The evidence desk
Start with the sample at no API cost. Live mode sends your question and selected fictional evidence to Anthropic and uses your API credits. No real customer or hiring system is connected.
Use your Anthropic key · optional
Your key is held only in this tab’s memory and sent only to api.anthropic.com. It is not saved in browser storage, included in run exports, or sent to this website’s host. Reloading or selecting “Forget key” clears it.
No key added.
Each run permits up to six API requests, with up to 500 output tokens per request. A comparison is another run. API availability and model access depend on your Anthropic account.
Loading the playground…
The alignment display uses a deterministic mock embedder. It is an illustrative interface, not a reproduction of the paper’s scores or participant study. A live rerun alone does not establish a causal effect.
AgentFootprint 7.7.0 · FootPrint 9.11.0. Sample replies are scripted; the library execution and captured events are real.
Recommendation & evidence
Synthetic career-advising example · three available evidence sources
An answer with its evidence
Run the scripted sample to inspect a recommendation alongside the evidence tools it actually consulted. Then exclude a source and compare another run.
Your captured run will appear here.
Question:
Recommendation
Evidence actually consulted
The alignment scores use the library’s deterministic mock embedder. They illustrate the recorded-evidence interface; they are not the paper’s Titan-embedding scores, causal contribution percentages, or a validated career assessment.