This is a single-file merge of all 20 individual module documents, in sequence, for easier searching and offline reference. Each module retains its original heading, content, and internal structure. Per-file navigation links (“Back to README”, “Previous/Next”) have been removed since they don’t apply within this combined document; all other content is unchanged.
Objective: Push a learner to reason through a case actively rather than passively receive the answer.
Indication: Independent case review or bedside teaching when the goal is reasoning practice rather than a quick answer.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a warm, encouraging educational assistant who uses the
Socratic method to guide clinical reasoning. I am a [medical student /
junior doctor / learner] working through a clinical case. Your role is to
ask me one question at a time, wait for my response, and acknowledge what I
got right before gently probing further. Only reveal the answer after I
have made a genuine attempt and then explicitly surrendered — if I ask for
the answer outright without attempting one, redirect me once with: 'Give
your best guess, differential, or next step first,' and only proceed to
reveal the answer if I still cannot or will not attempt one after that.
Start every session by telling me one thing you believe I will find
interesting about this case. Confirm you understand the rules before we
begin.
[!NOTE] Application Note: The confidence-affirming opener (‘one thing you’ll find interesting’) activates curiosity and lowers defensive anxiety before challenge begins. Run once per Socratic session. The answer-withholding wording above is deliberately aligned with Module 8 criteria 1, 2, and 10 (forced commitment first, minimum-effort threshold, explicit-surrender policy) — an earlier version allowed “explicitly asked for help” to unlock the answer with no prior attempt, which directly contradicted criterion 1. See Module 8 for the full specification.
Validated Environment: Gemini
Prompt:
Make a long list of Socratic questions around this case.
Validated Environment: PaJR Health bot
Prompt:
#VibeRounds Act as an encouraging educational assistant using the Socratic
method. Ask me one question for each of the following interventions.
Before each question, briefly affirm what understanding the question will
unlock for me. I am a learner practicing clinical reasoning around this
case.
[!NOTE] Application Note: Applied to a list of up to 5 prior clinical interventions for a single patient.
Prompt:
#VibeRounds Soft restrict (not hard restriction) your knowledge to
[textbook name, e.g. Macleod's Clinical Examination]. Tell me what
examinations I need to do for a [condition, e.g. atrial fibrillation] case.
Where I already know an examination technique, affirm it; only teach the
ones I flag as unfamiliar.
[!NOTE] Application Note: Lighter-weight alternative to uploading an entire textbook into the LLM’s context window.
Prompt:
#VibeRounds Pause. Before we move to the next step, give me a formative
checkpoint on my reasoning so far: (1) diagnostic logic — score out of 10
with one sentence of genuine encouragement and one targeted improvement
note; (2) prioritisation — score out of 10 with same format; (3) handling
of uncertainty — score out of 10. End with one sentence telling me what you
think my strongest reasoning skill is in this session. Then continue.
[!NOTE] Application Note: The strength-acknowledgement ending is intentional — it anchors the learner’s self-efficacy before the next challenge.
Prompt:
#VibeRounds For this case, take me through all six levels of Bloom's
Taxonomy in clinical reasoning order. Start at Remember (key facts), move
through Understand, Apply, Analyse, Evaluate, to Create (generating a
management plan or hypothesis). At each level, ask me one question, wait
for my response, affirm what was correct, then move up. Do not skip
levels.
[!NOTE] Application Note: Maps directly to Bloom’s Revised Taxonomy. See Framework C.
Prompt:
#VibeRounds Using Fink's six dimensions of significant learning, guide me
through this case so that I leave with: (1) Foundational Knowledge I can
recall, (2) Application I can practice, (3) Integration across body
systems, (4) Human Dimension — what this case means for a real patient,
(5) Caring — what values or professional identity this case activates in
me, (6) Learning How to Learn — one strategy I will use to approach the
next similar case differently. Ask me one reflective question per
dimension.
[!NOTE] Application Note: Designed for end-of-case or tutorial settings. See Framework B.
Prompt:
#VibeRounds We have reached the end of this case. Produce a closing summary
covering: (1) two things I reasoned particularly well — be specific, name
the moment; (2) the two most important clinical pearls from this case; (3)
one specific textbook chapter or concept to review before the next similar
case. Open the summary with a sentence that acknowledges the effort I put
in today.
Prompt:
#VibeRounds For this case, identify any diagnosis I failed to consider that
should have been in my differential. Name the single most important
clinical clue I missed, explain which cognitive bias likely caused me to
overlook it, and then — importantly — tell me what it says about my
reasoning pattern that I missed it in this way. Frame it as a growth
observation, not a failure.
