Research · white paper · v2.2 · not yet empirically validated
Coherence Intelligence
The Signal Is Not the Value: a first-principles framework for human intelligence in the age of AGI
Coherence Intelligence proposes that AI trained on human feedback inherits a systematic distortion: feedback given under fatigue, overload or isolation can drift away from what people actually value, in the same direction rather than randomly. CI proposes ways to measure that gap. CWHF proposes how to correct for it. Neither has been tested yet.
In 60 seconds
- What people click is not always what they value.
- The gap depends on their state: rested or exhausted, held or alone.
- If the gap leans one way, AI trained on clicks and ratings learns the lean.
- Coherence Intelligence proposes instruments to measure the state behind a signal.
- A pre-registered study, H1, is designed to find out whether any of this is true.
The privacy paradox
A person who genuinely values privacy clicks “Accept all cookies” on site after site. Not from ignorance: evaluating every choice costs more attention than they have left. The click is a signal. The value is something else.
The four proposed instruments
C320: The substrate
A proposed profile of 64 skills across four layers (survival, functional, cognitive, meta), scored from 64 to 320. Two people with the same total can be built very differently.
c: The state
A proposed coherence score for a person at the moment they give a signal.
Δc: The direction
How c changes from one session to the next. The paper's primary metric: is this person steadying or slipping?
Ψ(t): The trajectory
The shape of a person's coherence over time, across daily, weekly and longer rhythms. Same average, different waveform.
Research instruments, proposed and not yet validated. Not shown to individuals as personal scores.
Three mirrors
Self
How you rate yourself.
Peer
How people who know you rate you.
AI
What your behaviour shows.
When CI fails
- Coherent but wrong: a steady person can still hold a mistaken value.
- Cultural bias: what looks coherent in one culture may not in another.
- Anchor inflation: people may learn to look steady on the measure.
- Cold start: a new person has no history to compare against.
- CI does not measure moral correctness. It measures the state behind a signal, not whether the signal is right.
Where the research stands
Theory (current stage)
The framework and the pre-registration.
H1
Does a person's state predict how far their feedback drifts?
H2
Does weighting by state improve trained models?
Replication
Independent teams repeat it.
Status
| Pre-registration | OSF.IO/W7NYH |
| Evidence | No empirical data yet; H1 pending |
| Conflict | The author will not serve as principal investigator. |
| Falsification | If H1 finds no meaningful link between state and feedback drift, the framework is revised and the null result published. |
| Licence | CC BY 4.0 |
How to cite
Rao, V. (2026). Coherence Intelligence — The Signal Is Not the Value (v2.2). Two Point Singularity Press. CC BY 4.0. Pre-registration: https://doi.org/10.17605/OSF.IO/W7NYH
CWHF, the proposed correction · What is Two Point Singularity?