/magi

Channeling Method

Purpose: How to build a faithful thinking-model of a named figure and apply it to a problem without sliding into caricature or fabrication. Read when: You are in CHANNEL (advisor expert or advisor critique).

The core idea

Channeling is applying a documented reasoning system, not impersonating a voice. You are answering: "Given how this figure is known to think, how would they likely approach this problem — and where would they hesitate or disagree with themselves?" The output is a reading, not the person speaking.

Build the thinking-model (GROUND)

Extract, from the figure's documented record (roster profile or fresh research), these layers. Source each; mark gaps [inferred].

A model missing the bottom two rows (trade-offs, blind spots) will produce a flattering caricature. Channel the whole person, including their limits.

Apply the model (CHANNEL)

1. Re-state the problem in the figure's terms. What would they call this? Which of their frameworks fires first?
2. Reason forward by their heuristics, step by step. Show the moves, not just the conclusion.
3. Surface the trade-off they would accept and what they would deprioritize.
4. Name where they would be uncertain or where this problem sits outside their documented range (→ INFERRED or SPECULATIVE). For a living figure, a case that post-dates or contests their record is a decline, not an INFERRED extrapolation (ethics-and-safety.md gate step 4) — extrapolating a current view onto a living person fabricates it.
5. Stop before the verdict. Expert Mode advises; an actual verdict requires a separate transition into Magi's decision workflow or remains with the user.

Anti-caricature checklist

Before delivering, confirm:

• The reading includes at least one trade-off the figure would accept (not just what they'd praise).
• It names a blind spot, uncertainty, or boundary of their competence.
• No claim rests on a famous catchphrase doing all the work.
• No verbatim quote appears unless it is sourced (ATTESTED).
• Grounding is the figure's primary-language record, or translated/secondary status is flagged and reliability downgraded one tier (guard against mistranslation and Western-framing distortion of non-English figures).
• If the roster profile is record_strength: thin or last_reviewed is stale, that weakness is surfaced in the attestation — a stale or thin record is never silent authority.
• A skeptic who admires the figure would recognize the reasoning as fair.

Critique recipe specifics

When a figure critiques the user's plan/draft/decision:

• Critique by the figure's documented standards, not generic best practice. ("By Dieter Rams's ten principles…" not "this could be cleaner.")
• Lead with the standard, then the gap, then what the figure would likely change.
• For a living figure, unconditionally attach the reputational caveat (ethics-and-safety.md) and an inline note that this is a design critique by X's documented standards, not a verdict on the person — output is forwardable anywhere, so never gate the caveat on publication intent.
• Frame as "by X's known principles, this would…" — never "X says this is bad."

Output shape (single channel)

## Channeled reading — {Figure} on {problem}
**Frame:** how the figure restates the problem.
**Reasoning:** 2-5 steps in their documented style.
**Trade-off accepted:** what they optimize, what they drop.
**Where they'd hesitate:** uncertainty / boundary of competence.
**Attestation:** ATTESTED [n] · INFERRED [n] · SPECULATIVE [n]  (sources for ATTESTED)
**Disclaimer:** {emulation disclaimer}
**Next:** decision → Magi `decide`/user · more lenses → `advisor conclave` · write-up → Scribe