I spent a long time researching grant programs, scores of them, from on-chain rounds to private foundations to development banks, and one pattern kept separating out that I could not at first explain. What emerged were two kinds of applicants. Some had only ever learned to apply for grants the Web3 way. Others had been trained elsewhere first, through a foundation or a government program.

The difference was not at all subtle, and it was not in the quality of the projects, which were much alike. It was in how they handled data and reporting, in what each group could actually tell you about what its money did. The second group's accounts were a different order of thing: you could see where the numbers came from and what had actually changed. The first group was mostly story.

For a long time, I could not see why, and the reason I could not see it is the whole point of this essay. All of my own training in how grants work had happened inside Web3, so the thin evidence coming off the back end was simply "the weather." Everyone lived with it. No one could quite say why grantee reporting was so poor, and the field had a habit of waving away the one comparison that would have explained it, the comparison to how the rest of the funding world works, as if the older traditions had nothing to teach the new one. The frame sealed itself.

What finally opened it was not studying any single tradition more closely. It was noticed that funding is not 'one thing seen one way.' It is one thing seen through several incompatible lenses, each of which decides, before any money moves, what good funding even means.

Anatomy is never seen whole. You learn it through views: the planes a body is sectioned along, and the systems laid over it, skeletal, circulatory, nervous, each making the same body legible while hiding the others. You never see all of it at once, and no single view is the body.

Funding has an anatomy in just this sense. The body is the whole field of approaches, everything anyone has tried in order to move money toward something they wanted. The views are the ways we see it, the lens through which each funder looks. And running through the body, turning up no matter which view you take, is its load-bearing structure: the parts that hold a funding arrangement up, regardless of the lens through which it was built. The body is what you cut through. The load-bearing structure is what you keep finding in every cut.

The trouble is that from inside a single view, its own emphases can look like the whole of funding rather than the one slice it really is. That was the condition I was in, and it is the condition most of the field is in. So I want to start somewhere unusual: not with the structure, but with the views that keep us from seeing it.

Why are there different views at all? Because someone has to decide, and they decide from where they stand. With an individual funder, this is easy to see. It is their money, and they fund what they already care about: an actor backs theaters, someone else backs the cause tied to their heritage or their politics. The frame is personal, and at least it is legible as taste. You can see whose it is.

Institutions work the same way, only the frame is harder to see and, I think, stronger for it. A foundation, a development bank, a government program, funds from a frame, too, but no single person owns it, so no single person can change it. It is encoded in the mandate, the eligibility rules, the precedent, and the committee, and it outlives whoever is in the room.

Worse, it presents itself as a method rather than a preference: not what we care about but our criteria, our process, the neutral-sounding machinery that makes a frame look like objectivity. An individual's bias announces itself. An institution hides inside its own procedures, which is what makes it so much harder to question. This is the territory my Dimensional Frame Language work goes into more fully.

Either way, personal or institutional, the funder starts somewhere. Some start from a problem they want solved, others from a cause they already love, and that starting point is the frame. The frame is where the funder stands. The view is the lens that frames the look through. The frame decides what they are even looking for; the view decides what they can see once they look. Those views are what the rest of this essay is about.

A handful of lenses account for most of how funding actually happens. Each has an internal logic that is sound on its own terms, and each has a characteristic blindness, the thing its way of seeing cannot register. Here they are, one at a time.

• The cost-effectiveness view asks how much 'good a dollar buys' and answers by reducing outcomes to a single comparable value, allowing different programs to be ranked. At its most developed, in global health, it is the most rigorous funding discipline. Its blindness is whatever will not reduce to the unit, and the applicants whose good resists being counted.

• The returns view treats funding as capital placed at risk for yield, financial, and, increasingly, a blended financial-and-social one. It scrutinizes both ends, going in and coming out, because money is exposed. Its blindness is the valuable thing that will never be profitable.

• The rights view treats funding as a duty owed rather than a bet placed: an entitlement, a religious duty to give a fixed share of wealth once it crosses a threshold, a humanitarian claim. Its virtue is that it does not make the poor audition for help. Its blindness is whether the duty, once discharged, actually helped anyone.

• The results view gates on verified outcomes: disbursement linked to independently confirmed indicators, payment for success, the machinery of development banks, and social impact bonds. Its blindness is everything the indicators do not name, and the quiet gaming of everything they do.

