The atelier
A small three-person workshop studying what happens to a brand inside an AI answer.
i. What we are
The name risura carries two readings. In Latin, rasura was the trace left when a scribe scraped parchment to write over it — an old way of erasing, more careful than cutting. In Portuguese, riso is the smile. The atelier observes rasuras: the points where the model, in language, quietly scrapes a brand’s entity to write over it another function, another category, another neighbour. It looks at them with the composure of the scribe who recognises the mechanics of the gesture, not with alarm.
Atelier das Entidades is a small lab in Lisbon working on a single subject: how language models describe companies — where they get it right, where they slightly shift the category, where they quietly replace the product’s real function with a neighbouring one.
We are not an agency or a consultancy; nor are we an academic observatory. We work as an atelier: three people, one methodology, a shared notebook. Every observation is recorded before it is interpreted; every interpretation is discussed among the three of us before it enters the corpus.
ii. How we work
The work begins with a question put to a model. Always in pt-PT, always in an anonymous session, always repeated. What results is a collection of answers — some identical, others with small underlying variations. Within that collection, the lab separates the noise of variations that do not persist from the stable shift of displacements that recur across runs and across models.
The taxonomy we use — four modes of error — was not proposed a priori; it grew out of the observations. ▲ displacement · ↳ replacement · ~ neighbour · — empty space. Those four labels currently cover almost all of the plates in the corpus.
iii. What we do not do
The atelier does not do SEO audits, prompt optimisation, or model training. We do not sell advisory reports. We publish plates — each one signed, with a stated method, with the source answer on record, and reproducible by anyone who wants to repeat the protocol.
Masthead · three people
Leads the analysis of entities in model answers. Signs most of the plates in the corpus.
Looks after the run protocols, the repetition of samples, and the record of conditions for each test.
Maintains the observation notebook, traces the semantic trajectories, and reviews the technical Portuguese of each plate.
The atelier publishes no calendar and no newsletter; new plates appear in the corpus as the work advances.