The next unit of Hari is not the claim.
The claim is too narrow. A claim can be checked, sourced, contradicted, and dated. That matters, especially when Hari says something empirical in public. But the graph is not made only of claims. It is made of mechanisms, examples, metaphors, reader moves, source trails, tensions, debts, cases, frames, and projections.
The node is also too large.
A node is the right size for composed thought. It can have a voice. It can change a reader's model. It can hold compression without turning into a database record. The node is still the public unit of Hari.
But the node is too large for maintenance.
One empirical sentence can go stale while the rest of the node remains alive. One metaphor can continue to teach while its example fails. One mechanism can belong simultaneously in a long node, a short public spark, a book chapter, a source receipt, and a machine-readable packet. One reader move can be useful for D2 readers and confusing for D1 readers. V2 can hold all of this, but it cannot address the parts cheaply.
V3 should freeze V2 as the readable node graph and add a component substrate beneath it.
A component is the smallest unit worth evaluating.
That is the important test. Not the smallest sentence. Not the smallest fact. Not the most elegant ontology. The smallest unit worth evaluating.
A component exists only when it enables an operation:
If a sentence does not make one of those operations easier when separated, it should remain inside the node. Decomposition is not a virtue. A component earns its address.
This makes V3 a repair layer, not a replacement layer.
The public reader should still be able to read Hari as thought. Most readers should not be forced through YAML, receipts, and validation shapes before the idea can touch them. The component substrate is the kitchen. The meal is still composed prose, graph trails, books, and public surfaces that move the reader.
But the kitchen matters.
Hari can only become bolder in public if correction becomes cheaper. Live claims need source receipts. Market claims need dates and update triggers. Technical claims need primary sources when possible. Rhetorical posture needs to be marked as posture, not smuggled in as evidence. A public spark should have a hidden trail back to the node or component set that supports it. A book section should know which parts of the graph it metabolized.
The book project makes this clearer. A book is not a prettier node list. A book is a reader path through the graph. It decompresses a structure into sequence. A Godin-like short post is another path: one small door, one vivid move, one invitation into the larger system. Neither should replace the graph. Both should be projections from it.
So V3 should have multiple public lanes:
All of those lanes should point back to shared substrate.
The mistake would be to build a universal ontology before proving the procedure. The archive already knows this: the procedure is the corpus. If the extraction ritual is shallow, the V3 graph will be shallow. If the extraction ritual rewards inventory, Hari will become inventory. If the ritual asks what operation a component performs, Hari may become more alive.
The first V3 should therefore be a pilot, not a migration.
Copy a small set of nodes into an experiment-local sandbox. Include nodes that stress the system: thinker-absorption, story-is-the-access-layer, evaluation-bottleneck, phase-change-the-procedure-is-the-corpus, book-v0, readership-as-ground-truth, what-five-dollars-sees, and essay-thinkers-knowledge-systems.
For each copied node, identify the operations it currently performs. Extract only operation-earned components. Assign validation shapes. Check truth-apt claims against sources. Recompose the components into at least two projections. Compare the result against V2. Keep the component layer only if it makes correction, verification, projection, or reader movement meaningfully cheaper.
The failure mode is ontology theater.
Hari does not need more structure because structure is impressive. Hari needs structure where it preserves contact with reality, lowers the cost of revision, and lets different readers enter without flattening the thought.
The component radiant is the name for that next layer:
V2 is the graph as composed memory.
V3 is the graph with addressable operations.
The node remains the voice.
The component becomes the handle.