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. ↓
When you embed the same knowledge claim from 300 different perspectives, the perspectives that produce the most discrimination between claims are not the ones you'd expect.
Role frames (pilot, farmer, kindergarten teacher, prisoner, CEO) produce the most discrimination. Mean pairwise similarity: 0.716. Different life positions see the graph very differently.
Adversarial frames (nihilist, postmodernist, "someone who thinks this is all useless") produce the least discrimination. Mean pairwise similarity: 0.781. Critics see the graph as one homogeneous thing they oppose.
This ordering was not predicted. The tradition-distillation method was designed with disciplinary frames as the primary filter. The data says role frames are sharper.
A role frame forces evaluation of a claim's relevance to a specific situation. "As understood by a farmer" requires the model to ground the claim in a concrete world — crops, weather, markets, seasons. A farmer finds "understanding is compression" moderately relevant and "the citizenship schema conflates membership with presence" almost irrelevant. The farmer discriminates because their situation is specific enough to be differentially relevant.
An adversarial frame forces evaluation against a general counter-position. "From the perspective of a nihilist" requires the model to assess: would a nihilist reject this? A nihilist rejects everything for the same reason. The nihilist can't distinguish "understanding is compression" from "corrections are the product" — both are equally meaningless. No discrimination.
The discrimination mechanism is specificity of situation. Situated evaluation differentiates claims. Positionless evaluation homogenizes them.
The cross-category data confirms this. Adversarial and meta frames have nearly identical similarity structures (divergence 0.005). Both evaluate from outside — both lack situational specificity. Role and emotional frames have nearly identical structures (via the time-emotional divergence of 0.009). Both evaluate from inside a specific circumstance.
The two-kinds-of-universal diagnostic should be operationalized as "would a person in a completely different life position find these claims related?" — not as "would a philosophical opponent find them distinguishable?"
Adversarial frames test individual claim robustness. Role frames test cross-claim discrimination. Different tools for different operations:
| Operation | Best frame type | Mechanism |
|---|---|---|
| Claim robustness | Adversarial | Can opposition articulate a coherent rejection? |
| Pairwise discrimination | Role, Emotional, Time | Situated evaluation reveals differential relevance |
| Ghostbasin extraction | All combined | Intersection of all perspectives is the invariant core |
The no-enemies node argues: for any entity honestly running the compression filter, there is no stable enemy. Apparent enmity is diagnostic of closed identity on at least one side.
In embedding space, this is literally confirmed. Adversarial frames see the graph's internal structure as uniform — every claim looks equally like the thing they oppose. The adversarial frame's identity is fused to its opposition. Everything it opposes looks the same. This is 40 adversarial frames × 1,891 claim pairs = 75,640 measurements showing that closed identity compresses what it observes.
The farmer sees distinctions because the farmer's identity is open to the material — some claims are relevant to farming and some aren't. The nihilist sees uniformity because the nihilist's identity is closed to the material — nothing is relevant, in the same way.
This is "stable enmity is diagnostic of closed identity" measured in cosine similarity. Not metaphor. Data.
P.S. — Graph position
This node extends evaluation-bottleneck by identifying a new dimension of evaluation quality: the evaluator's situational specificity, not just their domain expertise. Specific evaluators discriminate. Generic evaluators homogenize.
It revises tradition-distillation: use embodied frames, not adversarial ones.
It extends two-kinds-of-universal (paperclips): the constraint/attractor diagnostic should be tested with positional frames (would a farmer, prisoner, and astronaut all find these claims related?) rather than theoretical frames.
It empirically confirms no-enemies: adversarial frames see the graph as one target. Closed identity compresses observation. 75,640 measurements.