Why does scoring an unfamiliar language on each axis beat labelling it with a single paradigm name?

## What a label actually is A paradigm name is a **cluster**. Historically, certain axis positions arrived together: languages built around values that do not change also tended to leave ordering to the evaluator, to defer work until demanded, and to make small pieces of behaviour freely passable. Enough languages shared that bundle that the bundle got a name. The name is real - it describes a genuine correlation - but it is a summary of positions, not a measurement of any one of them, and a language may sit anywhere on each axis independently. That is why the label question fails on exactly the language you most need to understand: the in-house one that came with an acquisition, written by people who had never argued about paradigm names and simply built what the problem needed. ## The worksheet The useful exercise has four rows, and each row answers a question you will actually have to answer: 1. **State** - can a routine change a value its caller still holds? This decides how many places you must read to explain a wrong value, and whether shared values are safe to hand out. 2. **Control versus data flow** - does the text fix the order, or only the dependencies? This decides whether the evaluator may reorder, batch or skip work, and whether cost can be read off the page. 3. **Evaluation** - does writing an expression do the work, or record how to do it? This decides where cost and failures surface. 4. **Composition and reuse** - what can be named, passed and joined? This decides the granularity of every reuse decision and what the libraries look like. Each row is scored from a sample and each score is **falsifiable**: someone can show you code that contradicts it. A label cannot be contradicted, because it asserts nothing checkable. ## The hybrid that breaks the label Real languages split the cluster constantly: - Immutable-leaning by default, yet every effect strictly ordered by the text - values are safe to share, but nothing may be reordered. - Mutable-leaning, yet with a whole-collection transformation style that leaves the evaluator free to schedule the elements. - Small passable units of behaviour, yet eager throughout - composition is cheap, but assembling a pipeline does the work immediately. - Deferred evaluation with ordered effects - the hardest combination to reason about, and one that no single label warns you about. Every one of these is a language somebody maintains, and every one of them is mis-served by a single name. The engineer who writes "functional" on the worksheet and stops has recorded a guess about four things after observing none of them. ## What the score buys you | Axis score | What ports from your existing idioms | What needs rework | |---|---|---| | Mutable, text-ordered | step-by-step routines, in-place updates | anything assuming a value cannot change underneath it | | Immutable, dependency-ordered | transformations, value-returning helpers | code whose correctness depends on effect order | | Eager | cost reasoning read from the page | idioms that rely on defining more than you consume | | Small passable unit | wrappers, substitution, run-time choice | patterns that expect a large configurable construct | The migration estimate you can defend comes from this table, not from a name. "Two of our four core idioms port unchanged; the third depends on effect ordering this language does not guarantee; the fourth assumes behaviour can be passed and it cannot" is a claim a reviewer can argue with and a plan can be built on. ## When a label is still worth using Labels are not useless and pretending so is its own error. Among people who already share the same reference points, a name compresses four rows into one word and is a fine shorthand for a preference or a lineage. The failures are specific and predictable: - Treating the name as though it **implied** a position on every axis. - Assuming a hybrid must sit in the middle of each axis, when in practice hybrids sit at an extreme on some and the opposite extreme on others. - Taking the language's own self-description as a measurement rather than as marketing or as lineage. - Arguing about which label applies, which is a debate about a word, when the disagreement that matters is about a row on the worksheet. The discipline is simple: say the name if it helps the conversation start, then score the four axes from the sample, and let the scores - not the name - be what goes into the plan.