The splice loses 0.03 dB and passes testing. Then it breaks: the missing proof isn’t optical
6 min read
An underground closure gets reopened to splice in a new drop: the crew shifts the trays to make room, without touching the fibres fused three years earlier. One of those splices lets go on its own while the tray next to it is lifted — not pulled hard, not stressed beyond what a reopened closure normally sees, just handled the way a closure organiser always gets handled. The test record from back then, pulled from the as-built, reads 0.03 dB: an excellent splice, within the range a well-made fusion splice normally reaches and well inside any threshold ever written into a specification. That night, nobody had asked for a second number, and nobody thought they needed to.
Two properties, only one measured
The Recommendation that defines how an optical fibre splice is made does not talk only about loss. ITU-T L.400/L.12, Optical fibre splices, states it in its own Summary: “High quality in splicing is usually characterized by low splice loss and tensile strength near that of the fibre proof test level.” Two properties, named side by side, neither one standing in for the other. A specification that measures only the first half has verified half the definition — and the half it skips is the one that decides whether the splice is still there in three years, not whether it worked on the night it was made.
The editorial history is worth reading before citing it. The Recommendation began as ITU-T L.12 on 31 July 1992, reached a second edition still as L.12 on 12 May 2000, and only with the third — approved 8 March 2008 — took on the double designation L.400/L.12. The edition in force today, 4.0, was approved on 13 February 2022 by Study Group 15: every figure in this note comes from that text.
Cleave quality, listed twice
What decides whether a splice holds? Describing fusion splicing, the Recommendation lines up the variables in two separate sentences, not one: “The cleave quality and the intensity and the duration of the arc, as well as the differences between the fibres […] determine the splice loss. In addition, the quality of coating removal, fibre cleaving, and splice protection contribute to long-term mechanical reliability in the field.” Cleave quality appears in both lists: it can spoil the loss reading, or — separately — leave it untouched and compromise only the splice’s durability.
Clause 7.4 sets tolerances rarely seen written into a specification. For fusion splices, the cleave angle must stay “typically less than 1° from perpendicular for single fibres and less than 3° to 4° for ribbons”, depending on the fibre type. The cut surface “should be mirror-like without chips or hackle.” A dirty or poorly adjusted cleaver “can cause flaws that make the fibre break at the wrong location or reduce the strength of the completed splice” — without the optical alignment, and so the loss measured that same night, ever showing it. A splicing machine with active core alignment will still line up two cores well enough to read a good number, even when the glass it is aligning carries a hairline flaw from the cleave that has nothing to do with alignment at all.
The test the splicer runs on its own
There is a moment, in the procedure described at clause 7.5.1.3, when the splice just made is put under tension before it is even protected: “it is recommended to check its minimum strength […] the splice is subjected to a tensile proof test for a short period of time […] Splices that have strength below the proof test level will be re-done.” The machine does this on its own, automatically or on command, and typical values “range from 2 N to 8 N, depending on the type of equipment and desired strength.”
This is a test that already exists, on every single splice made in the field — not a sample, not a laboratory statistic: that splice, that night, in that pit. But it lives for a few seconds inside the machine’s logic, between the fusion and the heat-shrink sleeve, and the report that ends up in the as-built almost always carries a single figure: the estimated loss. If the splicer model named in the specification does not also export the proof-test outcome — a point to check machine by machine, which we do not assume here, since the Recommendation only requires the capability, not a particular report format — that second number does not survive past the evening on site. It existed for a few seconds, decided nothing was wrong, and then vanished with the electrode arc.
A hundred splices to validate a procedure, one to accept an installation
What a specification can still demand, short of the value of a single splice, is proof that the fibre-splicer-protector combination in use holds up over time. Clause 8.3 describes exactly this: “At least 100 splices should be made to check the requirements of test Nos 8.3.1 and 8.3.2 in Table 2” — made on the same fibre, to validate the procedure before it becomes the site standard. For fusion with active core alignment, Table 2 sets the number: average loss ≤0.05 dB, and no more than 0.1 dB in 97% of the hundred splices.
A second list, in Table 1, adds mechanical tests no specification ever asks for splice by splice: torsion under a 2 N load at −180° and +180°, ten cycles at 25 cm from the sleeve; fibre retention under 2 N on the primary and 5 N on the secondary coating, at 30 cm from the sleeve for sixty seconds; a bending moment of 2 N at the middle of the sleeve for ten seconds. These are modest loads — of the same order a closure takes when it is reopened for work nearby — and they are what proves whether a splice survives life in the field. But clause 8.2 calls for five test samples per test, not a hundred: these are two different regimes, and neither of them is the splice made last night in that pit.
The point
A splice that loses 0.03 dB has passed only one of the two tests the same Recommendation lists from its very first page. The quality that lasts for years depends on a properly made cleave and a verified tensile strength, not just a figure in decibels. It is the check we build in when we splice a run, and the clause we write — the declared splicer model, the tensile proof-test outcome required in writing, a validation certificate for the fibre-equipment-protector combination in use — into a specifications and compliance engagement that does not stop at the first figure the display shows. Every splice, its declared cleave and the outcome of its proof test feed into the same network map — not an isolated report per job — where an AI flags which runs have splices outside the documentation standard and one of our teams steps in, for an operator, a data centre, an industrial site, a public administration body, a healthcare facility or a defence site. Where an AI is needed on network data, it runs within the client’s own perimeter — on-premise or on dedicated cloud with a data centre in Italy — alongside CSIDIA, the group’s other company.
Are you writing the splicing clauses for a new specification, or do you need to reopen a closure tested years ago? Talk to an engineer: the site visit is at no cost, and we check together whether your splicer logs the tensile proof test before trusting the loss figure alone.