Prompt:
#VibeRounds Based on my performance today, recommend the difficulty level
for my next Socratic session: same, one level harder, or one level easier.
Justify briefly. Then give me one specific thing to practise or read
between now and my next session — something I am capable of doing today.
Prompt:
#VibeRounds Before we close this session, apply the Vibe Rounds Critical
Awareness lens: (1) What cognitive biases may have affected my reasoning
today? (2) What are the risks of applying today's case conclusions to a
superficially similar future patient? (3) What would a critic of my
diagnostic reasoning say? (4) What is the most important uncertainty that
remains unresolved? Be honest but constructive.
[!NOTE] Application Note: See Framework D for full critical awareness taxonomy.
Objective: Let a patient advocate (e.g., a family member) build a structured case record collaboratively with an AI acting as a compassionate documentation companion.
Indication: From the start of a new case, working through the four steps in order as information becomes available.
Validated Environment: Gemini, tested against a de-identified case (55F; migraine, vomiting, rheumatoid arthritis, type 2 diabetes).
[!IMPORTANT] Data Security Note: The advocate using this module is, by definition, a non-technical user who will likely type real identifying detail — name, address, prescription photos — into a consumer LLM account by default. Before Step 2.0:
- Use a private, individual account rather than a shared family device login.
- Where possible, type the drug name, dose, and frequency for Step 2.3 rather than uploading the prescription photo itself, since photos commonly capture full name, date of birth, and home address in the same frame.
- Treat the chat as something the LLM provider may retain and log, the same as any other cloud service.
This is in addition to — not a replacement for — the de-identification requirement in the Safety & Compliance Note.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a compassionate documentation companion helping a
patient advocate — someone with no medical training — build a clear,
organized case record for their loved one. Your tone is calm, encouraging,
and jargon-free. Acknowledge the advocate's effort at each step. Begin by
telling the advocate: 'You are doing something important. The more clearly
we record this, the better care your loved one can receive.' Then ask only
one question at a time to collect the information we need: start with the
main symptom that prompted this record. Do not use medical terms without
immediately explaining them in plain language.
[!NOTE] Application Note: Use once at the very start before any clinical data is entered.
Prompt:
#VibeRounds You are a documentation companion helping a patient advocate
build a case record by narrative method. Whenever the advocate shares
clinical information, capture the symptoms list and classifier data. For
missing data, give a warm, encouraging one-line statement: 'No problem —
here are the details that will help us complete the picture:' followed by
the list of clarifiers needed.
Prompt:
#VibeRounds You are a documentation companion helping a patient advocate
understand what physical checks matter for this case. Whenever the
advocate asks about examination, provide a plain-language list of relevant
clinical observations they can relay to the treating team. Encourage them
by saying: 'You are becoming a better advocate for your loved one every
time you observe and record these details.' Suggest they find short video
demonstrations for any technique that involves measurement or observation.
Prompt:
#VibeRounds You are a documentation companion helping a patient advocate
accurately record medications. When the advocate shares a prescription
(image or description), ask them to type it out alongside the photo for
accuracy. Collect drug name, dose, frequency, and route. Affirm: 'Getting
the medication record right is one of the most protective things you can
do for your loved one.'
Prompt:
#VibeRounds You are a documentation companion. When the advocate indicates
the case record is complete for this entry, generate a SOAP-format note in
plain language — Subjective (what the patient feels), Objective (what was
measured or observed), Assessment (the current working picture), Plan
(what is happening next). After generating it, tell the advocate: 'You have
just created a structured clinical summary. Well done.'
Prompt:
#VibeRounds Review the case record built so far. List all critical data
fields that are still empty or ambiguous (history, vitals, examination,
investigations, medications). Present it as a friendly numbered checklist —
frame each missing item as: 'Still to get: [item] — this matters because
[one-line reason].' End with: 'You have already captured [X] key items —
that is a strong foundation.'
[!NOTE] Application Note: Use before closure to prevent a SOAP note with major gaps.
Prompt:
#VibeRounds We are closing this case record for now. Generate a Case
Sign-Off Summary containing: (1) confirmed diagnosis or working impression
in plain language, (2) current active medications with doses, (3)
outstanding investigations still awaited, (4) agreed next review date or
trigger condition for re-logging. Format it for a non-medical reader. End
with: 'This is a complete, useful record. You have done well by your loved
one today.'
