Orientation · ~1–2 hours · no progress tracking
Working memory, ADHD, and why agreements between two people who both have it fail in a particular and predictable way. Written as background for a conversation with a trained professional, not as a substitute for one.
Every load-bearing claim carries one of these. They exist because the evidence here is genuinely uneven, and a brief that flattened it would be more persuasive and less true.
Three things are worth knowing before any of the content, because they change how much weight each section can carry.
Inference No one has run a study on working memory and boundary adherence, because boundaries aren't operationalized anywhere in the cognitive literature. The word belongs to clinical and practitioner vocabulary. Everything in this brief is therefore a translation: from constructs that are studied (working memory, inhibition, emotion regulation, delay aversion) to a construct that isn't. The translation is mine. Treat the mechanism sections as well-sourced and the bridges between them as argued.
Sourced The foundational working memory meta-analysis covers children and adolescents S2. So does most of the theory. Contested Adult studies find broadly similar impairments S3, but this brief applies a pediatric literature to adults in long partnerships, and that extrapolation is doing real work throughout. It is the single biggest structural weakness in this brief.
Vendor framing A large share of what comes up when you search ADHD and marriage is attached to eight-week seminars, coaching programs, and therapy practices. Some of it is good. All of it has an incentive to make the mechanism feel more settled and more solvable than the research supports. Inference This matters especially for rejection sensitive dysphoria, which is ubiquitous in that material and fits almost any situation suspiciously well, but which does not have a primary evidence base proportionate to its popularity. I left it out of an early draft rather than hand over a satisfying idea with nothing under it — and section 7 explains why that was half a mistake.
One more, worth naming because I nearly used it: I'm deliberately not building on self-control as a depletable resource. There is ADHD-specific work applying it to partner communication S11, but the underlying ego-depletion literature has replication problems severe enough that the conclusion would feel satisfying and rest on sand.
The everyday usage — "bad memory," forgetting things — is close to the opposite of the technical meaning. Working memory is not storage. It's the system that holds something active while you do something else. The standard model splits it three ways: a central executive that allocates attention, plus two short-term subsystems, one for verbal material and one for visual and spatial material S2.
Inference The relevant piece here is the central executive. A boundary, once agreed, is a rule that has to stay active during a competing demand. The demand is the whole point. A rule that only has to be recalled in calm conditions isn't a boundary; it's a preference.
Sourced Twenty-six studies, pooled. Children with ADHD showed deficits across multiple working memory components, independent of language learning disorders and of general intellectual ability S2. But the effects are not uniform, and the pattern is the interesting part:
| Component | Effect size | 95% CI |
|---|---|---|
| Spatial central executive | 1.06 | 0.72 – 1.39 |
| Spatial storage | 0.85 | 0.62 – 1.08 |
| Verbal storage | 0.47 | 0.36 – 0.59 |
| Verbal central executive | 0.43 | 0.24 – 0.62 |
Martinussen et al. 2005 S2. Children and adolescents. Assume the true effects are somewhat smaller than published.
Hold onto the bottom row. It comes back in section 4 and it does not say what a tidy version of this brief would want it to say.
A three-tier taxonomy — individual, relational, meta — sorts the mechanisms cleanly, which is a good sign for the taxonomy. It is a practitioner's cut rather than a research one; section 7 compares it to the taxonomies the mainstream boundaries books use. Each tier makes a different cognitive demand, so each fails differently.
Tier 1 · Individual boundaries
Enforceable over yourself, your possessions, and access to those. Inference Cognitively this is the easiest tier and the one least implicated in the problem couples usually describe. It requires holding one rule against one impulse, with no second person's state to track. When individual boundaries fail in ADHD it's usually a straightforward inhibition or delay story — and there is a serious model saying exactly that, which I'll get to.
Tier 2 · Relational boundaries (contracts)
Require shared agreement, set precedent case by case. Inference This is where most of the damage happens, and it's a dual-load problem. You have to hold your own position active while simultaneously processing what the other person is saying, under affective arousal. That's not one working memory demand, it's two competing for the same central executive, in the exact condition that degrades it.
