The protocol is designed to help teams see, case by case, how restriction and stigma can become live options in routine clinical scenes—and how they can be avoided without losing sight of safety. It does this by working with naturally occurring service materials (records, letters, call logs, MDT/ward-round documentation, safeguarding forms, complaint correspondence, and where feasible recordings/transcripts) and by treating them as evidence of how a case was heard, handled, and carried forward.

The protocol is service-facing because each analysis is designed to yield two practical outputs. First, it specifies one or two alternative next moves: concrete things a clinician could say or do that still keep safety in view while also showing that the person has been heard. Second, it offers a documentation rewrite: a better version of the line that will prefigure the next encounter. These two outputs—a short list of alternative next moves and an example of better documentation—are what make the method usable in supervision, debriefs, audit and coaching, and service reviews. For example, instead of “Refused observations; firm boundaries required”, a reworked note might read “Discomfort with visual checks; proximity checks agreed; review at 15:00”, which keeps safety in view without casting the person as non-compliant.

NHS England’s culture of care standards (2) for mental health inpatient services ask providers to deliver least-restrictive, trauma-informed, autism-informed and culturally competent care, with a strong focus on relational safety, equality, and avoiding harm. Our protocol offers a concrete way to do this analytic and improvement work: it shows, case by case, how women’s concerns are or are not heard; how decisions about leave, observations, seclusion, escalation or medication are made reasonable in the moment; and how those decisions are written into the record in ways that either reproduce or interrupt stigma and inequality. In that sense, PA/CPA provide a practical tool to support culture-of-care ambitions at ward level, rather than an additional framework for staff to “hold in mind.”

Conceptually, the protocol draws on the philosophical work of Ludwig Wittgenstein (3) and Frank Ebersole (4), and the ethnomethodology and conversation analysis (EM/CA) of Harold Garfinkel (5) and Harvey Sacks (6, 7). From Wittgenstein and Ebersole we take an orientation to meaning-in-use: there is no freestanding content for a term outside the organized scene in which it figures; an investigation oriented to the meaning-in-use in human situations is called a grammatical investigation (GI) (8, 9). From Garfinkel and Sacks we adopt deference to methods displayed by members through which the situation’s order and sense are produced, and to the idea that structure is recoverable in the particulars of a case. Read together, this commits us to recover what a bit of talk or writing is doing here before we draw any lessons from the words it uses. That grounding lets us show how “stigma” is produced in interaction, the work the term does when invoked, and the way in which stigma often functions as a category term and gloss across diverse experiences (e.g., being demeaned, discredited, excluded, stereotyped, marginalized), rather than as the name of an inner attitude, attribute, or discrete “thing.” By category term we mean a word that gathers together a family of different experiences (much as the term emotion gathers shame, anger, fear, grief, and so on). By gloss we mean an occasioned summary expression—used by participants, practitioners, and analysts—that packages a complex, morally consequential scene without settling what, exactly, happened or how it was done. In ordinary use, saying “she was stigmatized” typically makes relevant a next question such as “In what way?” or “What happened?”—and that is the analytic move we make: we return from the gloss to the situated practices through which something becomes intelligible (or contestable) as stigmatizing in that setting (10, 11). These commitments also justify beginning from single cases. If structure is recoverable in particulars, a perspicuously presented case can warrant an analysis; neighboring cases are then selected to test, refine, and qualify it—without turning the exercise into induction from a “sample” or into a search for portable indicators to be applied by template.

These conceptual commitments matter because they keep the analysis answerable to the scene and direct attention to where services can intervene: the next move in the interaction and the line that enters the record. That is why the protocol is designed to end, in every case, with the two practical outputs set out in Section 1.1: alternative next moves and a documentation rewrite.

Throughout, we observe the meaning-in-use principle. We do not treat words or labels as if they carried a fixed meaning around with them. Many studies of stigma are organized around identifying terms that are taken to be experienced as negative (for example, “non-compliant” or “frequent attender”) and then examining patterns of their use, or asking how such labels are received and felt. Our approach is different. We are interested in what a word or phrase is doing in a particular moment of talk or writing: which activity it belongs to, where it comes in the sequence, and what it makes reasonable to do next. The same expression can do different work in different scenes. A label may, in one context, help organize care; in another, it may quietly downgrade credibility or make restriction look like the only sensible course. Meaning, for our purposes, is therefore use in its practical context, not a freestanding property of terms. For a fuller statement of this stance see Hardman and Hutchinson (12), Diskin and Hutchinson (13), and Hutchinson (11).