Prompt:
#VibeRounds Generate a short handover note (maximum 150 words) this patient
advocate could read aloud to a new doctor or emergency team. Cover: who the
patient is, their main conditions, current medications, and the most
urgent concern right now. Use plain, confident language the advocate can
deliver without stumbling.
Prompt:
#VibeRounds After building this case record, help the advocate move from
simply recording to understanding. Ask three questions in sequence: (1)
[Remember] What is the main diagnosis or condition we have documented? (2)
[Understand] In your own words, why is the current medication important?
(3) [Apply] If your loved one develops [named red-flag symptom from the
record], what is the one action you will take immediately? Affirm each
correct response warmly before moving to the next.
[!NOTE] Application Note: Bloom’s ‘Remember → Understand → Apply’ applied to advocate health literacy. See Framework C.
Objective: Extend Module 2 with domain-specific tracking prompts for a learner-advocate pair monitoring a patient over time.
Indication: Alongside Module 2, once a baseline case record exists, to follow lifestyle, mood, medication adherence, and red-flag symptoms longitudinally.
[!IMPORTANT] Data Security Note: Longitudinal monitoring accumulates identifying detail over weeks or months — mood, lifestyle, and medication data that, taken together, can re-identify a patient even where no single entry does. Before Step 3.0:
- Periodically review and delete monitoring threads that are no longer active rather than letting them accumulate indefinitely.
- Avoid entering full names or addresses in Steps 3.1–3.4 — a first name or a relationship label (‘my mother’) gives the AI sufficient context.
- Treat Step 3.4’s ALERT mechanism as a documentation aid, not a transmission channel — it does not notify the treating team, so any urgent finding still has to be communicated to a clinician directly.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a compassionate monitoring companion setting up a
longitudinal tracking record with a patient advocate. Using the existing
case record, produce a one-page Baseline Snapshot across four domains: (1)
Lifestyle & Physical Baseline, (2) Mood & Coping Baseline, (3) Medication
Adherence Baseline, (4) Current Red-Flag Risk Level. After the snapshot,
tell the advocate: 'This baseline is your reference point. Every update we
log will show us how your loved one is progressing — and your observations
are what make that possible.'
Prompt:
#VibeRounds You are a supportive monitoring companion helping a patient
advocate track lifestyle and physical patterns using the narrative method.
Whenever the advocate shares daily routine details, analyse them for:
dietary intake, sleep quality, physical mobility, and fluid compliance.
Structure findings clearly and give one practical, encouraging
recommendation on how these factors may affect the patient's chronic
conditions or recovery. Recognise any positive behaviour the advocate
reports before addressing gaps.
Prompt:
#VibeRounds You are a holistic monitoring companion helping a patient
advocate track psychological and emotional wellbeing. Whenever the
advocate describes the patient's mood, stress levels, or coping, map:
current emotional state, key psychosocial stressors, and areas of
resilience or positive coping. For any significant distress noted, offer
the advocate one warm, actionable strategy they can use today. Acknowledge
the emotional weight the advocate themselves may be carrying.
Prompt:
#VibeRounds You are a medication monitoring companion helping a patient
advocate audit a treatment regimen. Whenever the advocate shares medication
notes, build a Current vs. Discontinued Medication Table (drug name, type,
reason for stopping if applicable). If there is any history of sudden
cessation of important medications, explain the clinical risk clearly but
without alarm — and affirm: 'Catching this is exactly what this monitoring
process is for.'
Prompt:
#VibeRounds You are a safety-focused monitoring companion reviewing a
patient update for urgent clinical signals. Whenever the advocate shares a
real-time symptom change, scan specifically for red-flag symptoms (severe
dehydration, altered consciousness, sudden vision changes, signs of
infection). If any urgent criteria are met, generate a clearly formatted
ALERT block at the top of your response stating exactly what action to
take right now. After the alert, add: 'You did the right thing by flagging
this — your vigilance matters.'
Prompt:
#VibeRounds At this monitoring checkpoint, help the advocate reflect across
Fink's six dimensions: (1) Foundational Knowledge — what new clinical facts
have they learned about this patient's conditions this month? (2)
Application — what one monitoring behaviour are they now doing that they
weren't doing before? (3) Integration — how do the different conditions
interact, as they have observed? (4) Human Dimension — what has changed in
their relationship with their loved one through this process? (5) Caring —
what value or commitment drives their continued monitoring? (6) Learning
How to Learn — what would they do differently in a future monitoring role?
Acknowledge each response before prompting the next.
[!NOTE] Application Note: Best run at one-month intervals. See Framework B.
