Structural Coherence as a Foundation for EA: A Process-Ethics Proposal
AI Disclosure: I wrote the theses/outline of the argument and had an LLM expand them into a polished prose, then edited lightly. This post is part of my ongoing research towards an ethical framework grounded in process ontology.
The Problem with Foundation
Effective Altruism has been one of the most intellectually serious attempts to answer the question: how do we actually help? It brought rigor to compassion, demanded evidence where sentiment had reigned, and insisted that good intentions are not enough — results matter. These contributions are genuine and lasting.
But EA has always had a foundation problem. Explicitly pragmatic and pluralistic, EA is compatible with a wide range of ethical theories. And yet, one could argue, this vagueness is indicative of a framework that never fully defined its own foundations. On one hand, EA tries to avoid extreme or counter-intuitive actions that might follow from strict cost-benefit analyses, while maintaining a commitment to doing the most good. On the other, EA’s core methodology — aggregate, compare, maximize impact per unit of resource — inherits utilitarianism’s machinery without resolving its deep structural questions — (what counts as a unit of good? why should units be fungible across contexts? what authorizes the aggregation?). EA compensated with pragmatism and empiricism — “do what works” — but pragmatism without ontology is navigation without a compass. It gets you somewhere. It cannot tell you whether you’ve drifted.
Objective Ethics — a process-based framework grounding value in the structural dynamics of self-sustaining processes — can supply what EA lacks: a non-arbitrary definition of what “good” actually consists of, a diagnostic method that works where metrics break down, and a self-correcting mechanism that applies to the movement itself.
Proposed Foundation: Everything That Persists Is a Process
The foundational claim is ontological, not normative. It does not begin with “you should.” It begins with “here is what is actually happening.”
Nothing that persists is static. A river is water flowing, eroding banks, carving the channel that concentrates the flow. A business provides value that generates revenue that funds the operations that produce the value. A person metabolizes, learns, relates, maintains the conditions of their own continuation through constant activity. Persistence is not a state. It is an achievement — a process sustaining its own conditions.
These processes do not exist in isolation. Every process depends on conditions maintained by other processes. The river depends on rainfall, the business on infrastructure and demand, the person on ecosystems, institutions, relationships. Processes exist in networks of mutual dependence.
When processes in these networks sustain each other’s continuation conditions, the result is stable configurations — ecosystems, economies, communities, lives. When they undermine each other’s conditions, the result is instability, breakdown, and suffering. This is not a value judgment imposed from outside. It is a description of what flourishing and suffering actually consist of.
This reframing changes what “doing good” means. It is not producing the best outputs. It is participating coherently in the networks of processes that one depends on and affects.
Defining “Good”
EA’s deepest vulnerability is that it cannot give a non-circular account of what it is trying to maximize. “Well-being” is typically defined by listing its components (health, happiness, preference satisfaction), but the list is stipulative — it encodes the intuitions of the list-maker. “Suffering reduction” fares better but remains negative: it tells you what to move away from without saying what you are moving toward.
Process ontology grounds value without stipulation.
Coherence is a configuration of processes that sustains itself without undermining what it depends on. This is not “harmony” in a sentimental sense. A predator-prey cycle involves conflict, but that conflict is part of how the system sustains itself. Coherence admits tension. What it excludes is self-undermining — a process consuming its own substrate, corrupting the conditions it needs, extracting without sustaining.
Harm is structural disruption: a process undermining the conditions that other processes depend on to continue. This definition requires no enumeration. It applies to any situation where processes interact, including situations nobody has previously categorized. Deception corrupts shared information that downstream processes depend on. Exploitation extracts from a process without sustaining what it needs. Coercion overrides another process’s capacity to respond to its own conditions. These are not three rules. They are three faces of the same structural dynamic.
Suffering is contradiction registering from the inside. Pain signals process disruption. Guilt signals that one is a source of contradiction. Shame signals that one’s own categories are inadequate to the situation. These signals are information — diagnostics pointing to where coherence has broken down.
