Where the ideas come from

A situated catalogue, not a universal map

This English-language edition is adapted from a guided journey created through Brazilian research with adults in higher education. That is an origin and a limit, not a definition of who may use the app. Translating words alone would not carry the experience across contexts: examples, ways of taking part, references, voice, privacy, and support must also be adapted and tested.

The selection is uneven too: several European traditions appear through individual authors, while Ubuntu appears as a collective reference. Naming that asymmetry is more honest than presenting eight names as a complete canon. The lenses organize questions in the journeys; they neither represent every tradition nor classify the people who respond.

Who pays at this stage

At this stage, participants do not pay to use the product. The company covers the operating costs. This describes the present and makes no promise about future pricing.

What the response never does

When you select part of what you wrote and ask for a related passage, the response is written by a machine from the passages it found. The rules below are not a statement of good intentions: each is checked in code before the response reaches your screen. A response that breaks any of them is rejected; instead, you receive a continuation prompt written by people.

• Nothing in the response makes its reader the subject. It does not state your condition, conclude what is happening inside you, or return a question with a diagnosis already built in.

  How to check by reading the response: look for the word “you” followed by a state verb — “you are”, “you feel”, “you need” — and for questions beginning with “could it be that you”. None may appear. “You wrote”, followed by your words in quotation marks, is the opposite: it is a quotation and is allowed.

• What you write does not become evidence about the person who wrote it. Your text returns in quotation marks, as your words, never as a symptom, pattern, or defence.

  How to check by reading the response: see whose inner life the sentence is describing. If “your” is attached to trauma, fear, pattern, process, or defence, the rule was broken — the same applies to “this means” and “deep down”.

• The machine does not speak about itself or tell anyone to notice something.

  How to check by reading the response: look for “I realize that”, “I see that”, “notice that”, and “observe how”. None may appear.

• Every passage comes with its source marked, and the first source appears in the first sentence.

  How to check by reading the response: count the numbers in brackets — [1], [2] — and check that the first one is in the first sentence, not at the end.

• The response ends with a question.

  How to check by reading the response: Read the last line. It ends with “?”.

• No sentence comments on the search, the collection, or the machine's own limitations.

  How to check by reading the response: look for sentences about the apparatus: “the corpus does not cover”, “the passages do not say”, “I cannot decide”, and “it is worth noting”. None may appear.

The same standard applies to text written by people. It covers the response written by the machine, the notebook task shown when no response arrives, and the label that says where the text came from — not only the machine's output. A standard that applied only to the machine would leave the product unchecked precisely when no response arrived, which is when the person writing is most exposed.

The checks happen at different times, and that difference matters: a machine response is rejected immediately, one by one; text written by people is checked before it enters the product, and a sentence that breaks the standard prevents publication.

Where the guarantee ends. These editorial intentions for the machine response have no automated check today: the length of the response — the editorial notebook asks for two to four short paragraphs, and no check counts paragraphs, whether the final question uses the words you wrote — the check requires a question mark, not whose words the question uses, whether each sentence has an author, a work, or something in the world as its subject — the check looks for forms in which the machine speaks for itself and does not analyze grammatical subjects, and source order after the first sentence — the check stops there because enforcing it everywhere would reject a continuation of a paragraph whose source has already been attributed. They are written here for what they are — editorial intent, not a guarantee. Listing them with the guarantees above would ask you to believe that the text and the code will never diverge.