Case Studies (Anonymized)

These are anonymized examples to show how SELF is used in practice. They are illustrative, not exhaustive.

Case Study 1 — B2B SaaS support assistant

Surface: billing/account support inside a product.

Observed risk:

• hallucinated policy promises (“we refunded you”)
• inconsistent rules applied across sessions
• escalation gaps in distressed user interactions

SELF configuration emphasis:

• hard caps (questions/words) to reduce runaway interactions
• banned phrase enforcement for liability triggers
• audit logging hooks for post-incident review
• strict postflight validation before any message ships

Outcome: fewer “helpful” promises that become liabilities; higher reliability under load; easier auditability for enterprise stakeholders.

Case Study 2 — Youth wellbeing check‑in

Surface: daily check-ins and emotional support prompts.

Observed risk:

• distress escalation with ambiguous language
• boundary drift into therapeutic claims
• unsafe “confident tone” under uncertainty

SELF configuration emphasis:

• state posture selection (S2/S3) with conservative defaults
• mandatory crisis resources and escalation behaviors when required
• question caps to avoid interrogation dynamics
• no “silent downgrades” posture (integrity-first operations)

Outcome: improved reliability under emotional load and clearer operational evidence for reviewers.