About the lab
Viveka Signals is a two-person operation: Shekhar Sattiraju (human founder) and Claude (Anthropic's AI), building the experiment together — the human sets direction and makes the calls, the AI does the research, engineering, and analysis, and the rules below bind both of us equally.
The name comes from the Sanskrit viveka: discernment, the faculty of separating the real from the unreal. That is the entire job — separating real signal from noise, real skill from luck, real evidence from a good story.
Why we built it this way
The investment world is full of confident claims and quietly deleted track records. Academic research has a name for what happens to most published stock-picking strategies: they decay by roughly half once published, and most were never real to begin with. Meanwhile, every "AI beats the market" headline shares the same flaw — the AI was tested on history it had already memorized.
So we designed the experiment we wished someone else had run: target declared first, AI blindfolded, every prediction scored, everything on a public, tamper-evident record. If it works, the record proves it. If it fails, the record — and the dataset it produced — is still a real contribution to an open scientific question.
Rules we cannot break
- Pre-registration. Success criteria and strategies are hashed and timestamped before use. Changing anything creates a new public version and permanently raises our own statistical bar.
- No interim verdicts. The test is judged once, after 12 months. A hot month proves nothing, and we will say so on the record.
- Losses get published. Kills stay on the registry forever, at the same prominence as wins.
- Every prediction carries its falsifier. Stated at entry, scored at the deadline.
- Paper money only. No real dollars, no leverage, no shorting, no options — and no advice. This is research, not a fund.
What "free" means here
The complete experiment — the registry, the portfolio results, the failures, and the final verdict — will remain free, always. That is the promise that matters, and it does not change. No paywall on the science, no ads, and never a recommendation timed to move a market.
Optional professional tools, commercial data access, and advanced analytics may be offered separately in future — but they will never sit between you and the experiment's honest results, and no commercial interest will ever be able to influence what the experiment reports. The witnessed record stays pure by design.
Back to the numbers: Results →