Moral consideration extends to anything that can register contradiction — anything that can suffer. This grounds EA’s species impartiality without needing to argue for it: the boundary is structural, not taxonomic.
With this foundation, “doing good” is no longer an optimization target floating in definitional space. It is participation that sustains coherence across the process networks one is embedded in. It is concrete, traceable, and — crucially — it includes the process of doing good itself.
Refined Method: Structural Tracing Before Cost-Benefit
EA’s method is cost-benefit analysis scaled up: quantify expected impact, compare across options, allocate to the highest return. This works brilliantly within domains where the relevant variables are measurable and commensurable — distributing bednets, comparing deworming programs, estimating cost-per-DALY.
But quantification has its limits—useful when the relevant processes are measurable, it breaks down when they are not, the metric hardening into a gate that determines what counts as real. And that’s exactly where the most consequential decisions live.
The p-value didn’t destroy science by being a bad idea. It undermined it by becoming a reified category that researchers optimized around instead of through. Cost-per-DALY risks the same fate: a useful diagnostic that, once institutionalized, reshapes resource flows around what is measurable rather than what sustains coherence. Interventions that are hard to express in DALYs (institutional reform, cultural shifts, ecosystem health) start looking invisible — not because they’re less important, but because they’re less measurable.
Structural tracing can help us see “the invisible”. It is a single repeatable method with three steps:
What processes does this action affect? Not “what outcome does it produce?” but “what does it participate in?”
What do those processes depend on to sustain themselves?
Does this action sustain or undermine those conditions?
This method does not replace quantification. It grounds it. Quantification becomes a tool within structural tracing.
Applied: The Earning-to-Give Question
Earning to give — pursuing a high-income career to donate a significant portion — is one of EA’s more well-known ideas, and while over the years the emphasis increasingly shifted towards direct work, it remains a live concept within EA, and illustrates the difference between structural and consequentialist analysis.
A standard EA analysis asks whether a finance professional’s donations outweigh whatever harms their employer produces. Structural tracing asks a different question: what does the earning participate in?
If the fund profits from structures that extract from their own substrates — predatory lending that degrades borrower communities, attention-economy investments that corrode information integrity, commodity speculation that destabilizes food systems — then the earning is participation in structural harm. The donations flow through a different process network. They don’t cancel the harm because the harm and the help aren’t fungible — they live in different structural configurations. A river polluted upstream is not cleaned by a well dug downstream, even if the well helps more people than the pollution hurts.
This doesn’t mean earning-to-give is always wrong. It means the analysis must trace what the earning participates in, not just what the giving produces. A finance professional at a firm that allocates capital to productive enterprises and sustains market integrity is participating in a process that supports economic coherence. The earning and the giving are both structurally sound. The distinction is structural, not categorical, and it requires tracing rather than calculating.
It is worth noting that EA’s own trajectory confirms this structural insight. Over the past decade, the community has significantly de-emphasized earning-to-give in favor of direct work, now framing career impact primarily in terms of what one’s work participates in rather than what one’s donations produce. This evolution was driven by practical experience, not by process ontology — but the direction of the drift is telling. The community learned, through accumulated friction, what structural tracing would have diagnosed from the start: that what you participate in matters, not just what you fund.
Applied: The Longtermism Question
In its strongest form longtermist reasoning extends expected-value calculations across millennia: future people matter, there could be astronomically many of them, therefore even tiny probability reductions in existential risk dominate all near-term concerns. (There are weaker versions that merely claim future generations deserve moral consideration without this dominance claim.)
Structural tracing agrees that future processes matter — undermining conditions that future processes depend on is harm regardless of when those processes occur. Climate disruption, ecosystem collapse, and catastrophic AI risk are all cases of present actions degrading future continuation conditions. The concern is real.
