9 Why the documents read the way they do

9.1 Why a standard is written in a strange dialect

3GPP prose is not written to be enjoyed. It is written so that two lawyers, two engineers and a test house reading the same sentence reach the same conclusion about what a conforming product does.

That is harder than it sounds, because a standard has no force of its own. The obligation comes from somewhere else — a law that points at the document, or a contract that says a product conforms to it. So the document's only job is to be unambiguous about which of its own sentences are requirements ^[1].

9.2 Where the rules actually live

Not in TR 21.900. The working methods point elsewhere twice: for how a specification is drafted and which tools may be used §4.1.1, and again in the list of management tools, where the permitted word processors are the ones named in the drafting rules §7.5.

The drafting rules are their own document TR 21.801 v19.1.0, and the shared vocabulary is another one again, TR 21.905 v19.2.0, holding 579 defined terms ^[3]. Both appear in TR 21.900's reference list as the first two entries, before anything technical §1A.

The IETF has the same split. Its style guide is a document of its own RFC 7322 — the counterpart to 3GPP's drafting rules, doing the same job for a very differently shaped series.

9.3 The two verbs

One set of verbal forms marks a requirement.

Another set marks a recommendation.

In practice that means "shall" for the first ^[1] and "should" for the second ^[2] — and no "must" anywhere, so that nobody reads a standard as a law.

The third verb is "may", which grants permission rather than imposing anything. One paragraph of TR 21.900 uses all three, and is worth reading as a specimen.

Three sentences, three different strengths. The Support Team is permitted to make the fix. If it does, incrementing only the third digit is not optional. And holding the fix over is preferred but not required.

Once you are reading for verbs, whole clauses change shape. Compare two lines about change request numbers: numbers "shall not be re-used" §4.6.1, but the plenary "should approve, reject or postpone a CR in its entirety" §4.6.1. The first is a rule a system enforces. The second is advice a meeting can depart from, and does.

9.4 How the documents handle their own history

Two habits make 3GPP text look odd to anyone used to ordinary prose, and both exist for the same reason: numbers are addresses, and an address that moves breaks every reference pointing at it.

Emptied numbers stay empty. TR 21.900 has a clause 4.5 whose entire content is "(void)" §4.5, and a clause 4.9B.1 that says the same §4.9B.1. Table 5 is marked void where it sits §4.6.4. Change request category E is listed in the categories table as "not used" §4.6.2. In none of these cases has the number been given to something else. The same principle governs change request numbers, which are never reused even after a rejection §4.6.1.

Terms are defined before they are used. Clause 2 is a glossary, and it opens by saying that the terms defined there and those in the vocabulary document both apply §2. It is where "frozen", "closed", "draft", "feature", "building block", "work task" and a dozen others get their precise meanings — and where you should look first when a word in a later clause seems to be carrying more weight than usual.

9.5 Reading a NOTE

Alongside the numbered clauses, 3GPP documents carry paragraphs marked NOTE, and they do a different job from the text around them. In TR 21.900 they explain, warn or name things rather than impose:

• the note in clause 4.1.1 explains why a document jumps from 2.0.0 to 4.0.0, and that no 3.y.z will ever exist §4.1.1;

• the note in clause 4.6.1 records that "rejected" became unacceptable in some groups and was replaced by "not pursued" §4.6.1;

• the note in clause 4.7 gives the nickname FASMO to an essential correction §4.7;

• the note in clause 6.2 says the order of work item codes on a cover sheet carries no meaning §6.2.

Read them. They are where the document explains itself, and each of those four answers a question that comes up constantly.

The same goes for the paragraphs marked EXAMPLE, which are how the document shows rather than tells — six of them in the file-naming clause alone §5A, and a long worked one about work item codes in clause 6.2 §6.2.

9.6 Definitions borrowed and definitions local

A 3GPP document does not redefine what another 3GPP document has already defined. TR 21.900's definitions clause explicitly says the terms of the vocabulary document apply alongside its own §2, TR 21.905 v19.2.0.

That is what makes a term like "Change Request" mean the same thing in a security document and a charging document. It also means a definition you cannot find locally is probably not missing — it is one level up, in the vocabulary.

9.7 Where this leads

The IETF solved the same problem with a different instrument: three capitalised words, defined once in one short document and reused by every other. That is MUST, SHOULD and MAY — three words, one short document, and the comparison between the two approaches is in The two cultures, side by side and on the evidence.