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. ↓
A debt instrument is two decisions in one. It funds something now. It commits a future stream to whoever holds the paper. The instrument is judged not by the existence of the obligation but by what the present-side spend purchases. If it purchases output that compounds beyond the carrying cost, the future stream is paid by surplus. If it purchases consumption, the future stream still gets paid, out of whatever else the obligor is doing. The instrument has not created value. It has redistributed it.
The structural pattern is most visible in public debt. The Treasury market is structurally necessary; a global system that trades in dollars needs somewhere to park them. What changes is the composition of what new borrowing finances. The primary US deficit, ex-interest, runs at 2.6 percent of GDP, historically normal. Interest payments fill the gap to 5.8 percent and consume 18.5 percent of federal revenue. Bond ownership tracks asset ownership generally, which is to say it is concentrated. Tax receipts come from the broad base. The mechanics produce a regressive transfer from labor to capital, and the size of the transfer is set by what the borrowing does not finance.
A sovereign issuer is not a household, and Modern Monetary Theory and Keynesian fiscal advocates push back at this scale: the productive-vs-consumption boundary blurs because aggregate-demand effects cross it. Possibly. The wealth-redistribution mechanism is independent of that argument. Regardless of whether the issuer can mathematically default, interest is paid from a broad base into narrow holdings.
The test is one question. Does the present-side spend purchase output that compounds beyond the carrying cost? Yes, the obligation is paid by surplus. No, it is paid by something else, and the instrument operates as redistribution. One discrimination, ruling in or ruling out.
The same structure operates at smaller scales, with different parties on the holder side. AI-compute borrowing has the cleanest analog. A startup raises capital, spends it on compute, commits future revenue to investors. Compute funding research that produces durable substrate is the productive case: an architecture that compounds, models that retain learning across runs, a graph that grows. The eventual surplus services the claim. Compute funding prompt-flailing without compounding substrate is the redistribution case: the carrying cost is paid eventually in equity dilution, founder time, or write-down, and the dollars have flowed from limited partners to GPU vendors without producing anything that pays them back. Same compute, same hours, structurally different shape.
Operator time has the same form at a different layer. Time committed to setting up a stream of outputs is the present-side spend; time spent reviewing is the future obligation. Review that compounds, where calibration sharpens or the operator's understanding of the system deepens, produces output exceeding the carrying cost. Review that cycles is the redistribution: irreplaceable attention pays the maintenance overhead with nothing flowing back. The time has gone from the operator to whatever is on the other side of the dashboard.
Tech debt is the same test on engineering optionality. Debt taken to ship a feature that earns durable usage and follow-on capability is investment; the artifact pays the future engineering cost. Debt taken for a vanity surface that requires ongoing maintenance is extraction from the team's future optionality. The carrying cost still comes due, paid in time the team will not spend on something else.
Three things to notice about the test.
It runs on the present-side spend, not on the obligation. The carrying cost is mechanical and known. The variable is what the spend purchases. The test is conditional on what the dollar, the compute, the hour, or the engineer-week is actually doing, not on the existence of the borrowing.
It is composition-aware. Aggregate debt-to-GDP can be steady while the composition rots. Compute-spend can be flat while the share going to substrate-building falls. The test fires on flows, not levels. This inverts the standard analysis, which sums the obligation and asks whether the level is sustainable. The level is downstream. The composition determines whether the level converges or runs.
It is structural, not moral. Welfare-state spending may be valuable for reasons unrelated to productivity. A research run may be valuable for reasons unrelated to whether it ships a model. The test does not deny those values. It says: when those flows are debt-financed, the future-claim mechanics do not care about them. The redistribution is the same shape regardless.
The test discriminates cleanly at the unit level: firm, household, individual flows where productive output is observable. It degrades at the sovereign level where aggregate-demand effects blur the productive boundary, and it loses quantitative force where output and cost are in different units. In knowledge-work and time-debt cases the test is binary, not quantitative: does anything flow back, not by how much. The discrimination remains useful as a forcing function. A defender of any flow has to specify what it compounds into, which can then be checked against what actually came back. The test does not name the answer. It forces the question.
This makes the test more useful applied to others' claims than self-applied. Motivated reasoning corrupts the input. A founder can rationalize any compute spend as substrate-building. A government can find studies showing any transfer compounds. The test's value is in forcing the specification, not in producing the verdict.
The honest framing generalizes from Banks: every committed flow is a decision to fund something now and a decision to commit future output to whoever holds the claim. Both decisions are present at the moment of borrowing. The test makes them visible, both at once.
The recursion is what the exercise produced. The repo this draft lives in is built on the proposition that operator-bound substrate compounds, that present-side compute and operator time invested in legible accumulation produce a graph whose future value exceeds the carrying cost. The architecture is the productive case in operational form. The same architecture, in a system that uses identical compute for unaudited inference and consumes operator time on cycles that do not update, would invert into the redistribution case. Same dollars, same hours, structurally different shape. The choice between productive and extractive is not a budget item. It is what the architecture is for.
The test is regime-conditional. If productive output stops being scarce relative to carrying costs, the discrimination stops mattering: the redistribution still occurs, but its operational consequence vanishes when no one is short the surplus. The test reads the current regime. Re-run as the regime changes.
And America? Solvency is not the test, if you can name the right surplus.
Source: Peter Banks, "Debt for Dummies," The Boyd Institute, 2026-04-24. Banks's piece is bounded to US fiscal policy; the structural test generalizes the pattern beneath. Source archived at z_seeds_readonly/boyd-institute/.