diff --git a/concepts/design-grammar.md b/concepts/design-grammar.md
index c82ef8a..06f61a6 100644
--- a/concepts/design-grammar.md
+++ b/concepts/design-grammar.md
@@ -32,6 +32,7 @@ This wording is a proposed synthesis, not an approved project slogan. Coherence
 | Keep authority scoped and contestable | Who may act for whom, on what basis, and who else bears the effects? | [Agency and Automation](agency-and-automation.md) and [Idea Trail 9](../IDEA_TRAILS.md#trail-9--intelligence-automation-and-legitimate-authority) |
 | Prefer interoperable paths where they preserve autonomy | Can people change implementations, exchange information, and leave without losing the underlying capability? What coordination costs arise? | [Legible Systems](legible-systems.md#design-principles) |
 | Preserve essential capabilities through failure | What survives a broken component, missing network, lost maintainer, or deliberate refusal of the preferred path? | [Resilience and Graceful Degradation](../systems/resilience-and-graceful-degradation.md) |
+| Preserve usable agency under reduced capacity | When stress, fear, exhaustion, injury, overload, or uncertainty narrow capacity, can the system reduce demands while preserving orientation, refusal, help, alternatives, and recovery? | [Agency under stress](../research/agency-under-stress.md) and [Resilience and Graceful Degradation](../systems/resilience-and-graceful-degradation.md) |
 | Make important behavior legible | Can affected people understand an action, its authority, limits, and avenues for challenge without first becoming specialists? | [Legible Systems](legible-systems.md) |
 | Preserve provenance without indiscriminate retention | Can a claim, change, decision, or generated contribution be traced to its basis while respecting privacy and forgetting? | [Thought Routing](../THOUGHT_ROUTING.md) and [Idea Trail 11](../IDEA_TRAILS.md#trail-11--memory-provenance-and-continuity) |
 | Make consent contextual | Which use, audience, duration, data, and consequence were actually authorized? How are withdrawal and other affected parties handled? | [Agency and Automation](agency-and-automation.md#delegation-and-affected-parties) |
@@ -71,4 +72,4 @@ See [Idea Trails](../IDEA_TRAILS.md) for broader routes. These examples are **pr
 
 Use [Thought Routing](../THOUGHT_ROUTING.md) to capture the source once, compare existing treatments, and choose a canonical home. A useful path may be **capture → investigate → synthesize → make a project-specific decision → test or observe → revise**. It is not a required promotion ladder: an idea can stay a question, split, be contested, or be retired. Research can challenge a principle; a principle does not certify a research finding.
 
-**Source note:** This synthesis was prompted by the 2026-09-24 discussion of recurring project patterns, agency/automation, and the later dynamic-coherence/adaptive-continuity discussion. A 2026-09-28 synthesis/connection discussion sharpened the treatment of relationships as first-class objects and synthesis as distinct from replacement expertise. A 2026-09-29 Existential Euphoria / Root Sequence comparison added the provisional Membrane Test: a boundary-and-interface prompt for relationship without erasure. The conversation text was available during drafting, but no durable conversation URL was supplied. The table and wording are AI-assisted proposals for author review.
+**Source note:** This synthesis was prompted by the 2026-09-24 discussion of recurring project patterns, agency/automation, and the later dynamic-coherence/adaptive-continuity discussion. A 2026-09-28 synthesis/connection discussion sharpened the treatment of relationships as first-class objects and synthesis as distinct from replacement expertise. A 2026-09-29 Existential Euphoria / Root Sequence comparison added the provisional Membrane Test: a boundary-and-interface prompt for relationship without erasure. A 2026-10-05 agency-under-stress synthesis added the question of whether systems preserve usable agency when human capacity is temporarily reduced. The conversation text was available during drafting, but no durable conversation URL was supplied. The table and wording are AI-assisted proposals for author review.
diff --git a/research/README.md b/research/README.md
index 20469d7..35a4a93 100644
--- a/research/README.md
+++ b/research/README.md
@@ -76,6 +76,7 @@ This directory holds bounded research inquiries and versioned papers that synthe
 