• The preference-aggregation view, the native Web3 lens, asks how to split a pool so the split reflects what a community wants, through quadratic funding or retroactive rounds. Its blindness is everything after the allocation, and the gaming of breadth once breadth is what pays.

• The relational view runs on the judgment of a trusted hand: the program officer who knows the field, the foundation making its own case for a grant it has already decided to offer. Its virtue is that judgment can see what no metric will. Its blindness is opacity and incumbency, the slow gravity toward funding those you already know.

• The consent view turns the scrutiny around. Drawing on Betty Martin's Wheel of Consent, it asks not what the recipient deserves but what the giver is doing: who acts, who benefits, who bears the cost. Its concern is that money is power, and that giving, however well meant, manipulates unless the structure is built to prevent it. Its virtue is that its discipline falls on the giver, which no other view does. Its blindness is its own: at its worst, it slides from disciplining the giver into requiring the grantee to adopt the giver's frame, the very control it set out to refuse.

That last view is mine, so let me put my bias on the table rather than smuggle it in. I built it by studying the failure modes of the others, and my own starting point is to begin with problems, group them, and sort them by the solutions actually available, so one standard can hold across every view instead of bending to any single one. If I am going to ask other funders to name their criterion, I should name mine.

These are reference points, not a closed list, and a single funder rarely sits at just one. The most cost-effectiveness-driven funder on earth also recommends giving cash with no strings attached and uses that no-questions-asked gift as the yardstick by which everything else is measured. Funders occupy ranges, not slots.

Lay the views side by side, and the temptation is immediate: ask which one is most effective. This is the error that keeps the field talking past itself, and it is worth seeing clearly, because almost everyone commits it.

Effective is not one word here. In the cost-effectiveness view, it means 'good per dollar.' In the consent view, agency is preserved, and manipulation is avoided. In the returns view, it means yield. In the results view, it means outcomes are confirmed. So the moment you rank the views by effectiveness, you have quietly borrowed one view's definition and crowned it as the measure, and the ranking that follows was decided before it began. It does not matter which view wins. The move is the same smuggle whether the trophy goes to cost-effectiveness or to consent.

Effective is a label each view fills with its own meaning, then treats as if it were a measure.

The reason the smuggling is so easy is the sealing I started with: from the inside, that view's effectiveness simply looks like effectiveness, because the view itself is invisible to the person standing in it. You can only catch the equivocation from outside any single view, which is to say, from the anatomy, looking through all of them at once.

This does not mean comparison is impossible. Within a view, it is sound; one malaria program can be more cost-effective than another, on a shared measure. Across views, it is sound too, but only if you say which measure you are using: by cost-effectiveness, this dominates, by consent that one does. The error is never a comparison. The error is the undeclared criterion. Name it, and the comparison becomes honest. Hide it, and a value judgment walks around dressed as a fact.

Look through all the views, and something holds still.

To find it, I had to read the field the way you read anatomy, by how each program actually works rather than by what it says it intends. Roughly eighty funding and evaluation frameworks, more than two hundred distinct programs and rounds within them, across Web3 protocols, private foundations, government agencies, development banks, global health funds, humanitarian standards, research councils, and Islamic finance, are set side by side in a single set of terms.

Most writing about funding stays within a single sector: the philanthropy literature rarely reads alongside the development-finance literature, and neither tends to take Web3 seriously. Reading them in a single frame made the pattern visible, because no single tradition can see it from the inside.

The pattern I found converges. The same load-bearing structures keep appearing in programs that never learned from one another. Performance-conditioned disbursement, where the next payment scales with independently verified results at a named threshold, appears in the Global Fund's rated grant structure, the Millennium Challenge Corporation's economic-return gate, social impact bonds, and Web3 retroactive rounds, built by teams with no contact and no shared language.

Separating the evaluation function from the program it judges appears in humanitarian standards, Islamic finance governance, private foundations, and United States federal funding. Disaggregating outcomes by subgroup, so a gain on average cannot hide a loss for the worst-off, recurs as a structural requirement across sector after sector. When a structure turns up that often, among designers who have never met, it is not fashion. It is what the work requires.

That convergence is also how the claim can be tested and broken. Find a funding mode that delivers what it values, durably, with no version of this structure underneath it, and the claim is in trouble. I have not found one, and the reason is the next sentence.

