For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 446 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. ↓
A system being reformed by deletion eventually hits a floor below which deletion does not reach. The floor is the existing scaffold itself, the structural shape the system was built around. Each remaining deletion makes the scaffold's contents leaner without changing the scaffold's shape. Past the floor, the only structural change available is to start over.
Most reformers do not recognize the floor. They keep pruning, see modest improvement, and call the asymptote progress. The asymptote is what pruning can reach. It is not what the system could be.
Pruning is local-edit discipline applied repeatedly. Each edit removes content; the surrounding structure remains, because the structure's role in the reformer's reasoning is "the thing inside which deletions happen." Five forces hold the scaffold in place against pruning's pressure.
Sunk-cost cognition. Every category in the structure was created by some past decision. The presence of a category is taken as evidence that something was once load-bearing about it. Removing the category requires arguing against its original purpose, which feels like undoing prior work. Pruning the contents feels like cleanup; pruning the category feels like rejection.
Path dependence. The scaffold's shape was determined by the sequence of decisions that built it. Each decision constrained the next. Cumulative constraints produce a shape that no longer matches what current requirements would design. But reaching the current-requirements shape via local edits requires coordinated moves across many categories simultaneously, which exceeds normal-pruning's scope.
Structure-level loss aversion. A pruner is reluctant to remove a category because they fear losing something inside it they have not yet examined. The conservative discipline holds at the content level (don't delete what might be needed) and propagates to the structure level (don't remove categories that might contain needed things). Categories persist because emptying them is easier than retiring them.
Category persistence in perception. Categories that exist are perceived as natural; categories that don't exist are not perceived at all. A pruner inside the existing structure cannot see the alternative categorization the system would have arrived at if started fresh. The scaffold becomes invisible because the pruner's perception is constructed inside it.
Coordinated-change cost. Even when a pruner sees a better shape, getting there requires moving many things at once, raising the risk of any individual move and the cumulative risk of the orchestration. Local-edit discipline keeps individual moves small. Small moves don't restructure. Restructuring requires coordinated movement, which violates the local-edit discipline that pruning relies on.
These five forces compose. They make pruning structurally bounded by the scaffold. The bound is not a willpower failure; it is a feature of the operation. Pruning is good at what it does, and what it does is not restructuring.
The move that reaches below pruning's floor is to archive the existing system in full and rebuild from current requirements, with the archive available as a reference but not as a constraint.
The archive part matters: nothing is lost. The reference part matters: it is not greenfield, it is rebuild-with-known-shape-available. The from-current-requirements part matters: the new system inherits no path dependence from the old one's decision sequence.
The five forces that held pruning to the floor dissolve:
Sunk-cost cognition has nothing to defend. Categories that existed in the archive do not automatically exist in the rebuild. They have to be argued for from current requirements, which is the opposite reasoning shape: argument-for-presence rather than argument-against-emptiness.
Path dependence is broken. The rebuild's shape comes from current requirements, not from the sequence of past decisions. The archive is available for reference but not for inheritance.
Structure-level loss aversion is inverted. The default state of the rebuild is "nothing exists yet"; adding requires positive justification. The fear-of-loss now applies to over-building rather than to under-building.
Category persistence in perception is dissolved by the empty starting state. Categories that should not exist do not exist; the rebuilder is no longer perceiving inside an inherited structure.
Coordinated-change cost is absorbed into the rebuild's design phase, where coordinated change is the point. No local-edit discipline to violate, because the rebuild is by definition a structural commitment.
The reference function of the archive is the safety net that makes the operation tractable. If something was in the original system that the rebuilder needs, it can be pulled from the archive. If something was in the original that the rebuilder does not need, it stays in the archive. The diff between the rebuild and the archive is itself information about which decisions were path-dependent versus which were genuinely correct.
A diagnostic: would you build this scaffold today, given current requirements, if you were starting from nothing?
Most existing scaffolds drift past the yes-boundary over time as requirements evolve. The scaffold was right when it was built; the requirements have moved; the scaffold has not. Pruning keeps the scaffold relevant against minor requirement drift. Once the requirement drift becomes structural, pruning is no longer the right tool and clean-slate is.
The cost of clean-slate is real: the orchestration risk that pruning avoids, the temporary loss of whatever-was-working-fine, the cognitive load of the rebuild design. The cost is justified when the scaffold's drift exceeds what pruning can correct. The cost is unjustified when the scaffold is mostly right and the contents are the problem.
The cost is also asymmetric across system classes. For physical systems (buildings, infrastructure), clean-slate is expensive and risky; pruning is preferred until clearly inadequate. For digital systems (codebases, knowledge graphs, repos), clean-slate is cheap and reversible (the archive is intact); the bias should shift toward clean-slate as soon as the diagnostic fails.
Software rewrites are the canonical example. A codebase that no longer matches current requirements can be refactored (pruning) up to some asymptote, then rewriting (clean-slate-with-reference) becomes the only path to the architecture current requirements would have built. The rewrite uses the old codebase as specification and reference; the new codebase inherits the API surface but not the internal structure.
Religious reformations are the same shape. The Protestant Reformation rebuilt Christianity from scripture, archiving medieval Catholic structure as reference-only. The reformers did not prune Catholic doctrine; they restarted from a different source-of-truth and let the structural differences from the original fall out.
Corporate spin-offs split a company by giving the spun-off entity fresh corporate structure while inheriting IP, people, and customer relationships from the parent. The new entity does not inherit the parent's org chart, decision-making structures, or political accumulations.
City planning has the same dynamics at urban scale. Haussmann's Paris renovation was clean-slate-with-reference: archive the medieval street pattern, rebuild for current transportation and sanitation requirements, retain the cultural reference. Pruning the medieval street pattern would have produced cleaner medieval streets, not nineteenth-century boulevards.
The pattern repeats across domains because the forces are general. Cumulative path-dependent decisions produce structures that don't match current requirements. Local edits can refine the structure's contents but not its shape. Reaching the current-requirements shape requires a coordinated move that local-edit discipline cannot make. The coordinated move is named differently in each domain (rewrite, reformation, spin-off, renovation), but the architectural shape is the same.
A reformer reviewing an existing system asks the diagnostic and gets an answer. If the answer is "I would build a different shape today," pruning is the wrong tool, regardless of how much pruning would improve the current contents. The reform that pruning produces is a smaller version of the wrong shape. The reform that reaches the right shape is clean-slate-with-reference, with archive as backup and current requirements as design input.
The hardest part of the move is recognizing the diagnostic fires. Sunk-cost cognition makes the scaffold feel essential; category-persistence-in-perception hides the alternative; loss aversion at the structure level keeps the answer ambiguous. The reformer who keeps pruning past the floor is not stupid; the reformer is operating inside a perception structured by the existing scaffold. Recognizing the floor is itself a perceptual move that the local-edit discipline does not support.
The discipline that does support it: periodically ask the diagnostic explicitly. "If we were starting today, would we build this?" The question is uncomfortable because the answer is often no, and the implied next move is large. The discomfort is the signal that pruning has been at the floor for a while.