Islands of Coherence

Sensemaking Atlas

The agentic component within Islands of Coherence. An evolving reading instrument that applies a small set of structured lenses to a curated corpus.

The Atlas is one component of IoC, not all of it. The Map is the artifact. The Criterion is the test the map applies. The Atlas is the working surface where readings get done at scale, by an instrument the curator tends rather than by the curator alone.

What it is

A corpus, a small set of evaluation lenses, agentic skills that apply the lenses consistently to any project page or any question that crosses islands, supporting plugins (semantic search, reference linking), and a publishing layer that reshapes the same intelligence into briefings, signals, talks, and outreach drafts.

Not autonomous AI. Agentic-enhanced reading. Every output goes through curator review. The reading discipline stays with the curator; the Atlas scales the volume and the consistency of lens application.

Locked design (2026-05-14):

Tech substrate: LlamaIndex + sqlite-vec + a current-generation LLM API. Working directory in the repository: docs/plans/agentic-sensemaking-commons/.

The lens stack

The Atlas reads through ten lenses. Each on a different axis. Each with its own vocabulary. The Reflexivity meta-lens reads the Atlas's own apparatus.

The stack is deliberately stacked, not collapsed into a score. When the lenses disagree, the disagreement is the finding.

The substrate

The corpus, the vocabulary, the lens essays, and the curated external reading the Atlas reads against. Each of the linked pages is one face of the substrate that the agentic skills operate on.

Current build state

What exists in the repository today, May 2026:

What is not yet built: the agentic skill layer that runs the full lens stack against a project page programmatically; the public query interface; the longer-form Field Signal cadence at scale.

What a reading looks like

Bring a question that crosses islands.

"What would a credible bridge between the Global Ecovillage Network and people aligned to conscious community but unexperienced look like, given current fragmentation?"

What the Atlas gives back is not an answer. It is a reading.

The Atlas surfaces what each lens sees. It shows where the lenses agree, where they disagree, and where each lens says I cannot read this. It pulls patterns from the corpus where comparable bridges have been attempted. It names what the corpus does not yet hold. It points to specific people whose practice touches the question. It flags false bridges where the question's framing might rest on a decorative connection. It maps actions a person in your position could take.

What the Atlas does not do: tell you what to do. Issue verdicts about named organisations or living people. Pretend to have read sources it has not read. Smooth lens disagreements into a single tidy score.

Open questions about how the Atlas should work

The Atlas is under active design. The questions below are unresolved. They are made visible here so the curator's gaps are legible to anyone bringing work to the Atlas. A question is included only if it is real, field-level, and live (its answer would change how the Atlas reads).

  1. How does the depersonalise filter handle figures whose personal credibility is a load-bearing field event? The methodology says: read the work, not the personality. But sometimes the personality is the field event. When does the annotation become the story?

  2. Where does the line sit between "this island is thin in the corpus" and "this island is not on the map"? Some archipelagos are named-but-undeveloped here. At what threshold does an undeveloped archipelago cease to be on the honest-incompleteness list and become a failure of curation?

  3. What does the relational claim require of AI systems, operationally? "The unit is the relation" is an ontology. What follows from it for the design and use of AI? The corpus has not yet answered it cleanly.

  4. How does IoC relate to a successor or fork? The methodology says forks are welcome. The substrate is portable. The voice is single-author. The relational stance toward a serious successor is not pre-decided.

  5. What is the explicit test for whether a bridge is load-bearing? The Atlas needs to distinguish a load-bearing bridge from a decorative one. The test is currently more intuitive than explicit. Can it be made explicit without becoming brittle?

  6. How is the untested category visually marked on the map without becoming a dumping ground? The category is meant to be a working flag, not a parking lot for unread orgs.

Questions move off this list when they are answered, reframed, or distributed back to a single archipelago's conversation rather than to the field.

What to expect, when

The Atlas is not yet open to public queries.

When it opens, it opens with a smaller scope than this page describes. A synthesis reading on a question someone brings. A reading map for newcomers to the field. A reading of an engagement when offered. A way to give the curator a witness from inside one's own work.

The fuller modes follow over time. Sustained dialogue while writing something long. A curriculum for learning the lens discipline. A watch for emerging convergences in the field.

The two Field Signals already published are the first hand-made artefacts from this work. They show what a reading from inside the discipline produces. Read them as previews of what the Atlas will eventually do at scale.

How to engage now

If you have a question you would bring to the Atlas the day it opens, Reach In. Tell the curator what you would ask. The question shapes what the corpus becomes.

If you have an observation from inside your own work that the Atlas should know, Reach In. Witness offers shape the corpus.

Either way, you find your way onto the substrate.

Last touched 2026-05-15. The Atlas is in active build. This page is the visitor-facing description of what is being made.