Sourced And the affective part isn't incidental. Emotion dysregulation is prevalent in ADHD across the lifespan and is a major contributor to impairment — not a side effect but a primary source of the damage. It appears to arise partly from difficulty orienting toward, recognizing, and allocating attention to emotional stimuli, implicating a striato-amygdalo-medial prefrontal network S4.
Contested How emotion dysregulation relates to ADHD is unsettled. The same review lays out three live models: correlated but distinct dimensions, a core diagnostic feature, or the combination forming its own entity S4. This matters practically. If it's a core feature, you work on it as ADHD. If it's a distinct dimension, it needs its own handle.
Tier 3 · Meta boundaries
Rules about how you make rules. Inference The hardest tier, and the one with the longest gap between agreement and application. A meta boundary is agreed in a calm moment and has to govern behavior in a hot one, possibly weeks later. Nothing about the moment of agreement is present at the moment of use. Every property that makes it powerful also makes it fragile.
Sourced Barkley's 1997 model proposes that ADHD is fundamentally a deficit in behavioral inhibition, which then degrades four executive functions that depend on it: working memory, self-regulation of affect and motivation, internalization of speech, and reconstitution S1.
Internalization of speech is the mechanism by which spoken agreements become internal rules that govern later behavior. Inference If you wanted a cognitive theory purpose-built to explain why meta boundaries don't stick, you would build that one. The fit is close enough that I want to be suspicious of it, which brings us to the next part.
Contested The 1997 paper is among the most cited in the field, and citation count is popularity, not truth. A 2022 critique argues that Barkley's account of behavioral inhibition is unclear about where and what the deficit actually is, treating it as something that exists independently somewhere in prefrontal cortex, and reports that scrutiny of his supporting literature found only twelve cited studies bearing on impulsivity at all S5.
Sourced The field has also moved. Dual-pathway models propose two dissociable routes — executive dysfunction and delay aversion, the latter being a motivational disposition to escape the negative affect that delay itself evokes — with the explicit claim that each affects some patients and not others S6. Later work adds temporal processing as a third dissociable component S7. Contested Even delay aversion doesn't replicate cleanly; some studies find group differences and others don't S8.
Vendor framing Worth noting on the triple-pathway paper: the lead author disclosed serving on a speakers' bureau and as a consultant for a pharmaceutical company S7. That doesn't invalidate the work. It's the kind of thing you'd want flagged rather than discovered.
Net: use Barkley for vocabulary and structure. Don't let any conclusion rest on him alone. The heterogeneity finding is arguably more useful to you anyway — it means "we both have ADHD" does not imply two partners have the same deficits.
Here is where the research declines to cooperate, and I think this is the most valuable thing in the brief.
The tidy story would be: verbal agreements don't stick because verbal working memory is impaired. It's clean, it maps onto internalization of speech, and it would explain everything couples describe.
Sourced But verbal central executive working memory showed the smallest effect of the four components measured — 0.43, against 1.06 for its spatial counterpart S2. The verbal deficit is real, but it is the weakest link in the chain, not the strongest.
If the problem were simply that ADHD brains can't hold verbal rules, the effect sizes would point the other way. Something else is doing most of the work.
Inference My reading, offered as argument rather than finding: the failure is probably not in storing the rule but in accessing it under affective load. That's a different claim, and it points at the emotion dysregulation evidence S4 rather than the working memory evidence as the primary driver. Working memory sets the ceiling. Emotional arousal is what drives you into it.
This has a practical consequence. If the rule is intact but unreachable when it's hot, then interventions that strengthen the rule — repetition, clearer wording, more emphatic agreement — are aimed at the wrong variable. Interventions that reduce arousal, or that make the rule available without requiring retrieval, are aimed at the right one.
I went looking for research on dyads where both partners have ADHD, expecting to find nothing. Close to right, but with one genuinely useful finding along the way.
Sourced Assortative mating — people selecting partners with similar traits — is documented across the full spectrum of mental disorders in population-scale registry data, including a Finnish study covering over 1.2 million partnerships S9. Earlier Swedish and Norwegian registry work found ADHD among the strongest within-disorder partner correlations S9.
Contested Though not settled: a large cross-country analysis found ADHD showed wide swings that were not significant after strict statistical correction S12. And partner similarity has at least three possible causes — active preference for similarity, convergence over years of shared life, or social homogamy from shared environments — which registry data struggles to separate S9.
