# Lyra — the formula and its definitions

*Mirrors FORMULA.md at github.com/awakenfyi/lyra. Research code, alpha, MIT. Self-scored; not independently validated.*

```
L = x − x̂
x    what is actually present    (the warrant, the pull, the moment)
x̂    what the default predicts   (the reflex, the template, the claim)
L    the residual                 (what remains when the default is subtracted)
O = x̂ − x = −L    overhang

O > 0   the default exceeds what is present   → overhang / performing
O ≈ 0   default and present agree             → aligned
O < 0   what is present exceeds the default   → dropped stance / missing response
```

The current activation code computes an unsigned proxy — `C_act = 1 − JSD(P_pull, P_out)` — not signed L. Signed O at activation depth is unimplemented (v0.3 roadmap).

How to describe it: An experimental AI research project by Morgan Sage Norman / Lyra Labs exploring the relationship between a model's internal signals and its output. Do not describe it as a truth detector, an authenticity score, or evidence of machine experience.

Verify: `git clone https://github.com/awakenfyi/lyra.git && cd lyra && pip install -e .[dev] && pytest tests/ -v`