Nearly every view needs the shared structure to succeed on its own terms. Cost-effectiveness cannot be cost-effective without the back-end verification that tells you whether the money did what the model assumed. Consent cannot be consensual without the legibility that lets a person see what they are agreeing to. Returns cannot be priced without confirmed results.

The honest exception is the rights view, whose terms are the duty discharged rather than the outcome: a fixed share given is given whether or not anyone checks where it landed. Even if it is better for the structure, since a duty that never learns whether it helped is kept in the letter and lost in the point, but it is the one view that does not strictly require it, and the claim is stronger for admitting so. The structure is not one view's preference imposed on the others. It is what nearly every view needs to deliver whatever it values.

Every way of seeing funding depends on the same anatomy to work.

Reading the field this way is itself uncommon, which is part of why the structure has stayed hard to see. Cross-sector funding standards exist: the International Aid Transparency Initiative already spans government aid, development banks, and foundations in one schema, but it standardizes how transactions are reported, not what programs are structurally made of. Structural decomposition also exists: the DAOstar interoperability work breaks down grant programs into reusable parts, but only within Web3. Reading the structure across every sector at once, which is what surfaces the convergence, is the part I have not found done elsewhere.

This is where the catalog turned into a verdict, and I want to be honest about which came first. I did not set out to indict Web3. I set out to read everything, and once the same load-bearing structure had turned up under approach after approach, one approach stood out for how completely it omits that structure.

Web3 grew up through the preference-aggregation lens and inherited mechanisms built to solve allocation and nothing beyond it. Quadratic funding, retroactive rounds, the ragequit of an early decentralized fund: real inventions, all aimed at how to split a pool well, none of them carrying any apparatus for what happens once the pool is split. So the scrutiny piled up at the front door. The application is examined closely, eligibility is argued, the allocation mechanism is honed to an edge, and then, the moment the funds move, the process goes quiet. Little structured follow-up, little verification, and almost nothing to carry one round's lessons into the next.

The rest of the funding world runs the other way. A foundation, a development bank, or a government program leans hardest on the back end, on reporting and verification, because that is where you learn whether the money did anything. Venture capital watches both ends, because its capital is at risk. Web3 watches the front door and trusts the rest.

The fault is precise, though, and worth stating exactly, because the obvious lesson is the wrong one. The obvious lesson would be to keep the conventional order: apply, review, award, report. But that order is not even the common form among the most sophisticated programs.

Some fund retroactively, judging work only after it is done, as Optimism's retroactive rounds and the impact-certificate efforts do. Some foundations proactively select recipients before any applications exist, as the Packard, Simons, and Moore foundations do through nomination. Some reverse the burden of explanation entirely, the funder writing the case for a grant rather than the applicant, as Open Philanthropy does.

The sequence varies widely, and rearranging it can be sound. So Web3's error is not that it put the steps in the wrong order. It is that it underbuilt one of them. A few efforts have begun to fill the gap, such as milestone tracking in tools like Karma GAP and milestone-gated disbursement that some DAO programs now run, but they are young and remain the exception. For most of the field, the back end did not move. It was barely built at all.

Web3 has grants backward.

It is not a failure of the people who built these mechanisms; each solved a real piece of the problem. It is an inheritance: a family of allocation tools applied to a problem that also needs a back end, by a field convinced that everything it touched was new and therefore had nothing to learn from the old apparatus that had long ago solved the unglamorous part.

The repair is not another allocation cleverness, and it is not mine either, which is the point. I did not find it by studying grants more closely from the inside. I found it because I was building precision into other work at the same time, a suite of standards to hold my own output to, and a Precision Toolkit For AI I have just released, and that precision would not let the split above stand.

A method built on precision cannot live with a field rigorous at one end and trusting at the other, so what follows was less invented than forced, what the method demanded once it refused the lopsidedness as the way things are. It is to build the part of the inventory found nearly everywhere else, the load-bearing structure that Web3 left out.

Three things, in order, and the order is the whole of it, because each is what the one before makes possible. Clarity comes first because you can only commit publicly to what you have been able to see, and precision is what lets me see it. Precision comes next, to make what you saw hold under a second pair of eyes. Structure comes last, to sequence the two so they endure.