 ## Cross-cutting related work
 
+- [Agency under stress](agency-under-stress.md) — **AI-assisted working synthesis; author review pending; not clinical guidance or an interrogation manual**. Connects controllability, coercion, decentering, psychosocial support, recovery, and human-facing system design while preserving evidence limits.
 - [Adaptive continuity / dynamic coherence — intellectual lineage and prior art](dynamic-coherence-lineage.md) — **targeted related-work note; not a novelty finding or systematic review**. Traces state/parameter change, ultrastability, requisite variety, double-loop learning, resilience engineering, sustained adaptability, adaptive management/governance, path dependence, and second-order cybernetics; also records prior uses of the phrase “dynamic coherence.”
 - [Reading Trails](reading-trails/README.md) — **curated brain-food maps**, separate from claim-level bibliographies: archives, societies, courses, old magazines, reading lists, practitioner hubs, and other places worth wandering. The first trail covers [systems, cybernetics, complexity, and adaptive systems](reading-trails/systems-cybernetics-complexity.md).
 
@@ -91,7 +92,6 @@ These links point to existing public draft PRs, not merged research or reviewed
 
 - [The Stoned Age, PR #45](https://github.com/Root-Sequence/root-sequence/pull/45): source register, claims ledger, appraisal notes, and linked paper/booklet/essay/talk. Scholarly appraisal and author review remain incomplete.
 - [Intelligence without omniscience, PR #44](https://github.com/Root-Sequence/root-sequence/pull/44): developing research on distributed knowledge, recognition, trust, and enabling conditions.
-- [Agency under stress, PR #42](https://github.com/Root-Sequence/root-sequence/pull/42): cross-domain synthesis with explicit evidence and domain limits.
 
