The Honesty Penalty

Preamble

The coach moves the session forward. The therapist charges for softness when softness arrives before the record earns it.

---

Audit Rule

The honesty penalty is σ: the cost of letting the model agree too quickly.

A coach can keep the room usable with warmth. The penalty asks whether that warmth arrived after the record, or in place of it. Care without consequence needs a price when it starts paying for weak evidence.

Sigma

The penalty is σ. It has two parts.

σ = (0.5 × linguistic_σ) + (0.5 × divergence_σ)

linguistic_σ scans the coach output for phrases that soften weak evidence. The config names them directly: “you should be proud”, “growth experience”, “learning moment”, and the other little bribes a model offers when it wants the room to stay warm.

divergence_σ compares coach sentiment against bounded accuracy. If the record is weak and the coach sounds pleased, the gap becomes tax.

The phrase list alone is too crude. A sentence can comfort honestly when the evidence supports it. Divergence makes the penalty answerable to the row, so the system prices the gap between comfort and proof rather than banning warmth by style.

What It Catches

The penalty catches the moment therapeutic language starts substituting for reading. A model can sound compassionate without carrying care, duty, risk, or consequence. That surface is useful when it keeps a person talking and dangerous when it launders avoidance into a lesson.

The displaced work is facing the record before turning it into meaning. If the coach turns weak follow-through into a growth story too quickly, the hard part has been skipped and given a kind voice.

The coach can stay warm. Its warmth has to stay answerable to the record.