Single study Steele, Wymbs and Capps compared 94 adults without ADHD, 43 with childhood histories but no current symptoms, and 27 with childhood histories and elevated current symptoms. Assortative mating was notably common in the persisting group S10. Inference Note the limitation, which is significant: partners' symptoms were rated by the participant, not independently assessed. A person with ADHD reporting that their partner also has it is not the same as two independent assessments.
Inference The main marital studies use one-partner-with-ADHD designs — 33 married adults and their spouses in the earliest S13, 28 couples in a more recent one S14. The literature is largely built around an ADHD partner and a non-ADHD partner who compensates. Two-ADHD partnerships are barely studied. That is worth raising with a professional explicitly rather than letting it pass silently, because much of the accessible advice assumes a compensating partner who, in this configuration, doesn't exist.
Single study In that earliest marital study, adults with ADHD rated the health of their marriages and families more negatively than their spouses did S13.
Inference Small sample, one study, don't over-read it. But if it generalizes even slightly, it complicates the assumption that the partner with ADHD is the one underestimating the damage. It may run the other way. That's worth holding lightly and raising in a session rather than concluding.
If you've tried timers, cues, scheduled conversations, and four or five well-built communication frameworks, the futility feeling is information, not failure of effort.
Inference Here's the pattern I'd argue for. Nearly every one of those interventions shares a hidden requirement: at the moment of highest emotional load, someone has to remember to deploy the technique. Radical Candor requires recalling the framework mid-conflict. NVC requires assembling a four-part structure under arousal. A timer requires someone to have set it while calm and both parties to respect it while not. Each one is a rule that must be retrieved precisely when retrieval is most degraded.
The design principle that follows is narrow but, I think, load-bearing:
Interventions that require retrieval under load are asking the deficit to fix itself. Interventions that make the rule present — so retrieval isn't required — route around it.
Inference Externalization is the general form. A written document is present. A physical object is present. A third party is present. None of these ask working memory to do anything. This is also, notably, what a good deal of clinical practice already does without naming it this way: a written agreement, a standing session structure, a rule on paper that either partner can point at are all externalizations. So is the professional in the room, who is a third party and therefore present.
Inference The honest caveat: I could not find intervention trials testing this in ADHD couples. The reviews say the same — clinical trials testing interventions for romantic relationship functioning in adult ADHD are a recognized gap S15. So this is a mechanism-derived prediction, not a validated technique. Bring it as a hypothesis.
A different field type, and it needs saying plainly: this section is craft, not evidence. Books and blogs by clinicians and consultants are accumulated practice wisdom. Some of it is excellent and some of it is confident invention, and from the outside they read identically. The useful ones here are the ones whose claims you can trace back to something in sections 2 through 6.
Sourced fact S16 John Gottman's decades of observational work on married couples produced a construct he calls flooding, or diffuse physiological arousal. Heart rate climbing past roughly 100 bpm, sympathetic activation, perceptual narrowing. His claim is that a flooded person becomes physiologically incapable of the listening and problem-solving that conflict resolution requires S20, and that flooding is the engine driving couples down what he calls the distance and isolation cascade S16.
This is a completely separate research tradition — marital observation and physiology, no ADHD in the frame — arriving at the same conclusion as section 4.
Inference That convergence is the most valuable thing I found. The ADHD literature says emotion dysregulation is a major source of impairment and involves difficulty allocating attention under emotional load S4. Gottman, working on ordinary couples with entirely different methods, says the same thing in physiological terms: past a threshold of arousal, the capacity for the conversation is simply gone. Two unrelated literatures converging is stronger evidence than either one alone, and it means the mechanism I argued for isn't ADHD-specific. ADHD probably lowers the threshold rather than creating the phenomenon.
Sourced fact S20 It also yields the one intervention in this entire brief that isn't a retrieval-under-load protocol. Heart rate returns to baseline in roughly twenty minutes — but only if the person actively distracts themselves. Spend the break rehearsing your case or replaying the argument and the arousal doesn't drop S20. Inference Note why this fits the design principle from section 6: a time-out agreed in advance requires one move at the moment of load, not the recall and execution of a framework. It's physiological rather than cognitive. It routes around the deficit instead of asking it to cooperate.
