The lab · Signal ≠ Value · companion to the paper Signal ≠ Value · not yet empirically tested
The Signal ≠ Value Lab
A thumbs-up is a signal. It is not the value.
The lab separates what is known about the gap between what people signal and what they value, what is being tested, what a full simulation of the test shows, and what is only proposed. Every claim carries its tier.
Read the paper (PDF)Code on GitHub ↗
What is inside
- Evidence: 71 cases, each tagged with one of eight reasons a signal departs from a value.
- Reach: where human signals (thumbs, votes, upvotes, clicks) enter AI systems, and how far a result can travel.
- Flip test: answer 12 questions rushed and at your own pace, and see which way you drift. A demonstration, not data.
- Simulation: the full study run on simulated participants, including four design flaws it caught.
- Proposal, Method and Claims: CWHF v2 (record, prevent, weight, audit), definitions, and every claim with its status.
How sure are we
| A recorded signal is not a direct readout of value | Established |
| Pressure shifts choices toward a pre-named style (H1) | Hypothesis, untested |
| Power and false-positive rate of the planned study | Simulated |
| CWHF v2: record, prevent, weight, audit | Proposed, depends on H1 |
| Pilot or study data | None yet |
Questions
Is the lab evidence that CWHF works?
No. It sets out a testable hypothesis and the design to test it. No human data have been collected, and the simulation shows only how the study would behave.
Who built it?
Venkat Rao, drafted with AI assistance. The author has a commercial interest and will not serve as principal investigator of the test.
This is the research end of one story. The other end is Neev: who holds you. Read the story.
Related: CWHF and the research programme · Coherence Intelligence · Corrections