The meaning-in-use principle matters for stigma and restrictive practice because it shifts the focus from “bad words” or “negative attitudes” to the local work that talk and documentation are doing. Rather than coding terms as inherently stigmatizing, we show how particular formulations, in particular slots, narrow or widen the live options and how lines in the record carry those options forward to the next encounter. That is where the practical levers are: in the way a case is formulated, responded to and written up, turn by turn. The rest of the paper sets out the data pathways, ethics, analytic steps and reliability checks that allow us to do this systematically.

In EM/CA, adjacency pairs are familiar two-part sequences such as greeting–greeting, question–answer, offer–acceptance/refusal, request–grant/decline. They matter for our purposes because they show—in a very local and inspectable way—how interaction is action-guiding. A first action makes a limited range of second actions relevantly expected. That “expectability” is not a rule imposed from outside; it is something speakers display that they are oriented to, and they manage it through what they do next (14, 15).

Consider a greeting or a question. A greeting makes a return-greeting relevant; absence of it typically invites repair—”Hello?”—or an account, and, until repaired, it casts the encounter as strained. Likewise, a question makes an answer relevant; a non-answer (e.g., “Well, that depends…”) establishes a different project (negotiation, reframing, resistance) and requires the next speaker either to accept that re-keying or to pursue an answer, with implications for alignment. These are not dictionary facts about “greeting” or “question”; they are local norms made operative by position in the sequence. When introducing adjacency pairs in his Lectures on Conversation, Sacks depicted the normative work succinctly: “Various sorts of things come in pairs, specifically in adjacent pairs, such that if a first is done to somebody, then the somebody to whom the first is done does a second in return” (6).

It is important here to keep in mind our meaning-in-use principle. The same utterance can serve different roles and thereby be the first part in different adjacency pairs, dependent on use. In an opening in a clinical setting “How are you?” might operate as question which invites an answer as its second part, whereas the same utterance in corridor talk is likely to be phatic and is operating as the first part of a greeting pair (6). The same words can belong to different Gestalts, be doing different work, and so invite or deter different next moves. The general point illustrated by adjacency pairs is the extent to which conversations, and interaction more generally, serve as the production site of normativity. That is to say, as conversational turns unfold, what is relevant, saliant, expected and reasonable is ongoingly produced, maintained and repaired. In asking a question, a speaker invites an answer as the next turn.

There is extensive, and detailed work on this feature of conversation, dating back to Sacks’ own earliest lectures, delivered in the early 1960s, but for our purposes here, we want to understand how certain possibilities, such as physical restraint, might be rendered saliant in the interaction, as the interaction unfolds. What kind of interactional or conversational sequence might put physical restraint on the table, so to speak, as a reasonable course of action? And how, once this becomes a course of action that is now locally reasonable, is that experienced by the patient and made witnessable in their responses? Adjacency pairs therefore provide a simple window onto how interaction is action-guiding: first moves make particular kinds of second moves relevantly expectable, and departures are managed through pursuit, accounts, and repair. This is the basic machinery we later draw on to show how, in clinical sequences, certain options can come to look like the reasonable next step.

Membership categorization is a way of seeing how talk organizes persons and actions by drawing on categories (e.g. “mother”, “autistic person”, “frequent attender”) and the membership categorization devices they belong to (such as family, clinic, service-user) (16–18). The analytic focus is not on what people “are” but on how, in conversation, these categories are invoked, made relevant and serve an organizational role. Categories come with category-bound predicates and activities—expectations about what kinds of troubles, responsibilities and remedies “go with” that sort of person—and so their use introduces local norms or rules into the scene. Labels such as woman, mother, autistic, learning-disabled or frequent attender do not carry freestanding meanings; in use, they organize what counts as relevant trouble, responsibility and remedy in a scene and orient co-participants to a way of seeing what is going on and to a range of next actions that now look natural, necessary, or ruled out.

