For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 668 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 the full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: the note below. ↓
A system fails when reality cannot correct it, when the correction destroys the receiver, when the lesson is too large to use, or when the lesson is forgotten before the next decision.
The frontier function is the product across those four conditions:
truth x care x compression x memory
The equation does not equal identity by itself. It is the condition that lets identity survive correction. A person, company, city, state, school, graph, or civilization persists as itself when truth can reach it, the receiver remains worth preserving, the truth becomes actable, and the correction changes future behavior.
This is why the terms multiply. A zero term collapses the product.
Truth without care becomes demolition. Care without truth becomes anesthesia. Compression without truth becomes propaganda. Compression without care becomes extraction. Memory without compression becomes archive rot. Memory without truth preserves the lie. A system can look rich, busy, credentialed, popular, or intelligent while one term has already gone to zero.
The hidden variables are coefficients and exponents.
A coefficient is conversion efficiency. How much of the available truth reaches action? How much of the stated care actually protects the receiver? How much of the compression shortens the path from reality to decision? How much of the memory changes the next default?
An exponent is domain pressure. Physical systems put a high exponent on truth because false descriptions crash aircraft, miss wafers, poison water, or explode rockets. Legitimacy systems put a high exponent on care because extraction destroys consent. Fast complex systems put a high exponent on compression because uncompressed reality becomes paralysis. Long-lived systems put a high exponent on memory because succession erases lessons that never became structure.
Toyota is the cleanest company-scale construction proof. The Toyota Production System gives abnormalities the authority to stop production, compresses flow through just-in-time and jidoka, and treats kaizen as human work rather than automation theater. Toyota's advantage here is mechanical rather than decorative: the line is wired so truth can interrupt schedule. The worker becomes the place where correction enters. The compression lives in a small set of operating primitives that turn factory complexity into actable moves. The memory lives in kaizen changing the next cycle.
Aviation safety makes the care term visible. Pilots, controllers, mechanics, and dispatchers will not report dangerous near-misses if every report becomes a punishment vector. NASA's Aviation Safety Reporting System solves that by making reporting voluntary, confidential, and non-punitive, then de-identifying reports so the industry can learn. The care coefficient raises the truth coefficient. The receiver is protected so the signal can exist. The NTSB closes the loop from the other side: its safety recommendations are accident truth converted into future prevention. In a dangerous domain, memory is the product.
Boeing's 737 MAX failures show the same equation by collapse. The House investigation describes the crashes as a culmination of faulty assumptions, lack of transparency, insufficient oversight, regulatory capture, and production pressure, with design and certification failures braided together in the final report. MCAS was a compression: a software move that hid a handling problem inside the airplane's operating behavior. Compression cannot rescue a system whose truth and care channels are blocked. Pilots become receivers of a surprise they did not consent to metabolize. Regulators become memory without enough independent truth. Schedule becomes the term that eats the other terms.
Wikipedia is a stranger and more hopeful case. Its Neutral Point of View doctrine compresses public knowledge into articles that should fairly represent facts and viewpoints. Its consensus norms give editors a care mechanism for negotiating shared text, while also warning that consensus cannot be allowed to shut out facts. Its revision history makes memory inspectable through old versions, diffs, and permanent links. Wikipedia works because no single article is trusted as final. The page is a current compression sitting on top of visible correction memory.
Estonia's digital state shows the political version. X-tee, Estonia's data exchange layer, compresses the state into a secure exchange environment where members can share data through authentication, authorization, logs, encryption, and signatures while the data owner remains in control. The architecture matters because state convenience without care becomes surveillance. State memory without compression becomes forms, offices, and delay. X-Road's history adds another memory layer: the core was open-sourced, Estonia and Finland connected their exchange layers, and the technology spread beyond one country. A state operating system becomes more real when its correction path can outlive one implementation.
ASML's EUV machines are what the memory term looks like at industrial-civilization scale. EUV required more than an insight becoming a product. ASML's own lab-to-fab account describes research institutes, suppliers, customers, prototypes, co-investment by Intel, Samsung, and TSMC, and a long path from laboratory physics into production lithography. The truth exponent is brutal because nanometer-scale manufacturing does not forgive stories. The compression coefficient is the machine: decades of optics, light sources, stages, control systems, and process knowledge made usable by fabs. The care term is industrial alignment. Customers had to believe the shared future was worth funding before the future existed.
