# Ethical Framework

> Philosophical doc 32.16.3.10. The commitments that constrain the system's
> power.

A system that intervenes in human expression and reputation holds real power
over people. The ethical framework is the set of binding commitments that
constrain that power, so the platform's good intentions can't curdle into a soft
authoritarianism. These are not aspirations — they are encoded.

## 1. Human dignity is non-negotiable

Every person is treated as an end, never merely a means to a clean platform. No
one is reducible to their worst moment (see
[Buddha Nature](01-buddha-nature.md)).

## 2. Autonomy is preserved

Friction is offered, not forced; the skip option is preserved with a _no-shadow-
penalty_ guarantee; predictive features are opt-out; consent gates optional
processing and can be withdrawn at any time (`user-autonomy.ts`). The system
nudges; it does not coerce.

## 3. Humans stay in the loop

Irreversible actions (bans, terminations) are never finalised by automation —
`requireHumanReviewForBan` makes a human the gate. High-severity actions require
evidence and an appeal path (`validateEthicalCompliance`). Crises route to
people, not bots.

## 4. Fairness is measured, not assumed

Every model is gated through `@kuanyin/evaluation` for demographic parity,
equalized odds, the four-fifths rule, and cross-cultural/linguistic disparity. A
model that discriminates does not ship.

## 5. Transparency over opacity

Every decision is explainable ("why am I seeing this?"); moderation statistics,
appeal rates, accuracy, and false-positive rates are published; algorithm
documentation and third-party audit are supported. Power that hides is power
that drifts.

## 6. Privacy by design

Analyse the minimum, retain less, keep the user in control (see
[Trauma-Informed](05-trauma-informed.md) and the technical privacy guide). The
system understands moments; it does not surveil people.

## 7. Protect the vulnerable, unconditionally

When commitments tension against each other, the safety of the harmed and the
vulnerable wins. Compassion for a person who causes harm never licenses harm to
anyone else.

## Living the framework

These commitments are checked, not just declared: fairness audits, transparency
reports, audit chains, and human-review gates make them auditable. An ethics
that can't be verified is just branding. Kuan Yin's wager is that a system can
be both powerful and trustworthy — but only if its power is permanently, visibly
constrained by principles like these.