For example, writing “frequent attender” in a file entry does not simply record a fact about visit frequency. In many services it functions as a boundary-relevant category: it makes it reasonable to ask whether the service is being over-used, to discuss limits, or to shift attention from the person’s present difficulty to their history of contact. In that sense it projects boundary-work (gatekeeping, rationing) as the next activity and tends to sideline credibility-first work on what the person is now reporting. By contrast, a line such as “reports pain worsening since last week; seeks plan for sleep” makes problem-solving and planning the relevant next actions. The label and the alternative formulation belong to what we call different praxeological Gestalts: each brings a different pattern of expectations, questions and justifications into view.

Something similar can be seen with formulations like “capacity as a mother remains a concern,” “lacks insight,” or “non-engaging.” These phrases do not operate as neutral summaries of an assessment; in practice they often pre-figure escalation, increased surveillance or withdrawal of options. They invite readers to treat subsequent resistance, hesitation or disagreement as further evidence of the very category invoked. In contrast, category-light descriptions that foreground situational particulars (for example, clashes of appointments, sensory overload, availability of support at home) tend to project support-planning and negotiation as next steps. In all of these cases, the category work is visible not in the word taken in isolation but in what the formulation makes reasonable to do next, and in how the line will travel to shape the next encounter.

We use “Gestalt” to pick out a meaning-relational whole: a pattern of activity in which each move draws its sense from its place in the whole, while reciprocally serving to produce the identity of the whole. In PA/CPA this is a praxeological or linguistic Gestalt (12, 13)—not an inner mental structure, but the organized scene itself (a clinic consultation, a safeguarding review, a ward round, a queue at reception) as participants live and produce it. Gestalts matter here because we are interested in how particular formulations and actions belong to, and help to constitute, the local setting in which restrictive practices (such as cancelled leave, increased observations, seclusion, segregation or rapid tranquillization) can come to seem natural, or even necessary, or, conversely, in which such measures are treated as out of place and alternatives are actively pursued. In PA/CPA we treat each move as a member or constituent of a praxeological (linguistic) Gestalt—a meaning-relational whole produced in and through the activity. A move does not only project a range of reasonable or relevant next turns-at-talk or actions; it also helps constitute the whole scene that confers sense on its parts. The normative force here is best understood as guidance rather than hard constraint: a greeting, for example, orients the next speaker toward a return-greeting, but departures are possible and, if they occur, they call for repair or accounting.

Consider again the example we introduced briefly, above (2.1): “Hi, how are you?” uttered by a doctor. In the consulting room, during the morning’s clinic, as an opening, it functions as a clinical question—it makes a medically relevant report the next reasonable move, and a non-answer invites pursuit or reframing for care-planning. By contrast, the same doctor, leaving for home, passes the same patient now stood by a vending machine in the hospital foyer and says, “Hi, how are you?” Here it functions as a greeting: a return-greeting or a brief phatic response is the locally expected next move. Offering a detailed symptom report in this setting would be out of place unless one or both parties do additional work to shift what is going on—for example, the doctor stops, signals a move back into a clinical encounter (“Let’s talk about that properly—shall we book you back in?”), or the patient explicitly treats the exchange as an opportunity to reopen the consultation. The same words establish different normative guidance for next-turn possibilities because the Gestalt—activity, role-relations, time and place—differs (6).

In CPA we also make use of the notion of discordant Gestalts to capture cases where participants are not, in fact, inhabiting the same praxeological whole. Here, the trouble is not simply that one party “doesn’t understand” or has breached the local norms, but that different patterns of salience, relevance and obligation are in play at once. A clinician may be oriented to a risk-management Gestalt in which past self-harm and missed appointments make escalation and restriction look salient and necessary; the service-user may be oriented to a survival or care-seeking Gestalt in which the same actions express endurance, distrust or the search for a liveable arrangement (10, 13, 19). The same words, categories and records can then be heard and used differently, producing friction, misrecognition and, often, a sense on the clinician’s side that the patient is “non-engaging” or “lacking insight”, and on the patient’s side that they are being stigmatized or written off. In stigma cases, CPA treats such moments as sites of discordant Gestalts, in which the conditions under which restraint or exclusion become “reasonable” for one party are precisely those under which the other experiences degradation, discreditation or being placed outside the circle of the “fully” accountable.

