Model, Care, Execution
A three-part framework for diagnosing interpersonal harm: was it a wrong mental model, misaligned priorities, or execution failure? Each requires a completely different solution.
Read OriginalMy Notes (6)
Our capacity for harm is part and parcel of our humanity.
If you never encountered any Model, Care, or Execution failures, you would be God, and none of us are.
Remedies
Model problems?
- supply better information.
- interpersonal level: make the other person's preferences legible in advance, establishing standing defaults
- institutional level: the analogue is blunt explicit feedback
- premise: someone acting on a bad model is acting reasonably, so information rather than sanction is the lever
Care problems
- change the incentives or the relationship
- make the desired behavior cheaper or more rewarding
- ask directly to be prioritized higher
- if nothing shifts, the last remedy is to stop relying on that person
Execution problems
- treatment and structural workarounds
- find what actually works mechanically
- remove the failure point
- treat the underlying condition
Generic fallbacks when you're not confident in any of the three
- outsource the decision to someone with better models, care, or execution than you
- fall back on asking directly
- be conservative when errors are asymmetric, skew toward the cheaper mistake
- postmortem afterward: own the harm non-defensively, and on the other side, forgive
- without a culture that tolerates some mistakes, everyone plays defense and the whole framework stops producing good expected-value decisions
Fostering a culture of forgiveness with room for some mistakes makes MCE work way, way better. It allows us to make the best decisions in expectation for one another without having to play defense for every choice we make
Any failed action decomposes into didn't-know, didn't-want, or couldn't; belief, desire, and ability: the standard folk-psychological breakdown of why anyone does anything
When Alice harms Bob, it is likely that one of the following three things went wrong:
- Model: Alice’s mental picture of what Bob would want, or of what is going on in the situation (which we’ll call her model), was incorrect and led her to behave in ways that caused Bob harm. If she’d had a better model, she would have acted differently, resulting in a better outcome for Bob.
- Care prioritization: Alice prioritized something else—herself, another person, some abstract value—over Bob. Even if she cares a lot about Bob’s happiness, other factors outweighed her care for him in this situation.
- Execution: Alice knew what Bob would want and cared enough about him, but failed to execute on bringing about the best Bob-outcome. Her execution failure could be the result of physical limitations, impulsivity, depression, structural barriers, or any of a number of other constraints.
Identifying whether a problem is Model, Execution or Care prioritization
| How do you feel about your actions? | Endorsed while acting | Unendorsed while acting |
|---|---|---|
| Endorsed reflectively | Care prioritization | Your behavior is deeply confusing and you are beyond our help |
| Unendorsed reflectively | Model | Execution |
model failures are ones you endorse in the moment but not in hindsight, care failures you endorse at both times, execution failures at neither
Summary used for search
TLDR
• When Alice harms Bob, it's always one of three things: wrong model (she misunderstood what he wanted), care mismatch (she prioritized something else), or execution failure (she knew and cared but couldn't pull it off)
• The "endorsement test" reveals which: model failures you endorse during but not after; care mismatches you endorse during and after; execution failures you don't endorse even while doing them
• Consent is often insufficient—MCE is how healthy relationships actually operate, especially long-term ones where you know someone well enough to act on your model rather than asking permission every time
• The solution depends entirely on the failure mode: model problems need better information, care problems need incentive changes or priority shifts, execution problems need structural fixes or treatment
• If we never had MCE failures, we'd be God (omniscient, omnibenevolent, omnipotent)—our capacity for harm is inherent to being human
In Detail
The Model-Care-Execution framework diagnoses three distinct ways interpersonal harm occurs. Model failures happen when you act on an incorrect mental picture of what someone wants—you'd have acted differently with better information. Care mismatches occur when you prioritize something else over the person, even if you generally care about them. Execution failures happen when you know what they want and care enough, but fail to deliver due to physical limitations, impulsivity, structural barriers, or other constraints. The key insight is that each requires a completely different intervention: model problems need information sharing and better defaults, care problems need incentive changes or explicit priority negotiations, and execution problems need structural fixes or capability improvements.
The authors introduce an "endorsement test" to distinguish between failure modes. With model failures, you endorse your actions while taking them but wish you'd acted differently once you have full information. With care mismatches, you endorse your actions during and often still endorse them after—you made the right choice given your values, even if it hurt someone. Execution failures are when you don't endorse your actions even while taking them. This distinction matters because it determines whether the solution is more information, different priorities, or better systems.
The framework challenges the primacy of consent in relationships. While consent is the right default for low-trust situations (like college hookups), most healthy relationships actually operate on MCE—you act based on your model of what someone wants, weighted by how much you care, executed to the best of your ability. This is especially true in long-term relationships where constantly asking permission would be absurd. The authors argue this isn't just a variant of consent but a fundamentally different (and often better) way of navigating high-trust relationships. They connect this to theodicy: if we had perfect models (omniscience), perfect care (omnibenevolence), and perfect execution (omnipotence), we'd be God. Our capacity for harm is inherent to being human, and the framework helps us systematically reduce it.