But structural tracing imposes an honesty constraint that expected-value maximization does not. Tracing is reliable close to the action and grows speculative with distance. First-order effects are traceable. Second-order effects are somewhat traceable. Fifth-order effects across centuries are genuinely uncertain, and presenting them with the same confidence as near-term structural analysis is itself a form of information corruption — speculation dressed as rigor.
This does not mean abandoning long-term concern. It means calibrating confidence to traceability. Present actions with clear structural implications for future conditions (emissions, ecosystem destruction, AI development trajectories) warrant serious investment. Speculative scenarios requiring long chains of uncertain reasoning warrant serious thought but not the kind of confident resource allocation that strong versions of expected-value reasoning appear to license. The structural principle: stay close to what you can trace, preserve optionality where you cannot, and do not let the grandiosity of speculative future stakes crowd out the structural work that is clearly traceable now.
Here, too, EA’s own evolution is instructive. The most credible longtermist work has increasingly converged on exactly this principle, emphasizing present-day, traceable existential risk factors over speculative scenarios. The community arrived at these positions through internal debate and hard experience. That the drift has been consistently toward what structural tracing would recommend — stay close to the traceable, invest where structural implications are clear — suggests that the framework is tracking something real, not imposing an arbitrary constraint.
Axioms
What follows is not a list of rules but a set of orientations — structural properties that any coherent practice of doing good would exhibit. They are derived from the process ontology, not stipulated.
1. Coherence Over Consequences
Ethics is not about producing the best outcomes. It is about participating coherently in the process networks one is embedded in. Outcomes matter as evidence of coherence or its absence, not as the thing being optimized. This subtle shift prevents the pathologies of pure consequentialism — the willingness to accept any means if the expected outcome is good enough — without retreating to rigid rules.
2. Structural Tracing Over Quantification
Trace what your action participates in. Quantify where the processes are measurable. But never mistake the metric for the reality it was built to track. When the metric and the structural picture diverge, trust the tracing. The metric has probably reified.
3. Epistemic Humility
Assess with confidence what is traceable. Flag as uncertain what requires long inferential chains. Refuse to extend confident-sounding analysis past the boundary of the traceable, even when the speculative case is emotionally compelling. This is not timidity. It is information integrity — the condition that every other process depends on.
4. Impartial yet Situated
Geographic, species, and temporal boundaries are often reified categories that no longer track structural differences. Dissolve them where they don’t track reality. But do not replace them with the “view from nowhere.” You are structurally embedded in particular networks — family, community, ecosystem — and your continuation conditions are more tightly coupled with some processes than others. Structural proximity is not bias. It is the topology of your actual situation. Begin with what you can trace — which is near — and extend outward with appropriate epistemic humility.
5. Suffering as Diagnostic
Suffering is contradiction registering from the inside. It points at where coherence has broken down. Take it seriously as information, not just as a quantity to minimize. A framework that minimizes suffering metrics while leaving the structural contradictions intact has optimized the dashboard, not the system.
6. Self-Undermining Is the Cardinal Structural Sin
A process that consumes its own substrate — whether an economy depleting its ecology, a movement corrupting its epistemic standards to attract resources, or an individual burning out in the name of impact — is structurally incoherent regardless of what it produces in the short term. Watch for it everywhere, especially in yourself and your own movement.
7. Prevention Over Remediation
Once a contradiction takes root, structures form around it and other processes come to depend on it. Preventing contradictions from arising, or catching them before they become entangled with other processes, is categorically more effective than resolving them once entrenched. EA’s emphasis on tractability should include tractability-through-prevention, not just tractability-of-existing-problems.
8. EA Is a Process, Too
The community that practices these principles is itself a self-sustaining process, subject to all the structural dynamics it analyzes in others. It can reify its own categories, consume its own epistemic substrate, suppress signals that threaten its continuation. No movement is exempt from its own analysis. The test: does the movement’s self-sustaining activity (fundraising, community building, status hierarchies) sustain or undermine the conditions for its actual purpose (structural improvement in the world)?
