The pass that fails
A field crew signs off a 24-strand FTTX build at 2pm on a Friday. The OTDR trace is clean. The link budget is green. The customer portal says the service is live. Six weeks later, the subscriber at the end of that run opens a ticket: internet keeps dropping every evening. You roll a truck. It passes. You test a third time. Pass. By now the customer has stopped believing your test data — and honestly, so have I.
I'm a quality and compliance manager at a mid-sized US fiber operator. My job is to review field acceptance reports before they get signed off — roughly 400 of them a year across our FTTH and metro builds. In 2024 I sent back about 14% of them. Not because the instruments were wrong. Because the test was.
If you came to this article looking for a technical walkthrough, this isn't that. This is the version I wish someone had handed me three years ago, before I'd spent a full quarter chasing loops that kept passing.
Why "pass" and "works" are not the same thing
The obvious explanation for a re-dispatch is a bad splice, a dirty connector, or a kinked drop. Those are real. But they're also the answer most teams stop at, which is why they keep happening.
Here's what I actually found after pulling 18 months of passed-then-failed tickets:
The pulse width nobody checked
Most crews run acceptance testing at a short pulse width because it gets a sharper event signature. Sharper is fine for locating faults. It is not the same as characterizing the whole span. A 10-nanosecond pulse can miss a bend that's barely hovering at the edge of the attenuation budget — the kind of thing that shows up after a hot week, when the fiber expands and the angle shifts by a fraction of a degree. The trace passes on a cool Tuesday morning and fails on a warm Thursday evening. Nobody's lying. The instrument just wasn't asked the right question.
The one-way problem
This was the one that genuinely changed my protocol. The majority of our passes were single-direction. We tested A→B, recorded the loss, moved on. But attenuation is not symmetric. If a splice is offset slightly, one direction can read 0.4 dB and the other can read 1.6 dB. Both numbers go into the same report and only one of them matches the customer's experience. Bidirectional testing roughly doubled our per-span test time, and I still fought for it, because the recurrence rate on passed-then-failed tickets dropped by more than half once we made it mandatory.
The acceptance mindset
This is the part I have mixed feelings about. The crews are not lazy. They are being asked to do acceptance testing under a production schedule that assumes acceptance is diagnosis. It isn't. Acceptance asks "does this link meet spec today?" Diagnosis asks "will this link keep meeting spec under load, over temperature, and after a customer plugs in a $30 set-top box?" Those are different test sequences, different pulse widths, and honestly different frame of mind. The industry talks about them like they're the same task with the same tool. They aren't.
I have mixed feelings about how hard we pushed bidirectionality. On one hand, it added 10-14 minutes per span and the field group hated me for a quarter. On the other, the re-dispatch rate across our Q3-Q4 2024 builds dropped from 9.1% to 3.7%. I'd take that trade every time. It just took longer to argue for than it took to implement.
What a bad pass actually costs
Nobody budgets for the second truck. That's the core problem with the whole system — the first test is a line item, and the re-dispatch is a surprise.
Rough numbers from our own operation: a single re-dispatch to a suburban residential link runs about $410 fully loaded (crew time, vehicle, scheduling, the inevitable callback). Splice rework on a damaged state adds another $180-$260 depending on the enclosure. If the ticket escalates to a commercial SLA, the credit alone can exceed both. And that's before we talk about what the customer tells their neighbor.
The number that actually got budget approval, though, wasn't the direct cost. It was the churn rate on FTTH subscribers who had opened two or more trouble tickets in their first 90 days. We don't publish that figure externally, but internally it was more than 3x the baseline. That's not a test problem. That's a brand problem wearing a test problem's clothes.
Here's the point I keep making to anyone who'll listen: the damage from a bad pass isn't that the link failed. It's that the link failed after you told the customer it passed. That's a credibility event, not a network event. Nobody gets fired for a pass. But somebody absolutely gets fired for the third pass that didn't hold.
The fix is boring, which is why it's hard to sell
Nothing I'm about to recommend is exotic. It's just slower, and slower is a hard pitch to make to a production manager who's paid per span.
Make bidirectionality standard on any span over 500 meters. Not optional. Not "recommended." Standard. The asymmetric-loss cases are exactly the ones that pass quietly and fail loudly.
Match the pulse width to the question being asked. Short pulse for locating events. Wide pulse for link loss budget verification. Two passes, one report. If your test platform can't sequence both automatically, that's a specification problem, and it's the reason a lot of us end up standardizing on something like the OTDR EXFO FTB 400 — not because it's magic, but because the test sequence is repeatable across crews without leaving half the protocol up to the individual tech's judgment at 4:30pm on a Friday.
Record return loss, not just insertion loss. A connector can pass bend and pass attenuation and still have a return-loss figure that would make your head hurt — a clean mating sleeve can read a return loss approaching infinity, while the one right behind it drops 18 dB without blinking. Insertion loss tells you what the signal loses. Return loss tells you how much noise is bouncing back up the link. On a 4K stream, that second number is the one the customer experiences as freezing.
Define "pass" against the customer experience, not just the standard. ITU-T G.652 and Telcordia GR-326 give you the floor. The floor is not the job. Our internal acceptance numbers run tighter than the published minimums on residential drops, and we still argue about whether they're tight enough.
And one equipment note worth saying out loud: if you're searching for the best multimeter for electronics because a technician swears the fault is electrical, you're probably troubleshooting the wrong layer. Almost every "random drop" I've chased in the last two years has been an optical-layer event. The multimeter is a good tool aimed at the wrong problem.
My experience on all of this is based on mid-sized US operators running mixed FTTH and metro builds, and mostly on single-mode outside plant. If you're working with long-haul DWDM or submarine segments, the tolerances and the economics change a lot — I can't speak to those the way I can to residential last-mile, and I won't pretend otherwise.
The satisfaction, if there is one, comes from watching the re-dispatch number fall quarter over quarter and knowing exactly why. There's no dashboard for the ticket that never got opened. But I know roughly how many there were, and that's enough.
Bottom line: the OTDR is not wrong. It's answering the question you configured it to answer. The gap between that question and what the customer actually feels is where every second truck comes from — and it's the only part of the process most teams still haven't measured.