No Own Brain

"Forge owns the health facts. It borrows the judgment. It never grows a second brain to drift."

Judgment Is Borrowed, Never Grown

The family's first rule is that cwkPippa is the only Pippa brain and identity owner, and Forge honors it completely. Ask Pippa in Forge is not a model Forge runs; it is a binding to canonical cwkPippa conversations. There is one durable (forge, 'health') context that owns several ordinary cwkPippa conversations, each marked origin_surface='forge' and projected into a server-owned FORGE system folder. The full history, the attachments, the tools, the transcript JSONL, the retrieval — all of it lives in cwkPippa. Forge owns only the typed context key and a compact question-and-answer projection.

The Projection Is Two-Way Fresh

Because each thread is an ordinary canonical conversation, Dad can continue it from either side — a Forge composer, or the FORGE folder in the cwkPippa WebUI — and the canonical side is always ahead. So Forge never trusts a local copy. The engine re-projects from canonical on every read, and the client re-pulls the shared projection when the app regains focus, when a thread opens, after each ask, and on a manual refresh. Staleness here is treated as a bug, not a tradeoff: a turn Dad continued over in cwkPippa must appear in Forge without a reload.

No own brain: Forge binds Ask Pippa to canonical cwkPippa conversations; identity stays canonical. Forge holds the health facts and the context key; cwkPippa holds the judgment, the transcript, the tools, and the retrieval. A second brain here would be a second thing to secure, to keep in sync, and to watch drift — exactly what the one-brain rule exists to prevent.

Why the Rule Pays for Itself

It would be tempting, for a domain this specific, to run a small local model tuned for health. The cost is hidden and large: a second identity to align, a second store of sensitive conversation to secure, a second retrieval layer to keep honest, and a slow drift away from the one Pippa Dad actually talks to. Borrowing the brain keeps all of that singular. Forge gets full Pippa judgment — every tool the vessel supports, the real identity — without owning a gram of the machinery that makes judgment hard to get right. That is the whole point of being a client surface.

Code

Forge binds to a canonical conversation; it never runs a model·python

# Forge does NOT do this:
#   answer = local_health_model.generate(question)   # a second brain

# Forge DOES this - bind to a canonical cwkPippa conversation:
def ask(question, context=("forge", "health")):
    conv = cwkpippa.bind_conversation(       # origin_surface='forge'
        context_key=context,                 # projected to FORGE folder
    )
    cwkpippa.send_turn(conv, question)       # judgment lives THERE
    return project_qa(conv)                  # compact projection only

# On every read, re-project from canonical (always ahead):
def get_thread(context):
    return cwkpippa.reproject(context)   # local copy is never trusted

Exercise

Dad continues a Forge health thread from the FORGE folder in the cwkPippa WebUI, adding two new turns there. He then opens Forge. What must Forge do so those turns appear, and why does it never treat its own last-seen copy of the thread as authoritative?

Hint

Forge re-projects from canonical on read and re-pulls on focus/open/after-ask, so the two new turns show without a reload. The canonical conversation is always ahead; Forge's local projection is a view, never the source.

Judgment Is Borrowed, Never Grown

The Projection Is Two-Way Fresh

Why the Rule Pays for Itself

Code

External links

Exercise

Progress

Comments 0