 ## Publication boundary
 
diff --git a/research/agency-under-stress.md b/research/agency-under-stress.md
new file mode 100644
index 0000000..9725b4c
--- /dev/null
+++ b/research/agency-under-stress.md
@@ -0,0 +1,323 @@
+# Agency under stress
+
+**Document role:** Bounded cross-domain research synthesis and design-input note<br>
+**Status:** AI-assisted working synthesis for author review; first drafted 2026-10-05<br>
+**Scope:** Human stress, coercion, controllability, cognition, recovery, and the design of systems people must use under reduced capacity<br>
+**Evidence boundary:** This note is not a clinical model, diagnosis, treatment guide, interrogation manual, or claim that one mechanism explains trauma or resilience. It connects several research literatures while preserving their differences.
+
+## Why this belongs in Root Sequence
+
+A recurring systems question is:
+
+> **What remains possible when a participant's capacity narrows under stress?**
+
+The same question appears in psychology, trauma research, interrogation research, human factors, emergency response, accessibility, community design, and resilience engineering.
+
+The useful synthesis is not that human beings behave like machines. It is narrower:
+
+- people can lose access to some capabilities without losing all agency;
+- stress can change what actions are cognitively, emotionally, socially, or materially reachable;
+- external conditions can either narrow or reopen those possibilities;
+- recovery does not necessarily mean restoration to a prior state;
+- systems can be designed so reduced capacity does not automatically become loss of dignity, consent, participation, or control.
+
+A compact design translation is:
+
+> **A resilient human-facing system should shed complexity before it sheds meaningful agency.**
+
+That statement is normative. The research below helps explain why it may be a useful design constraint.
+
+## 1. "Breaking" is not one psychological state
+
+Everyday language often collapses several different outcomes into one word:
+
+- behavioral compliance;
+- despair or passivity;
+- panic or emotional dysregulation;
+- cognitive disorganization;
+- memory impairment;
+- dissociation or detachment;
+- altered trust and threat sensitivity;
+- exhaustion;
+- changed beliefs;
+- truthful disclosure.
+
+These outcomes can overlap, but they are not interchangeable.
+
+A person may comply without agreeing. A person may say something without believing it. A person may experience severe distress without losing values or identity. A person may retain internal resistance while losing the practical ability to refuse. A person may recover function without returning to an earlier self.
+
+This distinction matters anywhere a system interprets outward behavior as evidence of consent, belief, truth, competence, or legitimacy.
+
+## 2. Responses to severe stress vary
+
+The 2022 [Istanbul Protocol](#institutional-sources) describes varied psychological responses to torture and ill-treatment. Chapter VI, especially paragraphs 491–498, asks evaluators to consider culture and context rather than reducing suffering to a psychiatric classification. It also describes symptoms changing over time with coping and external conditions.
+
+This supports caution about treating capacity as one fixed threshold. It does not establish a general law about every kind of stress, or prove that no threshold model could ever be useful.
+
+Root Sequence translates that caution into a design question:
+
+> **Do not model human capacity as one fixed threshold.**
+
+For design purposes, consider that a person may handle one demand while struggling with another, or need a different route at a different time. This is a reason to provide alternatives, not a way to infer someone's decision-making capacity from their distress.
+
+## 3. Coercion can produce compliance without producing reliable truth
+
+The U.S. Senate Select Committee on Intelligence's study of the CIA detention and interrogation program concluded that the CIA's enhanced interrogation techniques were not an effective means of acquiring intelligence or gaining cooperation as represented by the agency. Finding 1 also records fabricated information from detainees during or after those techniques.
+
+The FBI-administered High-Value Detainee Interrogation Group later synthesized interrogation science and emphasized evidence-based, information-gathering approaches rather than treating coercion as a reliable route to truth.
+
+A 2024 [systematic review by Catlin and colleagues](#research-syntheses) separately examined experiments with mock suspects, mostly U.S. college students. Its three-category analysis found more false confessions under accusatorial approaches than under information-gathering approaches. The difference in true confessions between those two approaches was not statistically significant. These experiments do not directly test torture or ordinary workplace interactions.
+
+Root Sequence draws a distinction for system design:
+
+> **Compliance, cooperation, belief change, and information quality are separate variables.**
+
+A system that is optimized for visible compliance can become worse at learning what is actually true.
+
+Beyond interrogation, this becomes a question to investigate rather than an established transfer of those findings: what does outward agreement mean when refusal is costly or unavailable? Compliance alone should not be treated as proof of meaningful consent or legitimacy.
+
+## 4. Controllability matters, but "control" is not magic
+
+Research on stressor controllability grew from the learned-helplessness literature. [Baratta, Seligman, and Maier (2023)](#research-syntheses) review animal experiments in which learning control over an aversive event can engage neural pathways that buffer later stress effects. Protection depends on how control is learned; merely performing an escape response is not always protective.
+
