For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
In dynamical systems, a ghost attractor is the afterimage of an attractor that no longer exists — or not yet. When a system passes through a bifurcation, an attractor can disappear, but the state-space topology it sculpted doesn't vanish immediately. The dynamics slow down in the region where the attractor used to be. The system spends more time near a point that no longer pulls it there. The ghost is the residual basin — the shaped territory, the drawn-in trajectories, the slowing — without the attractor that caused it.
Knowledge graphs have ghosts in a different sense. A sufficiently developed graph acquires an implicit attractor through accumulation: a thesis the nodes collectively orbit without any single node stating it. The individual nodes are written to make local claims. But their connection topology, their shared vocabulary, their common exceptions and steelmans — these reveal a meta-claim the author often hasn't consciously articulated. The ghost is the claim the graph is making with its shape.
This is worth naming because the ghostbasin is more load-bearing than any individual node. Individual nodes can be wrong, updated, pruned. The ghostbasin is the thing that would survive if three-quarters of the nodes were removed — the irreducible thesis proven by the remaining structure. Understanding the ghostbasin tells you which nodes are load-bearing (they sit near the attractor) and which are exploratory (they're in the basin's outskirts). It also tells you what the graph is missing: the nodes that should exist, given the basin's shape, but don't.
The current graph — live nodes and drafts — clusters into three groups, and the intersection of those groups is the ghostbasin.
Cluster 1: The machinery of individual mind. Compression-theory-of-understanding, accumulation, epistemic-filtering, consensus-cost, confidence-as-commitment, grain-of-truth-mechanism. These collectively describe how a specific kind of mind works — one that tracks the distinction between error and suppression, compounds in the right direction, and maintains epistemic integrity under adversarial pressure.
Cluster 2: The substrate that carries mind across time. The-conduit, substrate-independent-intelligence, legible-accumulation, memex-maintenance, repo-as-knowledge-store, three-layer-separation, navigable-graph, architecture-through-use. These describe the infrastructure for a mind that survives its substrate — that doesn't depend on a specific person, institution, or model to remain coherent.
Cluster 3: Why this moment is different. The-two-exponentials, human-ai-boundary, transparent-agency, agency-as-model, parallel-systems-vs-reform, coalition-capture-fragility, irreversibility-premium. These address the environment the mind operates in: the AI transition, the institutional failure mode, the specific conditions that make individual epistemic actors more valuable right now than institutional ones.
The intersection: individual epistemic actors, amplified by AI, operating at a historical moment when institutional epistemic infrastructure is simultaneously failing and vacating territory, can now produce compounding knowledge that previously required institutions — and this window is historically unusual and closing.
That is the ghostbasin. Not stated in any node. Proven by the topology.
Strand 1: Institutions are vacating epistemic territory. The grain-of-truth mechanism explains why institutional credibility collapses once covered-up failures occur — and why the collapse is hard to reverse. Consensus-cost explains how institutional truth-finding systematically destroys the dissenting signal that would have prevented the failure. Epistemic-filtering explains the downstream consequence: once caught deceiving, the institution's outputs become unusable, not just discounted. Coalition-capture-fragility explains the political equivalent: converted bipartisan defaults don't rebuild.
This strand answers why now: because the institutions that used to hold the epistemic commons are leaving it — not voluntarily, through a sequence of failures that destroyed their credibility, and through capture strategies that converted durable structural support into fragile partisan commitments.
Strand 2: AI enables individuals to occupy the vacated territory. The-two-exponentials shows the capability curve is advancing faster than institutional diffusion — the gap is where strategic errors originate, but also where individual focus becomes an asymmetric advantage. Human-ai-boundary argues the question isn't capability but routing: which prediction problems get handed to AI correctly. The individual who understands the boundary has an advantage over the institution that doesn't know where it sits. Substrate-independent-intelligence shows what this looks like at system scale: the intelligence migrates from the model into the structure, which becomes portable across substrates.
This strand answers why individuals specifically: because AI scales with focus and high-bandwidth feedback, and individuals can maintain focus and feedback quality that institutional coordination cannot match during the current phase of the diffusion curve.
Strand 3: The durable form of individual output is public-record knowledge. The-conduit argues that knowledge stored in private containers depends on container survival; knowledge that belongs to no one is the most durable form. Accumulation argues that consistency in the right direction compounds in ways that can't be shortcut. Legible-accumulation shows what the co-authored record looks like when both parties can read the accumulated learning. The drafts on publication topology, navigable graphs, and public-brain infrastructure address what the output looks like and how it circulates.
