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 2002, Elon Musk went to Russia to buy rockets. He wanted to send a greenhouse to Mars — plants growing in Martian soil, photographed and transmitted back. A PR mission for space. The Russians wanted $8 million per rocket. Musk thought this was too much. He started asking why.
Not "is there a cheaper supplier?" — that's a search query. He asked: what does a rocket actually require? He worked through the materials: carbon fiber, aluminum, titanium, copper, steel. He priced the raw materials. He found that the raw materials for a rocket cost about 2% of the rocket's price.
The gap between 2% and 100% is not physics. It is manufacturing convention, incumbent pricing, and institutional inertia compounded over decades of government-contract aerospace. Musk founded SpaceX.
The first-principles method has a specific structure. It is not "think from scratch" or "ignore conventional wisdom" — these are approximations that miss the mechanism.
Step 1: Identify the physical ceiling. What does physics actually allow? For rockets: specific impulse is bounded by exhaust velocity, which is bounded by the chemical energy content of propellants. For batteries: energy density is bounded by chemistry. Lithium-ion cells have a theoretical maximum determined by the electrochemistry, not by manufacturing maturity. The physical ceiling is the oracle. It is the one input in the problem that cannot be negotiated with, lobbied against, or changed by incumbents protecting their position.
Step 2: Audit the gap. Everything between what physics allows and what the industry does is a hypothesis about why you can't close the gap. Some constraints are genuinely physical: you cannot violate thermodynamics. Some are institutional: regulatory frameworks designed for different technology. Some are economic: incumbent pricing, risk aversion, amortized tooling costs from decades of prior decisions. Some are historical accidents that became standard without anyone asking if they should.
Step 3: Treat the surviving gap as the design space. Hypotheses that survive examination — "this constraint is genuinely physical" — define the minimum achievable. Hypotheses that fail — "this constraint is conventional, no one has tried to remove it" — are the opportunity. Close the gap between the minimum achievable and current practice, constraint by constraint.
This is a knowledge generation method, not a knowledge organization method. The output is not a more navigable representation of what's known. It is new knowledge about what's possible.
The method's power comes from the oracle's trustworthiness, not from any particular intelligence of the person applying it.
Physics is a truthful oracle. It does not have incumbents with interests in maintaining high launch costs. It does not have regulatory agencies with frameworks designed for 1960s manufacturing. It does not accumulate error through institutional inertia. Physical laws are falsifiable and have been tested against reality extensively enough that their core claims are reliable at engineering timescales.
Everything else in a physical domain — industry practice, cost structures, regulatory frameworks, component specifications, professional norms — is a human construction that may or may not be well-calibrated to current materials, manufacturing capability, and market conditions. The first-principles method treats human constructions as hypotheses and physics as the ground truth against which hypotheses are tested.
Feynman made the same move from the other direction. His criterion for whether he understood something was: can I make it? Can I derive it from more primitive assumptions? The physicist's method and the engineer's method are the same epistemology applied at different timescales — Feynman's question is "what does nature actually permit?" and Musk's is "what has already been demonstrated about what nature permits, and how far is current engineering from it?"
The Space Shuttle cost approximately $1.5 billion per launch. Falcon 9's first commercial launch cost approximately $62 million. In 2025, reused Falcon 9 launches cost under $30 million for standard payloads.
The first-principles audit identified the dominant cost driver: the first stage. The first stage contains most of the hardware and is the most expensive component to build. In all prior launch vehicles, it was expended — separated from the rocket during ascent and discarded in the ocean. The question "can you recover and reuse the first stage?" is a physics question. The physics of returning and landing a rocket booster are not prohibitive: it requires additional propellant and control surfaces, which reduce payload fraction but don't violate any physical law.
The reason no one had done it was not that it was physically impossible. It was that the institutional context — government cost-plus contracts, heritage certification requirements, organizational structures optimized for expendable vehicles — made the engineering investment unattractive. Reusability would require redesigning for it from the beginning, and that investment was not recoverable within any existing government program structure.
