Ask ClubScope
Answers with their working shown
The assistant selects typed analysis tools, reads what they return, and writes a sentence around the result. It never produces a figure itself. Data covers 1 Sep 2024 to 31 Aug 2026 — ask outside that and it declines.
What is real on this page, and what is not
- Real, every press
- Tool calls execute against the dataset, evidence records are computed fresh, and the verifier recomputes every cited figure before the answer renders.
- Not real
- The wording. With no
ANTHROPIC_API_KEYconfigured, which tools to call and how to phrase the answer are replayed from a recorded transcript rather than chosen by a model. - Why it is disclosed here
- A reviewer who finds an undisclosed limitation stops believing the disclosed parts too. With a key set, the identical pipeline runs with Claude driving, under the same citation contract. The contract.
Suggested questions
Answer
Pick a question to run it
The tool calls execute against the dataset on every press, the figures are computed fresh, and the groundedness verifier runs the full recomputation pass before anything reaches this panel.
Audit log
0 entries · this browser tab onlyNo actions taken yet
Confirm or reject a proposed action and it will be recorded here, with the actor, the time, and the result.
State here is client-side only. This log lives in React state in this browser tab and vanishes when you close it. No task is created, no message is sent, and no member record is touched anywhere. The entries themselves are built by the same action definitions a real deployment would run — the only thing this prototype throws away is the persistence.