privacy/local-teaching.md

# Local OpenAI teaching boundary

Impara con Lucia is available in desktop Codex and, with verified capabilities,
actual local ChatGPT Work on the same OS user. The shared stdlib desktop_teaching
runtime stores a Lucia-specific professional summary, lessons, examples, progress
and optional feedback locally. Exact product identity, enrollment, revisions,
paths, hashes and native chat tokens are mechanical checks; legal reasoning and
understanding remain with the native model and professional. Lesson planning and
dispatch require a Lucia catalog workflow with a skill inside the same installed
Lucia root. Worker responses bind that product, root and skill path; native
teacher/worker instructions prohibit cross-product teaching.

Native OpenAI processes spoken conversation, the loaded profile and selected
material. Local storage is not offline inference. No custom model/voice API,
Mparanza interview, tutorial telemetry, change request or tutorial receipt is
called. Helpers do not save audio or raw transcripts. Tutorial files remain below
the compatibility local-only marker, including the actual private portable ledger.
A selected real assignment follows its specialist's data and review contract;
teaching state remains local even after completion. No Claude Cowork teaching.

Reviewed 2026-09-14 against scripts/local_onboarding.py, local_teaching.py,
local_onboarding_case.py, onboarding_session_start.py, the bundled shared runtime,
the learn-with-lucia skill and its local onboarding/session references.

The prepared course library uses local stdlib code, authored case material and
current-source hashes. It renders escaped HTML in a fresh directory without a
network/model call, external assets, profile changes or completion records.
Course sources remain bound to Lucia’s own catalog. The native host may read the
selected course for conversation. An authored specimen is not a newly executed
workflow; supplementary actual starter outputs preserve separate provenance.
Reviewed also against scripts/local_courses.py, assets/courses and the shared
courseware module. No additional recipient or transport is introduced.