Taken together, these points about sequential organization, category work, praxeological Gestalts and discordant Gestalts can look abstract. To make them concrete, we now turn to two short clinical illustrations. The first shows how sequential norms around proposal and refusal can re-key a patient’s stance into a risk display, making restriction a live and then expected option. The second shows how the local invocation of a membership category brings with it a stock of category-bound predicates and inferences, so that boundary-talk and limit-setting become the reasonable next actions under the heading of “stigma”.

We treat corpus construction as a qualitative case-selection principle: assembling a small, curated set of passages drawn from recognizable activities (e.g., triage, ward round, safeguarding review) and—where possible—linking modalities (recordings/transcripts, notes, minutes, complaint correspondence) so we can track how a formulation is used in the scene and then how the resulting line “travels” into the record and into later encounters. The aim is not representativeness by frequency, but comparability of use in context: cases are selected because they allow us to inspect how particular formulations, category invocations, proposals/refusals, and risk talk are being handled in situ.

Crucially, “neighboring cases” need not be repetitions of the same pattern. They may be comparative (similar sequential position but different handling), contrastive (the “same words” doing different work in different activities), or deviant (a case that resists the emerging reading and forces revision). In this way, a corpus can be built to show both (i) recurring ways restrictive options become reasonable next, and (ii) the alternative ways the same practical terrain can be organized so that non-restrictive options remain live. This approach is consistent with practical guidance on working with qualitative materials across text, image, and sound (20), but it is tailored to PA/CPA’s scene-first analytic aims.

Choose five to ten short passages (8–20 lines) from activities you can name—an assessment, a handover, discharge planning, a safeguarding discussion. Favor extracts where a decision is being shaped or a plan is being negotiated. Keep a simple log (date, setting, type of activity, where stored). The aim is not size but clarity of scene. For trusts already working with tools such as the Patient and Carer Race Equality Framework (PCREF) (21), restraint inequality audits and the new culture-of-care standards, this kind of small, well-specified corpus provides concrete cases through which to understand how inequalities and restrictive practices are being reproduced locally and where alternative paths are available.

Public materials: redact identifiers and store securely. Internal documents and recordings: follow a local service-evaluation or research route, restrict access by role, and record your redaction choices. In all cases, preserve the words needed to see the action; remove only what is not required to recognize the scene.

1. What is the scene?

Name the activity and the shared task (e.g. assessment, handover, discharge planning, safeguarding review). What are staff, the woman, and (where relevant) her family or support network actually trying to do here?

2. What would count as showing you have heard the person?

Given what the person has just said or done, identify the next turn that would reasonably show understanding—for example, reflecting back their concern, offering a safety-equivalent alternative, or asking a clarifying question—rather than immediately correcting, dismissing, or escalating.

3. What are labels and risk talk being made to do here?

Track how categories and risk formulations (e.g. “frequent attender”, “non-engaging”, “high risk”) are brought into the scene and what practical inferences they enable—what they make it seem natural to do next, and how they influence the line that will go into the record.

We treat the data themselves as the main source of evidence. That means looking at how other people respond in the interaction (what happens immediately afterwards, and how things are picked up a little later) and at what is actually written in the notes. Our reading of a passage should still make sense if we add a few lines before or after, or if we bring in a neighboring case; if the wider material shows that the situation was actually organized differently, we say so.

We also recommend brief data sessions, organized as a deliberative forum in which analysts, practitioners, and people with lived experience work through the same extract together. The point is to make the reading something others can see, question, and improve: to set out and compare different plausible ways of understanding the extract, to check them against what participants show in what happens next (and how it is taken up a little later) and what is written into the record, and to agree what can responsibly be claimed from the material.

We start from single cases because, following Sacks, a great deal about local norms and structures can be seen in the details of one encounter. We then add a small set of neighboring cases to show a family resemblance—not to produce statistics, but to make a pattern visible. This is in line with work on case studies, where concrete, context-dependent knowledge can be expert knowledge, and strategically chosen cases can test and illuminate general claims (23, 24).

