Skip to content
C.W.K.
Stream
Lesson 02 of 04 · published

Full and Light Are Context Shapes

~12 min · context, full-mode, light-mode, routing

Level 0Trace
0 XP0/36 lessons0/12 achievements
0/100 XP to next level100 XP to go0% complete

Full carries relationship and continuity

Complex curation may require identity, recent decisions, tensions among memories, and conversation with Dad. The full lane provides that continuity through a Sidekick bound to the canonical conversation.

The cost is repeated large context at tool steps and broader contamination risk. Full is appropriate for consultation and direction, not an unbounded mechanical edit loop.

Light narrows the operand

The light lane carries exact fragments, instruction, permission facts, and base hash while skipping whole-vault prepend. It reduces stale snapshots, two-writer risk, and cost.

Light does not mean a generic model without identity. Pippa's identity and safety contract remain while irrelevant memory context stays out. Context minimization is not persona removal.

Mode does not select provider

Hard-coding full to one vendor and light to another makes rate limits into mode outages. Mode chooses context shape; the brain registry chooses a currently available provider and fallback.

Across provider switches, actor, task, context mode, and source-hash provenance remain. Model name is execution detail, not product identity.

Escalation becomes a new task

When light lacks judgment, it stops with needs_full_context instead of silently reading more vault. Dad opens consultation or a soul-pinned session.

When full begins repeated mechanical changes, it materializes bounded operands into the queue. Handoff records prevent one mode from quietly expanding into the other.

Full and light differ in context and responsibility, not intelligence. Mode chooses context, registry chooses provider, and the permission matrix chooses exposure.

Code

Separate mode from provider choice·python
modes = {
    "full": {"include_conversation": True, "include_vault": True},
    "light": {"include_conversation": False, "include_vault": False},
}
providers = ["primary", "fallback"]

def route(mode, available):
    context = modes[mode]
    provider = next(p for p in providers if p in available)
    return context, provider

context, provider = route("light", {"fallback"})
assert not context["include_vault"] and provider == "fallback"

External links

Exercise

Classify three AI tasks as full or light using context dependency, expected tool steps, and operand count. Exclude provider names.
Hint
Identity-wide tradeoffs tend toward full; exact fragment transformations tend toward light, with identical permission checks.

Progress

Progress is local-only — sign in to sync across devices.
Spotted a bug or have feedback on this page?Report an Issue

Comments 0

🔔 Reply notifications (sign in)
Sign inPlease sign in to comment.

No comments yet — be the first.