9. Disagreement Is Structural Information
Two people tracing in good faith may reach different conclusions. The disagreement is about the structure of the case, not about whether coherence matters. Disagreement within the movement is diagnostic information — it points to where the structural picture is complex or where someone’s categories may have hardened. Suppressing disagreement suppresses the signal. EA’s norm of openness to criticism is structurally correct; this principle grounds it in something stronger than a norm.
10. Guilt is Finitary / Love is Generative
EA is rather self-aware in its acknowledgment that people can never fully live up to the level of altruism it demands. Looking at this point structurally, it’s not a concession to human weakness. It is a recognition that a framework generating ambient sense of guilt has itself become a source of contradiction. If the practice of doing good systematically degrades the well-being of the practitioners, the practice is quietly self-undermining. Sustainable participation — pacing, rest, joy, relationships maintained for their own sake — is not a compromise with effectiveness. It is a condition of it.
What Changes in Practice
Cause Prioritization Gets Richer
Scale, neglectedness, and tractability remain useful heuristics — compressed structural insights. But they operate alongside a deeper question: does this intervention sustain or undermine the process networks it enters? A cause area can score high on all three EA criteria while participating in structural dynamics that undermine its own conditions. Consider the tension when AI safety research is funded by the very companies whose development of AI technology creates the risk — the funding sustains the research while the funding source accelerates the problem.
Structural tracing adds a fourth lens: processual coherence. Does this intervention’s entire causal chain — how it is funded, how it operates, what it depends on, what it participates in — sustain or undermine the conditions it needs?
Metrics Become Tools Of Coherence
Cost-per-DALY, QALYs, and expected-value estimates remain useful instruments. But they are no longer the final court of appeal. When a metric says an intervention is effective but structural tracing reveals it is participating in self-undermining dynamics, the tracing takes precedence. This prevents the slow reification of metrics into institutional gates that reshape the movement around measurability rather than coherence.
The Movement Watches Itself
The most important practical change is reflexivity. Every meeting, every funding decision, every hiring choice, every public communication is participation in process networks. Does the way the movement raises money sustain or undermine its epistemic integrity? Does the way it builds community sustain or undermine its members’ capacity for independent judgment? Does the way it communicates with the public sustain or undermine the information environment?
These questions are not added overhead. They are the primary diagnostic. A movement that cannot pass its own structural test has no business applying that test to the rest of the world.
Emotional Signals Get Taken Seriously
Guilt, burnout, insularity, and the feeling that one is never doing enough — these are not weaknesses to manage. They are structural signals pointing at contradictions within the practice itself. A framework that produces chronic guilt in its practitioners is generating contradiction, not resolving it. The signal says: something about how you are framing “doing good” is contradictory. Trace it. Revise the framing.
What This Is Not
This is not a call to abandon rigor. Structural tracing is more rigorous than cost-benefit analysis, not less — it demands tracing what an action actually participates in rather than reducing it to a number. It is harder, slower, and less satisfying than a spreadsheet. That is because reality is harder, slower, and less satisfying than a spreadsheet.
This is not a call to abandon global concern for local comfort. The structural picture genuinely extends outward — to distant populations, to other species, to future processes. What it does not do is extend with uniform confidence. It extends with appropriate epistemic humility, staying close to what is traceable and flagging where speculation begins.
This is not a call to abandon EA. It is a call to give EA the foundation it has always needed — one that grounds its best impulses (take doing good seriously, demand evidence, care about what actually works) while correcting its structural vulnerabilities (reification of metrics, speculative confidence, self-undermining institutional dynamics, guilt as a feature rather than a bug).
The goal is a practice of doing good that is structurally coherent all the way down — in its theory, its method, its institutions, and its relationship to its own practitioners. Not optimization. Coherence.
Doing good is not an output to maximize. It is a way of participating in the world that sustains what it depends on and doesn’t undermine what it enters. Get that right, and the bednets and the deworming and the AI safety work follow — not as conclusions from a calculation, but as obvious structural responses to visible need.