SpaceX had no existing hardware to protect and no cost-plus contract to optimize for. The institutional constraints were absent. They could optimize for reusability from the start. By 2015 they were landing Falcon 9 first stages routinely. The cost-per-kilogram to low Earth orbit dropped from $54,500 (Space Shuttle) to approximately $2,720 (Falcon 9 reused) — a factor of twenty, achieved primarily by auditing which constraints were physical and which were institutional.
Musk gave the same description of the method in a 2013 interview, discussing electric vehicle battery costs:
"The first principles question is, what is the physical ceiling on battery energy density? What materials are we using? Cobalt, nickel, aluminum, carbon, some polymers for separation and a steel can. Break that down on a materials basis and ask, if we bought that on the London Metal Exchange what would each of those cost? It's like $80 per kilowatt-hour. So clearly you just need to think of clever ways to take those materials and combine them into the shape of a battery cell and you can have batteries that are much cheaper than anyone realizes."
In 2013, the industry consensus was approximately $600/kWh as a reasonable near-term target. The physics ceiling, as Musk calculated it, was $80/kWh. The gap was $520, and none of it was physics.
Tesla's battery costs were below $100/kWh by 2025. The roadmap from $600 to below $100 is a twenty-year project of auditing the gap between the physical ceiling and industry practice, then removing the non-physical constraints one by one. Manufacturing improvements, supply chain development, cell chemistry optimization, battery management systems — each is a hypothesis about where the gap comes from, tested by building and measuring.
The essay-thinkers landscape covers knowledge organization and transmission: Graham compresses within a domain to find the generative axioms. Cowen maximizes coverage, trusting volume to reveal patterns over time. Naval compresses for transmission, optimizing for portability across minds. Collison curates, trusting source material to speak to prepared readers.
Musk's method is orthogonal to all of these. He is not organizing what is known about rocketry. He is generating knowledge about what is possible with rocketry that has not been tried. The question his method answers is not "how do I represent what's known?" but "what are the actual limits, as opposed to the assumed limits?"
The distinction is: knowledge generation vs. knowledge organization. Generation comes first. If you don't know the physical ceiling, organizing the existing state of the art is organizing the wrong thing — you're optimizing within a constraint set that includes many hypotheses masquerading as constraints.
Physical constraints are exogenous; social constraints are endogenous.
This is the precise limit of the method. A physical constraint does not change because you try to remove it. If thermodynamics says the ceiling is X, the ceiling stays X regardless of how many engineers try to exceed it.
A social constraint — a regulation, a market norm, a professional standard, an organizational incentive — is endogenous. It can reorganize in response to attempts to change it. Incumbents lobby. Regulations update. Norms shift as new entrants force the question. The method assumes you can identify the constraint, audit its origin, and remove it if non-physical. In social systems, removing a constraint often produces a new constraint — the system adapts.
This is why the method works in physical engineering and fails in social engineering. SpaceX could audit launch costs and remove the non-physical constraints because the physics didn't fight back. Social systems do.
The steelmanning adds a second limit: the oracle works only when the physics is well-understood. The ceiling Musk calculated for batteries assumes current electrochemistry. Different chemistries have different ceilings. At the frontier of materials science, the ceiling is not yet known — the oracle is silent where the physics is not yet understood. The method provides the most guidance when physics is mature and engineering is immature. It provides less guidance when neither is settled.
One more limit, from the steelmanning: the method risks classifying hard-won engineering knowledge as waste. Some of what looks like "institutional artifact" in the gap between raw materials and finished product is accumulated problem-solving: QA processes that exist because earlier approaches failed in expensive ways, certification procedures that reflect real failure modes, safety margins derived from operational experience. The method correctly identifies this as not-physics, which is true. It doesn't automatically tell you which non-physical constraints are worth removing and which encode genuine experience.
What survives all four steelmanning challenges: physics remains a more trustworthy oracle than industry consensus. Even an imprecise physics ceiling is more honest than a consensus benchmark. The method's value is not that it always identifies the right ceiling — it's that it provides a check against reasoning anchored to convention. Convention-anchoring is the failure mode the method corrects. The correction is most powerful when the physics is clear; it degrades gracefully, not catastrophically, when it isn't.
Graph P.S.:
Written 2026-04-12.