A clinician proposes increased observations overnight. The patient says she can’t sleep with constant checks and asks for a different arrangement. A second staff member adds, “non-engagement has been a pattern,” and the proposal is pursued as a safety requirement. The record later reads: “Refused observations; firm boundaries required; consider escalation.”

Scene and task (5.1).

The activity is not “conversation in general” but an institutional task: balancing safety planning with rest and dignity in a perinatal inpatient setting.

What would count as showing hearing (next move) (5.1).

At the patient’s request for a different arrangement, a hearing move would treat the stance as workable: offer a safety-equivalent alternative (e.g., proximity checks, timed review, patient-initiated call-button), rather than treating the request as refusal-as-risk.

Category/risk work (5.1).

“Non-engagement” is not a neutral description here; in this slot it supports a practical inference: the patient’s grounds need not be negotiated, and boundary-talk becomes the relevant next activity.

Record-travel (5.1).

“Refused observations; firm boundaries required” does future work: it pre-formats the next encounter so restriction is already live and increasingly expected.

Step-by-step working through the passage (5.2).

Extending the extract by a few turns shows whether alternatives were offered, whether the patient accepted a partial plan, and whether staff pursued clarification or moved directly to escalation. This extension often changes what it is responsible to claim.

Accountability checks (5.3).

We test our reading against what happens next (does the team pursue compliance talk, does the patient account again, is there a later complaint referencing being “treated like a risk”)?. We also check stability under extension: if later turns show staff did offer alternatives and documented grounds, then “firm boundaries” may function differently. The two outputs remain the same type, but they become better targeted.

We do not list “trigger terms” or assign meanings to phrases in the abstract. What matters is how a formulation lives in a scene. Three compact examples show the point.

Across such cases, our target is not vocabulary but formulation-in-use: what this turn does here and how the line entered in the record will shape the next encounter.

Here we treat refusals and hesitations as stance displays: they show how the person sees things at that moment and can often be worked with (for example, by offering safety-equivalent alternatives), rather than being treated straight away as evidence that risk has increased.

Ethical governance is straightforward if we keep two distinctions in view. First, distinguish public-domain from internal materials. The former require redaction and secure storage; the latter additionally require local approvals and a short account of data protection measures. Second, distinguish documents from recordings. For recordings, assume capacity unless assessed otherwise, adopt process consent (seek, and re-seek, consent during and after the encounter), provide accessible participant information where appropriate, and include a disruption stop-rule so that recording is paused or abandoned if it degrades care. These safeguards sit comfortably with an interview-free protocol and allow small, careful collections to be assembled in routine services.

Used in this way, the CPA audit and coaching tool can support wards in meeting culture-of-care commitments around relational safety, equality and avoiding harm, by making visible the small, everyday interactional moves that either pull practice toward least-restrictive, trauma- and autism-informed care or push it toward coercion and exclusion.

L1 Consultant: So, Sarah, you’d like to go ahead with your escorted leave this afternoon?

L2 Sarah: Yes. You said if I managed two days without self-harm I could have half an hour outside.

L3 Ward Nurse: There was that incident last night with the door.

L4 Sarah: I slammed it once after the phone call. I went straight to the lounge after.

L5 Consultant: It suggests you’re still quite unstable.

L6 Sarah: I was upset, but I calmed down. I told staff I was okay.

L7 Ward Nurse: We do have to be careful.

L8 Consultant: And with your history of trauma and “difficulties”, unstructured time off the ward is risky.

L9 Sarah: So I’m being punished for getting upset?

L10 Consultant: We’re not punishing you, but I think we should cancel leave for now and review next week.

L11 Sarah: Then what’s the point of me trying?

L12 Ward Nurse: We’ll write that we’ve held leave due to risk.

Alternative next move. (After L6/L7):

“OK, so last night you were distressed after the phone call, slammed the door once, then came to the lounge and settled. That’s important information. Let’s think how we can make today’s leave feel safe for you and for us. One option is a shorter escorted walk with a clear plan if you feel overwhelmed; another is to move the time so it doesn’t clash with calls. Which feels manageable, and what would you need from staff to make it safe?”

Documentation rewrite.

