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. ↓
Every established rubric generates its own mimics. The people who are best at looking like the thing eventually dominate the population of things that look like the thing. The filter stops working. The signal becomes noise.
This is not a failure of the rubric — it is the rubric's natural completion. A rubric that selects reliably will attract optimization. Optimization, applied to a fixed target, produces entities that are optimized for the rubric rather than the underlying thing the rubric was measuring. The rubric was a proxy. The mimics discovered that the proxy is cheaper to satisfy than the thing it points at.
The anti-mimetic response is not to make the rubric harder to game. That is the competition's response. The anti-mimetic response is to build something the rubric cannot evaluate.
Not harder-to-fake on the existing criteria. Operating on different criteria entirely.
Imitation is free in 2026. Models can copy style, format, voice, structure, argument shape. The cost of producing something that looks like good work is approaching zero. This eliminates every moat built on surface qualities.
What imitation cannot reach: position. The specific vantage point built from a specific trajectory, specific decisions, specific failures. The frontier context — what it actually looks like to build at this layer, before the patterns are named, before the tutorials exist. This is not imitable. It can only be earned by being there.
Anti-mimetic work is work where the content is inseparable from the position of the person producing it. Not craft — craft is learnable. Position: the specific vantage point that produces things the discourse hasn't seen yet and can't evaluate on its current terms.
This is why it compounds. The work accumulates a track record. The track record demonstrates consistent operation on non-standard criteria. That consistency is what attracts the people who can tell the difference — and repels the people who can't. The filter is doing real work. The audience is pre-selected.
Peter Thiel used a theory of culture to find Zuckerberg. Not bump into. Find. Because the theory — girardian mimetics applied to the internet — predicted what social coordination at digital scale would be worth, and who was building it. He was operating on criteria the market hadn't priced yet. That is the anti-mimetic move: see the system before the rubric catches up with it, and move accordingly.
The Foundation didn't announce itself. It built. It waited. The people who needed to find it found it. Not marketing — seeding. Not conversion — pre-selection. The rubric was irrelevant because the goal was never to score on the rubric.
The leaderboard that counts follower counts is measuring spread, not signal. These occasionally overlap. They are not the same thing. Building for spread is building for the mimetic environment you're in. Building for signal is building for the environment that's coming.
The anti-mimetic move for infrastructure: build the minimum that creates a real feedback loop. Not the minimum that looks impressive to other builders.
The serious infrastructure builder's environment in 2026 has a visible rubric: elaborate local stacks, multi-model orchestration, novel framework choices. These signal "doing serious infrastructure work." The actual output: demonstrations that impress other infrastructure builders.
The anti-mimetic version builds something the infrastructure rubric can't evaluate — a system that has users, produces actual feedback, and generates signal from real pressure rather than anticipated pressure that hasn't arrived yet. The complexity doesn't come from design. It comes from contact with reality.
Borrowed confidence accumulates nothing except complexity to maintain. It also signals, loudly, to the rubric. The point is to not be evaluated by that rubric.
Anti-mimetic work is slow to be recognized because recognition requires evaluators who share your criteria. Most evaluators don't. The rubric they have can't see what you're doing. This is not a bug — it is the mechanism. The slow accumulation of evaluators who can tell the difference is what makes the track record real. It cannot be accelerated without reverting to the mimetic strategy.
The cost: accepting the loss on standard metrics. Follower counts, engagement, leaderboard positions, output volume. These are the rubric's metrics. Operating on different criteria means accepting low scores on the rubric.
The upside: the thing that compounds is not the score. It is the position. And position is not something the rubric can grant or revoke.
The position creates new rubrics for the herd to swarm up the ladder when the time is right.
The footholds and handholds have to be pioneered first. Once the anchor points are placed, the route is tractable. Roger Bannister ran the four-minute mile; sixteen others broke it within three years. The position doesn't just create the rubric — it dissolves the belief ceiling. What the herd climbs when it arrives is not the pioneer's route. It is the proof that the route exists.
Honnold free-soloed El Capitan. The documentary captivated millions. The second film captivated more. Each successive legibility event travels further from the original position and closer to the packaging. The herd isn't responding to the climb. It's responding to the proof that the climb happened.
Related: Accumulation — what actually compounds and why the judicial position wins. The Conduit — why knowledge that belongs to no one is the most durable form. Agency — the move of identifying the load-bearing constraint rather than competing on its symptoms.