Sprint developer

The Sweep button rereads this company's conversations and ratings and creates the new suggestions — always as PENDING, never approved. Running it twice doesn't duplicate anything, and what you already discarded doesn't show up again: insisting on what the person refused is how a curation queue loses the trust of those who use it.

Every suggestion on this screen waits for a human. This is not generic caution: self-learning without review is exactly how hallucination comes back. A single 👍 given absent-mindedly to a wrong answer would become a "verified answer" — and that question would start being routed to the wrong tool ALWAYS, with the authority of "verified" and no AI along the way to hesitate. The error would stop being occasional and become policy, born from an innocent click. Approving is a human act, and it is recorded with your name and the time.

On approval, the question starts going DIRECTLY to the chosen tool, with no AI routing — it answers in under 1 second with no possible hallucination along the way. What is stored is WHICH tool answers, never the answer itself: a frozen answer ages (yesterday's cash is not today's), a verified tool gives the fresh number from the database.

The system does not store which tool answered each question. So the tool that appears in the suggestion was deduced by word overlap between the question and the catalog: it is mechanical and reproducible, but it is a guess. By approving, you are confirming that it really is the one that answers — and that is why the decision is yours and not the generator's.

When someone asks using a word that only exists in your company ("carretel" where the system says "bobina"), the question matches no tool and the answer disappoints. The synonym suggestion appears when a rejected term LOOKS LIKE a term the system knows. On approval, it enters the company glossary and Tenkai starts understanding both — the same glossary you see in Tenkai Context.

Only those who administer the account — those who can already change Company Settings, Access or Authority Limits. The reason: approving here changes Tenkai's behavior for the WHOLE COMPANY, not just for the person who clicked. And no one can delete the row, not even administrators: the trail of who approved what is precisely what gives this record its value.

Today the chat's 👍/👎 falls into a table and changes nothing. This screen is where it starts to change things: the question your team asks every week and that worked becomes a verified answer; the term your company uses and the system doesn't understand becomes a synonym; and the question nobody can answer becomes a declared gap. You look at the queue, check the reason and decide.

When a question repeats and NO tool covers the subject, approving doesn't turn anything on: it only records the gap in the list of what Tenkai doesn't know yet. It looks like the lowest-value item on the screen and it is the highest: a known gap enters the backlog and one day gets built; a hidden gap becomes "Tenkai is useless" three months later, without anyone knowing for what.