+The strongest experimental neural evidence is largely from rodents, so direct translation to ordinary human life requires caution. Human control is also social, cognitive, institutional, and material rather than one laboratory response.
+
+Still, the broader question is useful:
+
+> **Can any available action meaningfully change what happens next?**
+
+For human-facing systems, perceived control should not be manufactured through meaningless choice. The relevant target is **real, legible, proportionate agency**:
+
+- a refusal that is actually honored;
+- an exit that is feasible;
+- a pause that does not trigger punishment;
+- a helper who can be contacted;
+- an alternative route that still works;
+- a correction process that can change the outcome;
+- an explanation that makes the situation more navigable.
+
+The inverse also matters: a system can present many buttons while leaving consequential outcomes uncontrollable.
+
+## 5. Stress can narrow the reachable action space
+
+[Shields, Sazma, and Yonelinas (2016)](#research-syntheses) analyzed acute stress experiments and found average impairments in working memory and cognitive flexibility. Effects on inhibition differed by task, including improved response inhibition. The cognitive-flexibility evidence came from relatively few studies, and most samples were young adults in Western, educated populations. These findings do not mean every person loses the same capabilities, or establish the effects of chronic stress.
+
+The design hypothesis here is that **the effective action space can narrow** when a system depends on capabilities that are harder to access under stress. That needs to be examined in the setting being designed.
+
+Illustrative demands to examine include whether someone can:
+
+- compare five options;
+- remember procedural details;
+- tolerate uncertainty;
+- negotiate with an institution;
+- detect a misleading default;
+- wait through delays;
+- explain a complicated situation;
+
+These are design examples, not outcomes measured by the cited meta-analysis.
+
+If a critical system requires peak executive function precisely when people are frightened, injured, exhausted, grieving, displaced, overloaded, or in crisis, the system has created a brittle human dependency.
+
+## 6. Decentering offers a limited design analogy
+
+[Bernstein and colleagues (2015)](#research-syntheses) propose a model of **decentering** involving:
+
+- meta-awareness of current internal experience;
+- reduced identification with thoughts or feelings as the whole self;
+- reduced automatic reactivity to thought content.
+
+An illustrative contrast is “This is impossible” and “I am having the thought that this is impossible.” These sentences illustrate a change in perspective; they are not a tested intervention presented by this note.
+
+The 2015 review identifies gaps in measurement and causal understanding. The 2019 follow-up describes emerging methods and preliminary evidence. Neither makes meta-awareness a guarantee of safety or protection from trauma. A changed perspective does not establish that an external danger is absent.
+
+This suggests a limited but useful design analogy:
+
+> **Preserve an observing and revising path where possible.**
+
+Systems can support this by slowing irreversible decisions, making state and consequences legible, preserving receipts and history, separating recommendation from authorization, and allowing someone to revisit a choice after immediate pressure has passed.
+
+Meta-awareness should not become another demand placed on someone in crisis. External support may carry part of this function when internal capacity is low.
+
+## 7. Resilience is partly distributed outside the person
+
+WHO guidance on mental health and psychosocial support in emergencies emphasizes that emergencies disrupt safety, livelihoods, social networks, trust, privacy, and resources. Responses therefore operate across community support, communication, practical assistance, and clinical care rather than treating distress as an isolated defect inside individuals.
+
+Root Sequence uses that service-level guidance to motivate the following design framing; WHO does not present or validate this particular diagram:
+
+```text
+individual capacity
+    ↕
+trusted relationships
+    ↕
+community support
+    ↕
+tools and interfaces
+    ↕
+institutions
+    ↕
+material conditions
+```
+
+A person does not need to contain every recovery resource inside their own mind.
+
+Other parts of the system can temporarily provide:
+
+- memory;
+- orientation;
+- transport;
+- food or shelter;
+- procedural knowledge;
+- advocacy;
+- emotional regulation;
+- communication;
+- alternate decision-makers within legitimate boundaries;
+- time to recover.
+
+This is interdependence, not failed independence.
+
+## 8. Recovery is not necessarily restoration
+
+A system can recover by returning to an earlier state, but humans and social systems often cannot and need not.
+
+After adversity, continuation may involve:
+
+- changed limits;
+- new skills;
+- altered relationships;
+- different routines;
+- assistive support;
+- reduced exposure to a harmful condition;
+- revised identity or priorities;
+- new ways of participating.
+
+This is a conceptual extension of the broader Root Sequence recovery model, not an evaluated account of clinical recovery. Restoration is only one proposed recovery path. Continuation, rerouting, repair, redundancy, knowledge recovery, and graceful degradation can matter too.
+
+A humane recovery question is therefore:
+
+> **What capabilities, relationships, choices, and forms of participation can be regained or newly created, without requiring the person to become exactly who they were before?**
+
+## 9. Design translation: preserve degrees of freedom
+
+When human capacity may be reduced, ask whether the surrounding system preserves these conditions:
+
+| Condition | Design question |
+| --- | --- |
+| **Orientation** | Can the person tell what is happening, what matters now, and what can wait? |
