks for a tool that applies "a controlled, preset pressure".⁸

If your customer fits the harness, none of this reaches the 10 recalls we read as happening at fitting. Put the mating sequence on the drawing: latch click, a pull on the housings, CPA closed, and record it. A function test will usually find only a connector that is fully open. Then look at what moves the connector in service, because movement at the contact is a separate failure that a seating check cannot predict.

What to read in your licensed copy

We have not read these and give none of their values.

• IPC/WHMA-A-620. Cirris points to section 19.7.5 for push-click-pull on crimp-and-insert contacts.¹ Check the section number and class wording against your revision.
• SAE USCAR-2, for connector systems in low-voltage road vehicles, revised December 2024.¹¹ Look there for retention and lock test methods if your customer calls it up.
• The connector maker's product specification, which is usually free and is where most usable numbers live: lance-only and TPA-locked retention, TPA closing force, mating force, and the fixture interface drawing.

Notes

1. Cirris, Connector Contact Retention, 15 June 2023. Supplier guide. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

2. Japan Automatic Machine, Handling Manual, SN Connector, HM-0044-00, sections 5 to 7 and 9. Supplier document. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

3. Molex, MX150 Application Guide, AS-33472-100 revision D1, 2020, copy on the Italian site i2m.it. Sections 3 to 6, MX150 sealed range. Molex lists a later revision on its own site, which we could not open from our server. Supplier document. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

4. TE Connectivity, Instruction Sheet 408-7984, revision A, 2011, hosted by Farnell. Supplier document. ↩ ↩²

5. Molex, MX150 Unsealed Connector System, flyer hosted by DigiKey. Supplier document. ↩ ↩² ↩³ ↩⁴ ↩⁵

6. Driver and Vehicle Standards Agency, Check vehicle recalls, recall data file RecallsFile.csv, downloaded 6 November 2026. All recall numbers, vehicle counts and quotations here come from that file. Contains public sector information licensed under the Open Government Licence v3.0. Labels are by Assembly Wire, made with AI assistance and not checked by an engineer. ↩

7. Peak Test Services, Wire Harness Testing - Automotive. Supplier guide. ↩ ↩²

8. NASA, NASA-STD-8739.4A with Change 4, 2022, sections 13.8.2 to 13.8.6 and Table 13-2. Public domain. The force range in the table runs across contact sizes. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

9. Sumitomo Wiring Systems, US patent 5,664,966. ↩

10. Yazaki, US patent 5,769,664. A patent's background section describes a design problem and says nothing about your connector. ↩ ↩²

11. SAE International, USCAR2 publisher's page, scope and revision date only. ↩