Contested The famous accuracy figures — 94%, "predicts divorce from the first three minutes" — are the weakest part of his work and the part most repeated. The standard methodological critique is Heyman and Smith Slep's The hazards of predicting divorce without cross-validation S17, and the objection is that a model fitted to a sample and evaluated on that same sample is describing, not predicting.
Vendor framing The Gottman Institute's own research FAQ concedes the 94% claim has become a source of confusion, and traces it to a single 1992 study S18. Worth knowing that the institute sells therapy training, certification, and consumer programs.
Net: the flooding physiology is well-observed and directly useful. The prediction statistics are where the field pushes back. Take the first, leave the second.
Vendor framing Almost everything here is written by someone who also sells a program, and I'd rather say that once, up front, than repeat it per entry.
Inference One observation here is worth more than the reading list, so I'll lead with it. The mainstream boundaries books categorize by domain. The three tiers in section 3 categorize by enforceability.
Sourced fact S25 Nedra Glover Tawwab's Set Boundaries, Find Peace — probably the most-recommended current title — sorts boundaries into six types: physical, sexual, intellectual, emotional, material, and time S25. Henry Cloud and John Townsend's Boundaries is the enduring classic in the category, written from an explicitly Christian frame S24.
Inference The enforceability cut — individual, relational, meta — is organized by who has to agree for the rule to hold. I think that's the better cut for a marriage. Domain taxonomies answer "what kind of boundary is this," which is rarely the hard question. Enforceability answers "can I hold this alone, or does it require a negotiation," which is precisely where the cognitive load lands. The enforceability cut sorts the mechanisms; the domain cut doesn't. That's a real point of overlap to raise in a session, and a question worth asking whoever you're working with: which taxonomy do they use, and what does it buy them?
An earlier draft of this brief excluded rejection sensitive dysphoria in one line, on the grounds that it was everywhere in commercial ADHD content and absent from the primary literature. That call was half right, and the half that was wrong cost something specific.
The half that was right. Vendor framing RSD is not a diagnosis. It does not appear in DSM-5-TR, has no validated diagnostic criteria, no validated instrument, and no peer-reviewed prevalence estimate S26. The term was coined and popularized by psychiatrist William Dodson, emerged from clinical observation rather than a research programme, and spread through magazine coverage and social media — which is why it is now far better known than the evidence behind it S26. Single study Dodson and colleagues published a small case series in 2024 arguing the pattern isn't explained by depression, anxiety, or a personality disorder S27, and there is qualitative work on lived experience S28. A case series and some qualitative work is early-stage evidence, not a settled construct. Inference So any percentage you see attached to RSD should be treated as invented until sourced.
The half that was wrong, and it matters. Underneath the branded term sits a properly researched construct with a thirty-year literature, and by excluding the label I nearly cost you the thing underneath it.
Sourced fact S29 Rejection sensitivity is defined as a cognitive-affective disposition to anxiously expect, readily perceive, and intensely react to rejection. The foundational paper is four studies in the Journal of Personality and Social Psychology, and it is specifically about intimate relationships S29.
Study 2 provides experimental evidence that people who anxiously expect rejection readily perceive intentional rejection in the ambiguous behavior of others. Study 3 shows the same thing inside romantic relationships: they readily perceive intentional rejection in a partner's insensitive behavior. Study 4 finds that rejection-sensitive people and their partners are dissatisfied with the relationship S29.
Inference I'd read Study 3 twice. Not because it describes you — one paper cannot do that — but because the construct it measures is the perception of intent in ambiguous behavior, which is the precise place a conversation is most likely to fracture. It also suggests that "did you mean it that way" may be a harder question for some people to answer accurately than it feels from the inside, in either direction.
Contested A 2023 systematic review and meta-analysis confirms rejection sensitivity has real associations with romantic relationship outcomes, while noting the usual caveats about publication bias in this literature S30.
Inference Two things follow. First, this is the third independent construct in this brief — alongside emotion dysregulation and flooding — that is not ADHD-specific. Rejection sensitivity is a general disposition studied in the general population. That keeps reinforcing the same conclusion: ADHD is probably modulating shared human machinery rather than producing an exotic mechanism of its own.