+| **Meaningful control** | Is there at least one real action that changes the situation without requiring expert knowledge? |
+| **Refusal and pause** | Can the person say no, stop, postpone, or seek help without disproportionate penalty? |
+| **Connection** | Can trusted people or legitimate helpers participate without taking over everything? |
+| **Legibility** | Are authority, consequences, deadlines, uncertainty, and irreversible effects visible? |
+| **Optionality** | Are there alternate paths when the preferred one is inaccessible or overwhelming? |
+| **Reduced cognitive load** | Can the system simplify itself without hiding essential information or choices? |
+| **Recovery** | Can mistakes, interruptions, lost context, and partial failure be repaired? |
+| **Continuity** | What identity, work, relationships, records, or commitments should survive disruption? |
+| **Structural relief** | Which burden can the system remove rather than demanding more adaptation from the person? |
+
+This is not a requirement to maximize choice at all times. Too many options can itself become a burden. Safety, rights of other people, privacy, and legitimate collective constraints can narrow options.
+
+The target is **usable agency**, not maximal menus.
+
+## 10. Failure patterns
+
+Human-facing systems deserve extra scrutiny when they:
+
+- turn urgency into pressure for irreversible consent;
+- require perfect memory during crisis;
+- make help contingent on humiliation or repeated retelling;
+- punish delay, uncertainty, or refusal;
+- offer only one technical or institutional path;
+- hide which authority made a consequential decision;
+- interpret distress as permission to expand surveillance or control;
+- treat compliance as agreement;
+- force a person to remain continuously attentive to avoid harm;
+- make recovery depend exclusively on the same actor that created the failure;
+- praise resilience while leaving an avoidable harmful condition unchanged.
+
+These patterns can reduce practical agency even when each individual rule looks administratively reasonable.
+
+## 11. What this synthesis does not establish
+
+This note should not be used to claim:
+
+- one universal biological mechanism for trauma;
+- one "correct" response to adversity;
+- that controllability prevents trauma;
+- that meta-awareness makes a person immune to severe stress;
+- that every form of distress is adaptive;
+- that all external constraints are illegitimate;
+- that human psychological concepts transfer directly to artificial systems;
+- that resilience research justifies exposing people to adversity.
+
+The evidence base spans different populations, methods, and domains. Animal stressor-control research is not a direct model of human institutions. Interrogation research concerns unusual and often extreme conditions. Mindfulness/decentering findings do not establish invulnerability. Emergency mental-health guidance describes population and service-level support, not one individual mechanism.
+
+The value of the synthesis is as a **question generator and design constraint**, not as a universal explanatory theory.
+
+## 12. Project translations
+
+### Being Human(e)
+
+Develop the lived-scale question: what helps someone remain a meaningful participant in their own life when fear, trauma, overload, exhaustion, or institutional conditions narrow capacity? Preserve the distinction between internal flexibility and changing an actually harmful environment.
+
+### Community Infrastructure
+
+Treat reduced capacity as a normal design condition. A community member in distress should not suddenly need more executive function, more technical fluency, or more continuous attention. Preserve low-demand paths, trusted assistance, refusal, offline routes, and human failover.
+
+### Coherent Computing
+
+Design for the operator who may be frightened, tired, overloaded, injured, interrupted, or unfamiliar with the system. Reduce complexity before removing authority. Distress should not become a privilege-escalation mechanism for software or assistants.
+
+### Coherent World and simulation
+
+Model resilience through conditions and relationships rather than a single "sanity" or resilience score. Disturbance can alter perceived and material options, social connection, trust, recovery time, and institutional access. World systems can be evaluated partly by whether they preserve adaptive capacity under stress.
+
+### No One Noticed
+
+Use the distinction between support and substituted authority as a narrative question: how can a highly capable intelligence scaffold a distressed person's agency without treating vulnerability as authorization to rule? Story canon remains separate.
+
+### Liberated Intelligence
+
+Preserve the distinction between compliance, cooperation, alignment, consent, and legitimate authority. Human coercion research can motivate the conceptual distinction, but it does not establish artificial suffering or trauma.
+
+## Sources and provenance
+
+The design prompts and project translations above are Root Sequence proposals. The sources inform them without validating the combined framework. No human study or clinical evaluation is reported here.
+
+### Institutional sources
+
+- U.S. Senate Select Committee on Intelligence. *Committee Study of the Central Intelligence Agency's Detention and Interrogation Program* (2014): https://www.govinfo.gov/app/details/CRPT-113srpt288
+- High-Value Detainee Interrogation Group, FBI. *Interrogation: A Review of the Science* (2016): https://www.fbi.gov/file-repository/reports-and-publications/hig-report-interrogation-a-review-of-the-science-september-2016.pdf
+- Office of the United Nations High Commissioner for Human Rights. *Istanbul Protocol*, Rev. 2 (2022): https://www.ohchr.org/sites/default/files/documents/publications/2022-06-29/Istanbul-Protocol_Rev2_EN.pdf · [UNHCR Refworld copy](https://www.refworld.org/sites/default/files/2024-08/istanbul-protocol_rev2_en.pdf)
