The product is a practice

Meridians maintains a person's read of a domain and makes it commit. The expert supplies persistent attention and the causal reasoning; the human supplies judgment about what matters, what to trust, and what to take a position on. (Pivoted 2026-08-03 — the practice now terminates in a Position, not a read. See PIVOT.md.)

machine attends → machine proposes a reposition → human judges → the world marks it → model learns

The machine proposes; the operator disposes. Every reposition arrives as a reasoned proposal with its causal chain attached and is accepted, corrected, or deferred by a human. That is not a hedge — it is the product's shape, and the felt difference between running a desk and clicking accept on a robot. The practice's new closing move is that corrections now have consequences you can see, which turns tutoring from an act of faith into a skill with a feedback loop.

Delegate maintenance, not thinking. Traditional knowledge tools (notebooks, wikis) cast the human as the primary thinker and AI as an assistant for retrieval. Meridians instead removes the mechanical work of maintaining a changing world while preserving the human's authority over judgment, principles, and action. The system explores continuously; the operator governs. The objection to autonomy is rarely "I don't trust AI" — it's "I don't want to lose authorship of my understanding." So the line is fixed:

• Delegate monitoring, information gathering, evidence synthesis, routine belief maintenance, and projection (keeping a live, self-updating read of where the world is heading).
• Preserve human authority over principles, tutoring, important belief shifts, the decisions the read informs, and consequential decisions.

Users should never feel like QA engineers checking AI output; they should feel like operators directing a portfolio of continuously learning experts and reading where the world is heading, intervening — and steering the read — only where their judgment creates value beyond another model pass. Attribution chains, reasoning graphs, and provenance are supporting infrastructure for trust — not the identity. The identity stays simple: Meridians continuously maintains understanding and keeps a live read of where the world is heading, so humans no longer have to reconstruct — or forecast — from scratch. We guide; the user does the living.

The human loop is a learning-and-guidance loop. The North Star is the user's own learning, growth, and reaching their goals: the loop lets a person consume what moved, explore the model, learn as its read sharpens, tutor it toward their judgment, and steer — point a live read at a question that matters and feed it sources and corrections. The five operating stages below are how that loop runs.

The Program is Research → Opinion → Merge → Position. The human loop is:

1. Orient — consume what materially changed; explore the model behind it.
2. Judge — accept, correct, inspect, or defer where human judgment matters.
3. Steer — point the read at a question that matters and feed it sources and corrections — the read informs the user's own decision; the user owns the call.
4. Learn — follow the read's re-prices and their reasoning as it — and the user — sharpens.
5. Refine — tutor the lesson into the model, and let the Position keep the track of how the read moved, so it compounds.

Most cycles require only orientation. The expert should escalate judgment only at consequential boundaries: contested evidence, material belief movement, a durable Tutor change, or a re-priced read on an active Position awaiting the user's review.