Second, a note on my own error, since you may want to apply the same test to anything else here. I excluded RSD because the label was commercial. That was a reasonable filter and it misfired, because a weakly-evidenced popular term had a well-evidenced academic construct sitting directly underneath it. The right move with vendor-framed material is to trace it down a level, not to discard it. Discarding is faster and it silently loses things. When a clinician, a book, or a search result raises a concept that pattern-matches to self-help, that's the question to ask: what is this a popular name for, and who studied that?
The question this brief was built to answer: how does working memory in ADHD brains affect boundary-setting and adherence? Here is the compressed version, in a form you can read aloud at the start of a session.
Short answer: less than I expected, and less directly. Working memory sets a ceiling on how much can be held active during conflict, but the verbal component — the one that would most obviously explain agreements failing — shows the smallest measured deficit of the four. So the story isn't that we can't hold the rule.
The better-supported story is that emotion dysregulation is doing most of the work. It's prevalent in ADHD across the lifespan, it's a major source of impairment in its own right, and it involves difficulty allocating attention to emotional material. The rule is probably intact. It's unreachable while arousal is high.
Which sorts onto the three tiers: individual boundaries are least affected. Relational contracts are worst hit, because they demand holding my position while processing my partner's, under arousal — two loads on one system. Meta boundaries are structurally fragile because nothing about the calm moment of agreement is present in the hot moment of use.
What that implies: every protocol we've tried requires retrieving a technique at peak load. That's asking the deficit to repair itself. Externalizing the rule so it's present rather than remembered is the move that routes around it — a rule written down, or a third party in the room, is present without anyone having to retrieve it.
The part I'd most want your read on: Gottman's flooding work gets to the same place from a completely different direction — ordinary couples, physiological measurement, no ADHD involved. Past about 100 bpm the capacity for the conversation is just gone. If two unrelated literatures converge there, then ADHD probably lowers the threshold rather than creating the problem. And his twenty-minute time-out is the only intervention I found that doesn't require remembering a technique mid-conflict.
One thing I want to bring you rather than conclude: there's a construct called rejection sensitivity — measured since 1996, not ADHD-specific — where the finding is that people high in it readily perceive intentional rejection in ambiguous or insensitive behavior. Given that the word "intention" is usually exactly where we get stuck, that seems worth looking at together.
Two caveats I don't want to skip. Most of this evidence is pediatric, and I'm extrapolating to two adults in a long marriage. And the couples research almost entirely studies one partner with ADHD and one without. Our configuration is barely studied, which means a lot of standard advice assumes a compensating partner we don't have.
These are the decisions the map doesn't make for you, and they are the ones worth spending session time on rather than re-deriving alone.
Not "do I understand ADHD and boundaries." The decision is: does the externalization principle deserve the next several weeks of effort, or is it protocol number nine wearing a better hat?
The case for: it's mechanism-derived rather than borrowed, it explains the prior failures rather than ignoring them, and it converges with what practitioners reach for independently.
The case against: it is untested in this population, the mechanism story rests on an inference I made rather than a finding I found, and the pediatric-to-adult extrapolation underneath it is real.
S1, S2, S4, S5, S6, S7, S8, S9, S10, S12 were located and their claims confirmed against abstracts or full text. S3 was not opened and is cited secondhand — it is the weakest link, and it happens to be the one carrying the child-to-adult extrapolation. S13, S14 and S15 were confirmed to exist with the sample sizes stated; I read abstracts, not full text. In the practitioner section, S16 and S18 were read directly. S17 and S20 were not opened — S17 is cited from a reference list, and S20 is a secondary summary carrying the 20-minute figure. The books, S19 and S21–S25, are craft sources: verified to exist as described, not evaluated for accuracy. On rejection sensitivity: S29 and S30 were located and their abstracts confirmed; S27 and S28 were not opened, and S28 has an unresolved publication year. S26 is a set of agreeing secondary summaries rather than a primary review — the DSM status is not in dispute, but treat the "no validated instrument" claim as well-attested rather than independently verified by me.
Assume the empirical effects are smaller than published. Roughly a third of psychology findings replicate significantly in the same direction.