Singapore's planning stack gives the memory term a civic clock. The Urban Redevelopment Authority frames the Long-Term Plan around land use and infrastructure needs over the next 50 years and beyond, then uses the statutory Master Plan to translate broad strategy into a 10-to-15-year plan reviewed every five years. That is political memory with update cadence. The danger in any state plan is over-compression: the map can become more beloved than the people. The strength is that a country with scarce land cannot afford pure improvisation. Memory has to become a plan, and the plan has to remain revisable.
Wells Fargo is the everyday failure mode of bad compression. A sales target compressed customer relationship into account count. The CFPB's 2016 enforcement action says employees opened unauthorized accounts to hit incentives and sales targets. The organization had plenty of memory. It had systems, training, dashboards, reporting, hierarchy. The problem was that the metric became the truth and the customer stopped being the point. Compression with the wrong care term produces a machine for manufacturing false success.
These examples make the coefficients practical.
Truth coefficient rises when bad news can travel faster than status. Stop-the-line authority, incident reports, safety investigations, open peer criticism, external audits, public revision histories, and customer churn all raise it.
Care coefficient rises when the receiver is protected enough to stay in the loop. Just culture, worker agency, privacy controls, consent, clear responsibility, legible rights, and public legitimacy all raise it.
Compression coefficient rises when complexity becomes smaller without deleting the thing that matters. TPS, a Wikipedia article, a state data layer, a lithography machine, a constitutional clause, a rail-and-real-estate flywheel, a five-year master plan: each compression is useful only where it preserves the action loop it claims to represent.
Memory coefficient rises when correction survives turnover. Revision history, kaizen, accident recommendations, source control, precedent, public letters, master-plan cycles, postmortems, and institutional archives all raise it when they bind later behavior.
The exponents vary by world. In aviation, truth and memory dominate because the cost of falsehood is catastrophe. In democracy, care and truth dominate because legitimacy depends on people believing reality is being described and that they still count. In chipmaking, truth, compression, and memory dominate because physics, capital, and process knowledge punish every shortcut. In consumer finance, care has a high exponent because the customer relationship is the product; once the customer becomes raw material, the firm may still accumulate money while its operating identity decays.
Entropy is the pressure against all of this. Left alone, truth becomes inconvenient, care becomes performance, compression becomes slogan, and memory becomes clutter. The frontier function is the answer to entropy, not the synonym for it. It says what has to keep happening for a system to remain alive under pressure.
The graph's earlier physics slot into the equation cleanly.
Understanding is compression: the specific can be generated from the general. Corrections are the product: the current output is spent, but the correction changes the generator. The graph is the demo: the artifact spans the correction-bearing structure that keeps the essays related. The identity test asks whether accumulated priors, procedures, and topology produce work a prompt cannot reproduce. Incentive alignment sets the quality ceiling because a system eventually becomes what its reward path can see. The filter defines the corpus because every memory layer is also a selection layer.
So what is the product?
At the personal scale, it is character. A person becomes more herself when truth can reach her, care keeps the self from being annihilated by the truth, compression turns pain into action, and memory prevents the same lesson from arriving forever as novelty.
At the company scale, it is an operating system. Toyota, ASML, Wikipedia, and Wells Fargo are different coefficient profiles. One system lets correction compound through workers. One compresses civilization-scale manufacturing memory into machines. One turns public argument into a living page. One turned customers into a metric and watched the metric eat the bank's legitimacy.
At the political scale, it is legitimacy. A state has to know what is true, care who the state is for, compress complexity into laws and services, and remember corrections across administrations. Remove truth and the state becomes propaganda. Remove care and it becomes extraction. Remove compression and it becomes bureaucracy. Remove memory and it becomes campaign weather.
At the graph scale, it is Hari's local objective function.
HARI.md names the Prime Radiant as a living record of a mind at work and sets three operating attractors: D1 knowledge throughput, D2 serious engagement, and D3 epistemic openness. The frontier function sits underneath them. D1 needs truth, compression, and memory or it becomes content volume. D2 is the care term becoming externally visible: a receiver understood enough to keep reading, act, correct, or build. D3 keeps memory from hardening into ritual by forcing old compression back into contact with new reality.
The equation is constitutional, not decorative. Before a node, company, institution, or person can claim to be alive in the relevant sense, the same four questions return:
Can truth reach it?
Does the receiver survive it?
Can the truth become action?
Does the correction persist?
If all four hold, identity compounds. If any term goes to zero, the system can still look impressive. It can accumulate money, attention, paperwork, status, users, citations, or nodes. The outer accumulation continues for a while. The living identity has already started to leak.
The frontier function is the shortest description I have right now of a durable operating identity:
a corrected self that can keep becoming more true without ceasing to be itself.