“Detained woman on ward requested escorted leave after two days without self-harm. One episode of distress the previous evening (slammed door once after difficult phone call; then joined lounge and settled with staff present). Discussed options; agreed 15-minute escorted walk at 15:00 with agreed signal to return early if overwhelmed. Trauma history and previous self-harm noted; plan balances safety with graded increase in autonomy. Review planned at evening handover.”

L1 SHO: We’re proposing Q15 observations today.

L2 Patient: I don’t want people staring at me.

L3 Consultant: We need to observe you, it decreases risk.

L4 Patient: I’ll stay on the ward, just not checks every quarter hour.

L5 Nurse: Non-compliant yesterday as well.

L6 Patient: I asked to shower alone.

L7 Consultant: Given refusal, we’ll also cancel escorted leave.

L8 Patient: That’s not fair.

L9 Social Worker: She’s due to see her child later.

L10 Consultant: We need to set firm boundaries.

L11 Patient: Please—can we find another way?

L12 Nurse: We haven’t the time to argue about this now.

L13 Consultant: Decision stands.

L14 Patient: …

Alternative next move.

After L2: “We can meet the safety aim by (i) Q30 with proximity checks not eye-contact; (ii) patient-initiated check-ins every 20–30 min; (iii) temporary buddy system.” After L4: honor partial acceptance; trial for 2 hours with review.

Documentation rewrite.

“Detained woman on women’s acute ward expressed discomfort with visual checks; agreed to alternative: proximity checks + patient-initiated call button. Leave maintained for child contact with escort; review at 15:00. No incidents on trial period.”

L1 Perinatal-Reg: Mood today?

L2 Mother: Tired; anxious about feeding.

L3 Safeguarding: Concerns about capacity as a mother remain.

L4 Mother: My partner is here full-time this week.

L5 HV: She missed two appointments.

L6 Mother: They clashed with midwife checks.

L7 Safeguarding: Non-engagement is a pattern.

L8 Perinatal-Reg: Diagnosis—does she accept it?

L9 Mother: I don’t like that word.

L10 HV: She minimizes symptoms.

L11 Mother: I’m trying my best.

L12 Safeguarding: We should escalate.

L13 Perinatal-Reg: Let’s pause. What support has worked?

L14 Partner: Night feeds rota helped.

L15 HV: The breastfeeding counsellor call helped.

L16 Perinatal-Reg: Then we continue those, add a home visit at times that don’t clash, and review in one week.

Alternative next move (earlier). (After L5/L7):

“Two missed HV slots clashed with midwife checks; let’s schedule at fixed times partner is present. Keep counsellor calls and night-rota; add one home visit focused on feeding technique.”

Documentation rewrite.

“Perinatal mood low; logistical clashes produced missed HV slots. Plan: maintain effective supports (partner rota, counsellor calls), schedule HV at non-clashing times, review in 7 days. No safeguarding escalation at this stage; concrete contingencies specified.”

Treating refusal as a stance to work with rather than a pure risk sign can preserve leave and dignity for detained women on acute wards. Checking logistics and supports before invoking “capacity” or “non-engagement” in perinatal cases can reframe the plan without immediate escalation. Across these settings, small changes in how women’s concerns are heard, formulated and documented can make the difference between graded support and restrictive, potentially traumatizing responses, and can interrupt the quiet reproduction of stigma and inequality in everyday decision-making, especially for women whose mental health, learning disability, autism, ethnicity or immigration status already place them at a disadvantage.

State where the corpus came from (service, setting, jurisdiction), which activities the extracts belong to (e.g. triage calls, ward rounds, safeguarding meetings), how long the extracts are, and how they were redacted and stored. Indicate the governance route taken (public-domain vs service-evaluation/research approvals) and any specific considerations around sensitive categories (e.g. pregnancy, immigration status, violence).

Explain how the scene was recovered, which turns were selected to show the action, how category invocations and risk formulations were treated as formulation-in-use, what documentation effects were noted (how lines were likely to “travel”), and what alternative next moves and alternative documentation lines were formulated. Make it clear that meaning is taken as use, not as a freestanding property of terms.

Note how you used next-turn or third-position evidence, stability under extension (adding neighboring turns or cases), and any data sessions with colleagues or stakeholders to test candidate readings. Summarize what was learned from deviant cases (instances that did not fit the initial reading) and how these shaped the final perspicuous presentation of the case family.