By example
Show it. The library finds it, proves it, and keeps your example as the test.
The ground is already there
Softanza does not have to invent the example. The house writes its tests as a call with its answer under an arrow, and a court compares that answer with what really printed. That form is the library's one unit of promise, and it is already everywhere.
The ground: an example in the library's own test form, a call and its promised answer, which a court compares with what really printed.
# an example as the house writes one: the call, and under it the answer it promises ? @@( Q([ 3, 1, 3, 2 ]).DuplicatesRemoved() ) #--> [ 3, 1, 2 ]
[ 3, 1, 2 ]
run on 2026-10-08 21:47 inside the library at commit 4a184e6f0; taken from base/doc/narrations/stz-functions-as-linguistic-expressions.md(opens in a new tab)
More than nine thousand such promises sit in the test tree, the articles and the quick-start files, counted as a text marker at library commit 93d7a39, which is a count of promises written and not of tests passing. Of the distinct, non-trivial answers among the ones that pair a call with its result on one line, 276 of 299 are promised by exactly one method. The limit of the idea is coverage, 193 methods among thousands, and not ambiguity.
Three levels
Finding (every piece built, the verb not yet)
Pairs go in: "hello" to "HELLO", [3, 1, 3] to [3, 1]. Candidate methods come from the shape of the pairs set against the recorded promises and the forms grammar; each is verified by running it in a fresh process. No induction is needed.
Inducing (the agent is declared and reserved)
Examples that describe a shape, such as +216 21 298 374, induce a pattern of the pattern-language family. When two candidates remain, the conversation asks for the example that tells them apart. The library says of itself that this is the part not yet built.
Synthesising (designed)
A chain of two or three verbs found by a bounded search over the Q-chain grammar, judged by running, and narrated step by step the way the mental model narrates.
What exists today
Today the library finds a method from the words of a need, with no AI model. Finding it from an example of the need is the next step, and it is not built.
What exists today: the library finds a method from the words of a need, with no model. Finding it from an example of the need is the next step, not built.
# found today from the WORDS of the need, not yet from an example of it
oDoc = StzSelfDocQ("stzList")
? oDoc.HowTo("remove duplicates from a list")Q([...]).RemoveDuplicates() -- composed by grammar (remove duplicates)
run on 2026-10-08 21:47 inside the library at commit 4a184e6f0; taken from base/education/program/courses/elementary-introduction/chapters/13-the-gap-question.en.md(opens in a new tab)
The three outcomes
An answer by example is one of five registers in which the library accepts an answer, and it follows the same court as the others: the candidate is run and judged against the world's rules before it becomes knowledge. If one reading fits, it is accepted and narrated. If several fit, they are listed and the person chooses. If none fits, it is refused with the reasons and the nearest alternatives. The protocol is told on the wise coding page, and there is no fourth outcome.
What you show is what you get, and what you get is proved
designed An admitted candidate writes its example back as a promise, a call with its answer under an arrow, into the caller's file and the project's guard, and the same court judges it like any other. Specification, test and documentation become one line. This is a rule of the design and it is not built.
What it is not
Programming by example is an old idea. What this one adds is how a candidate is judged: by running it against a vocabulary that documents itself, and by a court that can refuse. What is found is a closed pattern or a chain of named verbs, never free code, and an AI model, if one is used at all, is a generator of candidates inside the protocol and the last in line.
The costs are named. Several candidates is the normal case, which is why the second outcome exists. Three examples can fit too well, and the remedy is the question that tells two candidates apart. The index is as clean as the tags it reads.
in construction Nothing induces yet. The three levels above carry their stage in their titles; the rest of this page is what exists.
Sources: the promise form is explained in the articles functions as linguistic expressions and the mental model; the answer protocol is in SOFTANZA_INTELLIGENCE_ARCHITECTURE.md(opens in a new tab). The counts were made by reading files, not by running them.