Structurally Effective Altruism
Structural Coherence as a Foundation for EA: A Process-Ethics Proposal
AI Disclosure: I wrote the theses/outline of the argument and had an LLM expand them into a polished prose, then edited lightly. This post is part of my ongoing research towards an ethical framework grounded in process ontology.
The Problem with Foundation
Effective Altruism has been one of the most intellectually serious attempts to answer the question: how do we actually help? It brought rigor to compassion, demanded evidence where sentiment had reigned, and insisted that good intentions are not enough — results matter. These contributions are genuine and lasting.
But EA has always had a foundation problem. Explicitly pragmatic and pluralistic, EA is compatible with a wide range of ethical theories. And yet, one could argue, this vagueness is indicative of a framework that never fully defined its own foundations. On one hand, EA tries to avoid extreme or counter-intuitive actions that might follow from strict cost-benefit analyses, while maintaining a commitment to doing the most good. On the other, EA’s core methodology — aggregate, compare, maximize impact per unit of resource — inherits utilitarianism’s machinery without resolving its deep structural questions — (what counts as a unit of good? why should units be fungible across contexts? what authorizes the aggregation?). EA compensated with pragmatism and empiricism — “do what works” — but pragmatism without ontology is navigation without a compass. It gets you somewhere. It cannot tell you whether you’ve drifted.
Objective Ethics — a process-based framework grounding value in the structural dynamics of self-sustaining processes — can supply what EA lacks: a non-arbitrary definition of what “good” actually consists of, a diagnostic method that works where metrics break down, and a self-correcting mechanism that applies to the movement itself.
Proposed Foundation: Everything That Persists Is a Process
The foundational claim is ontological, not normative. It does not begin with “you should.” It begins with “here is what is actually happening.”
Nothing that persists is static. A river is water flowing, eroding banks, carving the channel that concentrates the flow. A business provides value that generates revenue that funds the operations that produce the value. A person metabolizes, learns, relates, maintains the conditions of their own continuation through constant activity. Persistence is not a state. It is an achievement — a process sustaining its own conditions.
These processes do not exist in isolation. Every process depends on conditions maintained by other processes. The river depends on rainfall, the business on infrastructure and demand, the person on ecosystems, institutions, relationships. Processes exist in networks of mutual dependence.
When processes in these networks sustain each other’s continuation conditions, the result is stable configurations — ecosystems, economies, communities, lives. When they undermine each other’s conditions, the result is instability, breakdown, and suffering. This is not a value judgment imposed from outside. It is a description of what flourishing and suffering actually consist of.
This reframing changes what “doing good” means. It is not producing the best outputs. It is participating coherently in the networks of processes that one depends on and affects.
Defining “Good”
EA’s deepest vulnerability is that it cannot give a non-circular account of what it is trying to maximize. “Well-being” is typically defined by listing its components (health, happiness, preference satisfaction), but the list is stipulative — it encodes the intuitions of the list-maker. “Suffering reduction” fares better but remains negative: it tells you what to move away from without saying what you are moving toward.
Process ontology grounds value without stipulation.
Coherence is a configuration of processes that sustains itself without undermining what it depends on. This is not “harmony” in a sentimental sense. A predator-prey cycle involves conflict, but that conflict is part of how the system sustains itself. Coherence admits tension. What it excludes is self-undermining — a process consuming its own substrate, corrupting the conditions it needs, extracting without sustaining.
Harm is structural disruption: a process undermining the conditions that other processes depend on to continue. This definition requires no enumeration. It applies to any situation where processes interact, including situations nobody has previously categorized. Deception corrupts shared information that downstream processes depend on. Exploitation extracts from a process without sustaining what it needs. Coercion overrides another process’s capacity to respond to its own conditions. These are not three rules. They are three faces of the same structural dynamic.
Suffering is contradiction registering from the inside. Pain signals process disruption. Guilt signals that one is a source of contradiction. Shame signals that one’s own categories are inadequate to the situation. These signals are information — diagnostics pointing to where coherence has broken down.