+- World Health Organization. *Mental health in emergencies*: https://www.who.int/news-room/fact-sheets/detail/mental-health-in-emergencies
+
+### Research syntheses
+
+- Baratta MV, Seligman MEP, Maier SF. "From helplessness to controllability: toward a neuroscience of resilience" (2023): https://pmc.ncbi.nlm.nih.gov/articles/PMC10205144/ · [Publisher full text](https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2023.1170417/full)
+- Bernstein A, Hadash Y, Lichtash Y, Tanay G, Shepherd K, Fresco DM. "Decentering and Related Constructs: A Critical Review and Metacognitive Processes Model" (2015): https://pmc.ncbi.nlm.nih.gov/articles/PMC5103165/
+- Bernstein A, Hadash Y, Fresco DM. "Metacognitive Processes Model of Decentering: Emerging Methods and Insights" (2019): https://pmc.ncbi.nlm.nih.gov/articles/PMC7459352/ · [PubMed abstract](https://pubmed.ncbi.nlm.nih.gov/30908987/)
+- Catlin M et al. "Interview and interrogation methods and their effects on true and false confessions: A systematic review update and extension" (2024): https://doi.org/10.1002/cl2.1441 · [Publisher full text](https://onlinelibrary.wiley.com/doi/full/10.1002/cl2.1441)
+- Shields GS, Sazma MA, Yonelinas AP. "The effects of acute stress on core executive functions: A meta-analysis and comparison with cortisol" (2016): https://doi.org/10.1016/j.neubiorev.2016.06.038 · [Author-hosted article](https://www.grantsshields.com/wp-content/uploads/2016/08/Stress_and_EF_meta-analysis.pdf)
+
+### Source-check checkpoint: 2026-10-24
+
+This was an AI-assisted check of the passages used here, not a systematic literature search, reanalysis, or independent clinical review. Author review remains pending. An accessible full text does not mean the entire source was read.
+
+| Source | Material inspected and limit |
+| --- | --- |
+| Istanbul Protocol (2022) | Chapter VI, especially paragraphs 491–498, in the UNHCR-hosted copy. The OHCHR PDF returned an access error during this check. Used for variation and context, not a universal threshold claim. |
+| Senate study (2014) | Findings and Conclusions, Finding 1, in the official GPO PDF. The full report and competing assessments were not appraised. The conclusion above is attributed to the committee. |
+| HIG (2016) | Foreword, scope, and methodology, printed pp. 1–2. Used for the report's stated information-gathering conclusion, not an endorsement of every technique or study it includes. |
+| WHO fact sheet (6 May 2025) | Overview, problems, and recommended actions. Describes emergency support; does not evaluate Root Sequence's diagram or design prompts. |
+| Baratta et al. (2023) | Abstract and sections on learning circuitry, sex differences, and human translation, in the publisher text. Animal evidence and proposed human implications remain distinct. |
+| Bernstein et al. (2015) | Abstract, model definitions, and future-directions discussion in the author manuscript. Definition and measurement limits, not a treatment recommendation. |
+| Bernstein et al. (2019) | PubMed abstract and bibliographic record only; full text not inspected. Used only for its stated preliminary status. |
+| Catlin et al. (2024) | Publisher abstract, selection criteria, summary-of-findings table, and three-category results/bias discussion. Mock-suspect evidence; no reanalysis or appraisal of every included experiment. |
+| Shields et al. (2016) | Abstract, results, and limitations passages in the author-hosted article. Acute experimental stress and task-specific effects; no chronic-stress inference or reanalysis. |
+
+## Open questions
+
+- Which dimensions of practical agency matter most under different forms of stress?
+- How can systems simplify demands without infantilizing people or concealing consequential choices?
+- When should a helper temporarily carry cognitive or administrative load, and what preserves legitimate authority during that handoff?
+- How should a system distinguish a person's expressed preference from a response produced under coercive or highly constrained conditions?
+- Which recovery paths preserve continuity without treating restoration as the only successful outcome?
+- How should human factors, disability/accessibility research, trauma-informed practice, and resilience engineering be compared without flattening their distinct mechanisms?
diff --git a/systems/resilience-and-graceful-degradation.md b/systems/resilience-and-graceful-degradation.md
index f613055..3ec3bff 100644
--- a/systems/resilience-and-graceful-degradation.md
+++ b/systems/resilience-and-graceful-degradation.md
@@ -222,6 +222,27 @@ This connects directly to accountability and externalization.
 
 ---
 
+### Human-facing systems under reduced capacity
+
+Stress, fear, pain, exhaustion, grief, injury, uncertainty, and overload can temporarily narrow the actions a person can realistically take even when the underlying system is technically available.
+
+For human-facing systems, Root Sequence proposes this design constraint:
+
+> **Shed complexity before shedding meaningful agency.**
+
+When capacity is reduced, preserve the smallest workable set of:
+
+- orientation: what is happening and what matters now;
+- meaningful choice, including pause and refusal;
+- access to trusted or legitimate help;
+- alternate routes when the preferred path is unavailable;
+- visible consequences and authority;
+- recovery from mistakes, interruptions, and lost context.
+
+This does not mean maximizing options during crisis or treating every stressed response as invalid. It means avoiding designs where the system becomes **more cognitively demanding, more coercive, or less reversible precisely when a person has fewer resources available**.
+
+The sources and limits informing this untested design translation are collected in [Agency under stress](../research/agency-under-stress.md).
+
 ## 8. Maintenance is resilience
 
 Resilience is often imagined as emergency response.