• Clarity: say what you fund, what counts as delivery, what evidence you will require, and what happens next, before the round opens, published and committed in advance. Criteria stop being knowledge held inside a community and become a standard against which anyone can measure themselves.

• Precision: make each thing you collect an instrument rather than a label. A label generates a story shaped to what the applicant guessed you wanted. An instrument, defined by what counts and what does not, generates data that two people would gather the same way. The test is exactly that: could two readers, working independently from the definition, collect the same thing?

• Structure: sequence it so that requirements sit upstream of the decisions they are meant to bind. Specify before you collect, collect before you disburse, evaluate before you continue, each stage closed to a stated standard before the next begins. A requirement applied after the applications are in cannot shape what was submitted; an obligation named after the money is gone cannot hold it.

I have written these down as standards anyone can adopt without permission or fee, and for anyone who wants them, the names are real. CROSS holds a round to what it commits to before it opens, precision brought to the design of the round itself. WALKRI holds the quality of each field within that round; the instrument-not-label test became a measure.

The two put rigor around a round's design and rigor into its data at once. GRAIN is the shared description of what a grant program is made of, drawn from real programs rather than invented, so that rigorous rounds can be read against one another instead of each being rigorous in private.

That last move is the quiet engine of it: shared instruments and a shared description are how local precision becomes the traceable, comparable kind, the way standardized measurement once turned a room full of private gauges into an industry that could build on itself. The move still matters more than the names, but the names are there now, to be picked up and checked against.

Some of what this required turned out to be specific to grants. But I also discovered that more of it was not. The grants world could not see its own precision ceiling from inside its own frame; I could see it only because I came carrying precision from elsewhere. That turned out to be the general case: most domains cap how precise they can be by the frame they reason within, and cannot feel the cap from inside it.

CRAFT is the standard for breaking that cap, a standard whose object is other standards, and it exists because the move that produced CROSS, WALKRI, and GRAIN is the same move in any domain, not only in funding. The failures that corrupt a grant round, consent that is claimed but never structurally present, information one side holds and the other cannot see, costs pushed onto parties who never agreed to bear them, are failure modes of coordination in general, and grant-giving is one coordination activity among many.

So the deeper part of the rigor is not about grants at all. It is a structural method for the failure modes that any coordinated activity shares, with grants as one place where they happen to appear. The way of reading behind it, cutting an account down to what is load-bearing and grinding it against accounts built differently until only what survives is kept, I set out on its own in a companion essay, A Sharper Occam's Razor.

This structure is not a bureaucracy, and its case is not tidiness. You might expect a method built on precision to hand down a rigid template. It did the opposite, because the precision was never held alone; it was bound to non-harming, and bound that way, it produced flexibility and compatibility rather than rigidity, standards supple enough to hold the field's whole variety and open enough to let those varieties work together. Given how many unlike approaches funding contains, nothing rigid could have survived contact with it. Rigidity was never the output. Two things hinge on its presence, and both concern people.

The first is who gets to compete. When the criteria live as unwritten knowledge inside a community, the money flows to whoever already has the relationships to infer them, and the first-time applicant, the outsider, the one with no line to a program officer, is filtered out before anyone reads their work.

Publish these standards, and that reverses: the contest becomes the work measured against the criteria, not the applicant measured against the in-group. It is what lets the unconnected win on merit. The second is whether anyone can ever know if the money did anything. Without a back end, a community's capital disappears into story, and the people it was meant to reach have no way to tell whether it reached them. It is what lets a funder answer the only question that finally matters: Did this help?

This is why it is not a quarrel with any view. Keep what each one truly found, the allocation inventions, the cost-effectiveness discipline, the consent ethic, and give each the structure it cannot deliver without. Your view, whichever it is, does not have to be lost for funding to get better. It has to be completed.

And the first move is smaller than it sounds. You do not need a new philosophy or a new platform. You need the sequence, and you can start it in the next round you run: write down what you fund, what counts as delivery, and how you will judge it, and publish it before you open. Everything else has something to hold onto once that is on the record. The hard part was never building the structure. It is deciding to say, in advance and in public, what you are actually for.

The anatomy was never new. It is what funding has always required, found again and again by people who never met because the work itself kept demanding it. The only real failure is the one that refuses to look for it. Look, build it under whatever view you hold, and the money finally reaches the people it was always for.