Twenty-five acronyms, nine pieces

Why do dozens of competing frameworks for the same task turn out to be rearrangements of one short list?

Spend an afternoon reading about how to write a good prompt and you will meet RTF, APE, TAG, RACE, CARE, RISEN, CRISPE, CO-STAR, and a dozen more. Each arrives as a distinct method with its own diagram and its own advocates.

Read them side by side, though, and something odd happens: you stop learning. The fourth one teaches you nothing the first three did not. So the interesting question is not which framework is best. It is why a field generates twenty-five names for what appears to be one thing — and whether that is a failure of the field or a predictable consequence of how the thing is shaped.

Let us do the reading rather than assume the conclusion. RTF, as usually given, is Role, Task, Format. APE is Action, Purpose, Expectation. TAG is Task, Action, Goal. RACE is Role, Action, Context, Expectation. CARE is Context, Action, Result, Example. RISEN is Role, Instructions, Steps, End goal, Narrowing. CO-STAR is Context, Objective, Style, Tone, Audience, Response.

Now lay the expanded words out and look for the repeats. Role appears again and again. So does some word for the task — Task, Action, Objective, Instructions. So does context. So does a word for the shape of the output — Format, Response, Result. So does a word for what good looks like — Expectation, End goal, Purpose. And a handful of the frameworks add examples, constraints, or tone.

Collapse the synonyms and what is left is a short list of the things a specification can carry: who is speaking, what is to be done, what background is needed, what data is supplied, what examples show the shape, what constraints bound it, what form the output takes, what tone and audience it addresses, and how you would know it succeeded. Call that nine, while admitting that nine is a judgement about where to draw lines rather than a discovered law — someone who splits tone from audience, or folds constraints into the task, will honestly count differently.

Here is the step worth slowing down for. If those pieces really are the substrate, how many frameworks could exist? Choose three of the nine and put them in a memorable order and there are already 504 distinct possibilities. Allow four, five, or six pieces and the number runs into the hundreds of thousands. So twenty-five is not an embarrassment of riches. It is a rounding error against the space of available rearrangements — and if a field generates names faster than it generates ideas, that is precisely the pattern you should expect.

But why does the generation happen at all? Notice the incentives. A nine-item checklist is correct and nobody remembers it. An acronym is a memory aid, and a memory aid that spells a real English word — CARE, RACE, RISEN — spreads further than one that does not. Meanwhile, minting a framework costs nothing, requires no permission, and confers authorship in a way that saying "here is the standard checklist again" does not. Cheap to produce, rewarded for memorability, with no mechanism anywhere that consolidates duplicates: what other outcome was available?

Is this special to prompting? Ask the same question of goal-setting, where SMART, GROW, OKR and their cousins circle a similar handful of components; or of any young practice where the underlying moves are simple, the practitioners are many, and no standards body exists. Proliferation is the signature of a field whose substrate is a short list and whose currency is attention.

None of which makes the acronyms worthless, and it would be smug to conclude that it does. A checklist you cannot recall under pressure is not operative knowledge. What the acronym buys is retrieval — five letters you can hold while typing — and different work genuinely emphasises different pieces. A customer-service reply leans on tone and audience; a data-extraction task leans on format and examples. The frameworks that survive tend to be the ones whose selection matches a common kind of job.

The mistake, then, is not using one. It is believing the acronym is the knowledge. Someone who has learned CO-STAR has learned six slots in an order; someone who has learned the nine pieces has learned what can be specified at all, and can notice which one is missing when a result comes back wrong.

Three clarifications.

The first is about what "the same thing" means here. The frameworks are not identical: they select different subsets, order them differently, and that ordering carries a real claim about what matters most. That is a genuine difference in emphasis. It is just a much smaller difference than the branding implies, and it is a difference of selection from a common list rather than of underlying theory.

The second is a caution about the nine. Treat them as a working inventory, not a taxonomy handed down from anywhere. They were arrived at here by collapsing synonyms across a set of popular frameworks, which is an honest method but a soft one; a different reading list would yield eight or eleven. What is robust is the shape of the finding — a small substrate, a large name space — not the exact integer.

The third is the practical consequence. Once you can see the substrate, a new acronym costs you about ninety seconds: expand it, map each letter onto a piece you already know, and notice what it leaves out. If it introduces nothing new, you have lost nothing. If it introduces a genuinely tenth component, that is worth knowing — and now you have a way to tell.

• Which of the nine components measurably change a model's output, and which are folklore.
• How the same dynamic plays out in management frameworks and diet advice.
• Checklists versus mnemonics as instruments of recall under pressure.

Every term the collection defines is gathered in the glossary.