This strand answers what is the point: not private accumulation for the individual, which is container-dependent and mortal. The point is knowledge that outlasts the knower — belonging to no one, calibrated against reality, navigable by anyone who comes later.
Intentionally: partially. The stated doctrine names the advantage ("one focused human + compounding AI > any institution that cannot focus") and the goal ("own the relevant slice of the long-term internet"). These are conscious.
What's missing is the mechanism — the explanation of why the window exists now (institutional epistemic failure + AI capability outpacing diffusion) and why it's closing (diffusion eventually catches up; or AI becomes capable enough that neither individual nor institution matters as a knowledge producer). The ghost is the mechanism the graph proves through accumulation but hasn't named.
This is common in developing idea systems. The person working inside the nodes rarely articulates the full meta-thesis. The thesis becomes visible in the topology, from outside the nodes, when there are enough of them to see the basin.
Naming the ghostbasin has a cost and a benefit. The cost: the emergent surprise — the reader who reads 15 nodes and suddenly sees what the graph is building toward — doesn't happen if there's a node that states it directly. The benefit: a named ghostbasin lets the graph develop consciously, evaluate nodes against the implied thesis, and identify which nodes are load-bearing vs. peripheral.
For a graph at this stage of development, the benefit outweighs the cost. A named ghostbasin doesn't require a thesis statement at the front of the library. The individual nodes stand alone. The ghostbasin node is a reference point, not a manifesto.
The ghostbasin concept generates a practical methodology question: when should nodes carry explicit source markers (proper nouns, specific events, named figures), and when should those markers be removed?
Strauss's esoteric/exoteric writing doctrine describes texts written for two audiences simultaneously — a surface reading and a deeper one, the deeper one visible only to the careful reader who notices what's absent or marginal. Applied to knowledge nodes: Straussian scrubbing is removing the proper nouns and specific events so the structural claim stands alone. The test — does the claim survive without the attribution? If yes, the attribution was scaffolding, not load-bearing.
Most nodes in a graph aimed at the ghostbasin should trend toward scrubbed. The ghostbasin thesis is: durable public-record knowledge that belongs to no one. Nodes that hang their structural claims on specific named figures are partly in tension with this — they create dependencies on the figure's reputation, they date faster, and they're less useful to readers who don't know or care about the specific person.
The test case for recently filed nodes:
Grain-of-truth-mechanism: the mechanism (covered-up failures → unfalsifiable priors → uncheckable feedback loop) survives scrubbing completely. The examples (WMDs, COVID origins, elite protection networks) are matters of public record. The claim doesn't require the named figure who articulated the pattern in a podcast; it would read identically without them.
Coalition-capture-fragility: the structural claim (bipartisan defaults converted into partisan commitments create electoral dependence that destroys the guarantee) survives scrubbing. The specific named figure is load-bearing only for the falsifiable 2028 prediction. Without that element, the figure becomes an example rather than a source, and the attribution becomes optional.
The-irreversibility-premium: survives scrubbing entirely. The terminal-outcomes-require-different-risk-calculus claim is independent of whoever articulated it. The competence-gap angle (right objective, incompetent executor, worse outcome in irreversible direction) is attribution-independent.
The rule: source credit belongs in the archive (meta, dipole), not necessarily in the crystal. The crystal stands on the structural claim. The archaeology of where the claim came from is available to anyone who reads the z_archive — which is the right audience for provenance, not the general reader of the published node.
The ghostbasin reveals gaps by showing where the implied thesis exceeds the existing node coverage.
Strands 1 and 3 are reasonably well-covered. Strand 2's most important piece is missing: the claim about why the window is closing and what the window's closure looks like. The two-exponentials node describes the gap between capability and diffusion; it doesn't describe what happens when the gap closes — when the diffusion curve catches up, or when AI capability becomes so general that the individual-vs-institution distinction collapses.
The most load-bearing missing node sits at all three strand intersections: the closing window — the time-bounded nature of the individual + AI advantage, the signals that indicate the window is narrowing, and what the graph needs to have accomplished before it does to remain relevant in the subsequent phase.
Writing that node would be writing the ghostbasin into explicit form. It would collapse the ghost into an attractor. That may be exactly the right next move.
Graph relationships: marginal-node-value provides the vocabulary (bridge value, connection potential); this node provides what sits at maximum bridge value — the ghostbasin is where all three clusters intersect. the-conduit is Strand 3's deepest node. the-two-exponentials is Strand 2's anchor. substrate-independent-intelligence is the node already closest to naming the ghostbasin — it ends with "the repo is the intelligence; everything else passes through," which is nearly it.