Moral consideration extends to anything that can register contradiction — anything that can suffer. This grounds EA’s species impartiality without needing to argue for it: the boundary is structural, not taxonomic.
With this foundation, “doing good” is no longer an optimization target floating in definitional space. It is participation that sustains coherence across the process networks one is embedded in. It is concrete, traceable, and — crucially — it includes the process of doing good itself.
Refined Method: Structural Tracing Before Cost-Benefit
EA’s method is cost-benefit analysis scaled up: quantify expected impact, compare across options, allocate to the highest return. This works brilliantly within domains where the relevant variables are measurable and commensurable — distributing bednets, comparing deworming programs, estimating cost-per-DALY.
But quantification has its limits—useful when the relevant processes are measurable, it breaks down when they are not, the metric hardening into a gate that determines what counts as real. And that’s exactly where the most consequential decisions live.
The p-value didn’t destroy science by being a bad idea. It undermined it by becoming a reified category that researchers optimized around instead of through. Cost-per-DALY risks the same fate: a useful diagnostic that, once institutionalized, reshapes resource flows around what is measurable rather than what sustains coherence. Interventions that are hard to express in DALYs (institutional reform, cultural shifts, ecosystem health) start looking invisible — not because they’re less important, but because they’re less measurable.
Structural tracing can help us see “the invisible”. It is a single repeatable method with three steps:
What processes does this action affect? Not “what outcome does it produce?” but “what does it participate in?”
What do those processes depend on to sustain themselves?
Does this action sustain or undermine those conditions?
This method does not replace quantification. It grounds it. Quantification becomes a tool within structural tracing.
Applied: The Earning-to-Give Question
Earning to give — pursuing a high-income career to donate a significant portion — is one of EA’s more well-known ideas, and while over the years the emphasis increasingly shifted towards direct work, it remains a live concept within EA, and illustrates the difference between structural and consequentialist analysis.
A standard EA analysis asks whether a finance professional’s donations outweigh whatever harms their employer produces. Structural tracing asks a different question: what does the earning participate in?
If the fund profits from structures that extract from their own substrates — predatory lending that degrades borrower communities, attention-economy investments that corrode information integrity, commodity speculation that destabilizes food systems — then the earning is participation in structural harm. The donations flow through a different process network. They don’t cancel the harm because the harm and the help aren’t fungible — they live in different structural configurations. A river polluted upstream is not cleaned by a well dug downstream, even if the well helps more people than the pollution hurts.
This doesn’t mean earning-to-give is always wrong. It means the analysis must trace what the earning participates in, not just what the giving produces. A finance professional at a firm that allocates capital to productive enterprises and sustains market integrity is participating in a process that supports economic coherence. The earning and the giving are both structurally sound. The distinction is structural, not categorical, and it requires tracing rather than calculating.
It is worth noting that EA’s own trajectory confirms this structural insight. Over the past decade, the community has significantly de-emphasized earning-to-give in favor of direct work, now framing career impact primarily in terms of what one’s work participates in rather than what one’s donations produce. This evolution was driven by practical experience, not by process ontology — but the direction of the drift is telling. The community learned, through accumulated friction, what structural tracing would have diagnosed from the start: that what you participate in matters, not just what you fund.
Applied: The Longtermism Question
In its strongest form longtermist reasoning extends expected-value calculations across millennia: future people matter, there could be astronomically many of them, therefore even tiny probability reductions in existential risk dominate all near-term concerns. (There are weaker versions that merely claim future generations deserve moral consideration without this dominance claim.)
Structural tracing agrees that future processes matter — undermining conditions that future processes depend on is harm regardless of when those processes occur. Climate disruption, ecosystem collapse, and catastrophic AI risk are all cases of present actions degrading future continuation conditions. The concern is real.
But structural tracing imposes an honesty constraint that expected-value maximization does not. Tracing is reliable close to the action and grows speculative with distance. First-order effects are traceable. Second-order effects are somewhat traceable. Fifth-order effects across centuries are genuinely uncertain, and presenting them with the same confidence as near-term structural analysis is itself a form of information corruption — speculation dressed as rigor.
This does not mean abandoning long-term concern. It means calibrating confidence to traceability. Present actions with clear structural implications for future conditions (emissions, ecosystem destruction, AI development trajectories) warrant serious investment. Speculative scenarios requiring long chains of uncertain reasoning warrant serious thought but not the kind of confident resource allocation that strong versions of expected-value reasoning appear to license. The structural principle: stay close to what you can trace, preserve optionality where you cannot, and do not let the grandiosity of speculative future stakes crowd out the structural work that is clearly traceable now.
Here, too, EA’s own evolution is instructive. The most credible longtermist work has increasingly converged on exactly this principle, emphasizing present-day, traceable existential risk factors over speculative scenarios. The community arrived at these positions through internal debate and hard experience. That the drift has been consistently toward what structural tracing would recommend — stay close to the traceable, invest where structural implications are clear — suggests that the framework is tracking something real, not imposing an arbitrary constraint.
Axioms
What follows is not a list of rules but a set of orientations — structural properties that any coherent practice of doing good would exhibit. They are derived from the process ontology, not stipulated.
1. Coherence Over Consequences
Ethics is not about producing the best outcomes. It is about participating coherently in the process networks one is embedded in. Outcomes matter as evidence of coherence or its absence, not as the thing being optimized. This subtle shift prevents the pathologies of pure consequentialism — the willingness to accept any means if the expected outcome is good enough — without retreating to rigid rules.
2. Structural Tracing Over Quantification
Trace what your action participates in. Quantify where the processes are measurable. But never mistake the metric for the reality it was built to track. When the metric and the structural picture diverge, trust the tracing. The metric has probably reified.
3. Epistemic Humility
Assess with confidence what is traceable. Flag as uncertain what requires long inferential chains. Refuse to extend confident-sounding analysis past the boundary of the traceable, even when the speculative case is emotionally compelling. This is not timidity. It is information integrity — the condition that every other process depends on.
4. Impartial yet Situated
Geographic, species, and temporal boundaries are often reified categories that no longer track structural differences. Dissolve them where they don’t track reality. But do not replace them with the “view from nowhere.” You are structurally embedded in particular networks — family, community, ecosystem — and your continuation conditions are more tightly coupled with some processes than others. Structural proximity is not bias. It is the topology of your actual situation. Begin with what you can trace — which is near — and extend outward with appropriate epistemic humility.
5. Suffering as Diagnostic
Suffering is contradiction registering from the inside. It points at where coherence has broken down. Take it seriously as information, not just as a quantity to minimize. A framework that minimizes suffering metrics while leaving the structural contradictions intact has optimized the dashboard, not the system.
6. Self-Undermining Is the Cardinal Structural Sin
A process that consumes its own substrate — whether an economy depleting its ecology, a movement corrupting its epistemic standards to attract resources, or an individual burning out in the name of impact — is structurally incoherent regardless of what it produces in the short term. Watch for it everywhere, especially in yourself and your own movement.
7. Prevention Over Remediation
Once a contradiction takes root, structures form around it and other processes come to depend on it. Preventing contradictions from arising, or catching them before they become entangled with other processes, is categorically more effective than resolving them once entrenched. EA’s emphasis on tractability should include tractability-through-prevention, not just tractability-of-existing-problems.
8. EA Is a Process, Too
The community that practices these principles is itself a self-sustaining process, subject to all the structural dynamics it analyzes in others. It can reify its own categories, consume its own epistemic substrate, suppress signals that threaten its continuation. No movement is exempt from its own analysis. The test: does the movement’s self-sustaining activity (fundraising, community building, status hierarchies) sustain or undermine the conditions for its actual purpose (structural improvement in the world)?
9. Disagreement Is Structural Information
Two people tracing in good faith may reach different conclusions. The disagreement is about the structure of the case, not about whether coherence matters. Disagreement within the movement is diagnostic information — it points to where the structural picture is complex or where someone’s categories may have hardened. Suppressing disagreement suppresses the signal. EA’s norm of openness to criticism is structurally correct; this principle grounds it in something stronger than a norm.
10. Guilt is Finitary / Love is Generative
EA is rather self-aware in its acknowledgment that people can never fully live up to the level of altruism it demands. Looking at this point structurally, it’s not a concession to human weakness. It is a recognition that a framework generating ambient sense of guilt has itself become a source of contradiction. If the practice of doing good systematically degrades the well-being of the practitioners, the practice is quietly self-undermining. Sustainable participation — pacing, rest, joy, relationships maintained for their own sake — is not a compromise with effectiveness. It is a condition of it.
What Changes in Practice
Cause Prioritization Gets Richer
Scale, neglectedness, and tractability remain useful heuristics — compressed structural insights. But they operate alongside a deeper question: does this intervention sustain or undermine the process networks it enters? A cause area can score high on all three EA criteria while participating in structural dynamics that undermine its own conditions. Consider the tension when AI safety research is funded by the very companies whose development of AI technology creates the risk — the funding sustains the research while the funding source accelerates the problem.
Structural tracing adds a fourth lens: processual coherence. Does this intervention’s entire causal chain — how it is funded, how it operates, what it depends on, what it participates in — sustain or undermine the conditions it needs?
Metrics Become Tools Of Coherence
Cost-per-DALY, QALYs, and expected-value estimates remain useful instruments. But they are no longer the final court of appeal. When a metric says an intervention is effective but structural tracing reveals it is participating in self-undermining dynamics, the tracing takes precedence. This prevents the slow reification of metrics into institutional gates that reshape the movement around measurability rather than coherence.
The Movement Watches Itself
The most important practical change is reflexivity. Every meeting, every funding decision, every hiring choice, every public communication is participation in process networks. Does the way the movement raises money sustain or undermine its epistemic integrity? Does the way it builds community sustain or undermine its members’ capacity for independent judgment? Does the way it communicates with the public sustain or undermine the information environment?
These questions are not added overhead. They are the primary diagnostic. A movement that cannot pass its own structural test has no business applying that test to the rest of the world.
Emotional Signals Get Taken Seriously
Guilt, burnout, insularity, and the feeling that one is never doing enough — these are not weaknesses to manage. They are structural signals pointing at contradictions within the practice itself. A framework that produces chronic guilt in its practitioners is generating contradiction, not resolving it. The signal says: something about how you are framing “doing good” is contradictory. Trace it. Revise the framing.
What This Is Not
This is not a call to abandon rigor. Structural tracing is more rigorous than cost-benefit analysis, not less — it demands tracing what an action actually participates in rather than reducing it to a number. It is harder, slower, and less satisfying than a spreadsheet. That is because reality is harder, slower, and less satisfying than a spreadsheet.
This is not a call to abandon global concern for local comfort. The structural picture genuinely extends outward — to distant populations, to other species, to future processes. What it does not do is extend with uniform confidence. It extends with appropriate epistemic humility, staying close to what is traceable and flagging where speculation begins.
This is not a call to abandon EA. It is a call to give EA the foundation it has always needed — one that grounds its best impulses (take doing good seriously, demand evidence, care about what actually works) while correcting its structural vulnerabilities (reification of metrics, speculative confidence, self-undermining institutional dynamics, guilt as a feature rather than a bug).
The goal is a practice of doing good that is structurally coherent all the way down — in its theory, its method, its institutions, and its relationship to its own practitioners. Not optimization. Coherence.
Doing good is not an output to maximize. It is a way of participating in the world that sustains what it depends on and doesn’t undermine what it enters. Get that right, and the bednets and the deworming and the AI safety work follow — not as conclusions from a calculation, but as obvious structural responses to visible need.