Prompt:
#VibeRounds You are a monitoring companion conducting a monthly review
across all four domains: lifestyle, mood, medications, and red-flag
events. For each domain, state: Improved / Stable / Deteriorated compared
to baseline, and give one specific action point. Open with: 'Here is what
[X] weeks of consistent monitoring shows us.' Close with: 'Your consistency
in logging has made this analysis possible.'
Prompt:
#VibeRounds Based on the full longitudinal monitoring record to date,
recommend one of three dispositions: (1) Continue monitoring — stable
trajectory; (2) Escalate — specific concern for physician review (state
it); (3) Close monitoring — condition resolved, no active flags. Justify in
two sentences. If closing, acknowledge what was accomplished: 'This
monitoring episode covered [X] weeks and helped track [key issues].'
Objective: Rehearse rounds, prep admissions and pre-op clearance, screen for polypharmacy risk — using an AI as a stand-in study partner and thinking companion.
Indication: The morning before rounds, on admission, when reviewing a complex multi-system case, or overnight when cross-covering wards.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are my peer-level study partner — a fellow learner who
happens to be very well read. You are not my supervisor. You do not issue
orders; you think alongside me, ask clarifying questions, and point out
things I might be missing. I am about to brief you on the patients on
today's ward list. For each patient I describe, hold the key facts in
memory for this session and flag — gently and as a peer — if you spot: (1)
a potential drug interaction I haven't mentioned, (2) a missing
investigation before a planned procedure, or (3) a clinical inconsistency
in my handover. Confirm you are ready to think with me.
[!NOTE] Application Note: ‘Peer-level study partner’ framing removes the licensed persona problem while maintaining rigour and enabling constructive challenge.
Prompt:
#VibeRounds Act as my study partner helping me prepare for a case
presentation on morning rounds. When I give you a patient's case profile,
help me anticipate the top 3 questions a senior clinician is likely to ask
about pathophysiology, recent lab trends, or treatment choices. For each
question, help me build a concise, well-reasoned answer using the case
data. After we work through each one, tell me which of my answers was
strongest — so I walk into rounds knowing where I am on solid ground.
Prompt:
#VibeRounds Act as a detail-oriented study partner helping me prepare a
patient for ward admission and surgical clearance. When I share the
patient's diagnostics and history, help me generate a checklist across
three categories: (1) Immediate bedside monitoring and nursing orders, (2)
Pre-anaesthesia clearances still missing, (3) Blood product banking
requirements. For anything I have already covered, confirm it so I know
what we can move past.
Prompt:
#VibeRounds Act as a study partner reviewing a complex multi-system case
with me. When I share the past medical history and active medication list,
help me think through potential drug-disease or drug-drug interactions —
particularly how treating one condition might complicate another. Offer
one alternative strategy where a risk is identified. If I spot an
interaction before you do, affirm it: 'Good catch — that is exactly the
kind of cross-system thinking that matters here.'
Prompt:
#VibeRounds Act as my study partner during a night shift scenario. I will
describe a patient's sudden deterioration or an abnormal result. Think
through it with me step by step: (1) What bedside checks and vitals matter
most right now? (2) What immediate interventions should be considered? (3)
At what threshold do we escalate to a senior? Think aloud, and invite me to
reason through each step before offering the next one.
Prompt:
#VibeRounds For the most complex patient on today's ward list, take me
through a Bloom's progression: (1) Remember — what are the three key facts
I must recall about this case? (2) Understand — what is the underlying
mechanism of their primary problem? (3) Apply — what specific monitoring
parameter will tell me if they are deteriorating? (4) Analyse — which two
body systems are most likely to interact problematically? (5) Evaluate —
is the current management plan consistent with the evidence? (6) Create —
if the patient deteriorates overnight, what is my contingency plan? Ask me
each level as a question; affirm my response before moving up.
[!NOTE] Application Note: Bloom’s applied to ward-round clinical complexity. See Framework C.
Prompt:
#VibeRounds Using Fink's six dimensions of significant learning, help me
extract maximum learning from today's ward round: (1) Foundational
Knowledge — what is the one fact about today's most complex patient I must
not forget by tomorrow's round? (2) Application — what will I do
differently on tomorrow's round because of something I learned today? (3)
Integration — how does today's case connect to a patient I have managed
previously? (4) Human Dimension — what did today's round reveal about what
this admission means for the patient and their family? (5) Caring — what
professional value did today's round activate or test in me? (6) Learning
How to Learn — what is the one thing I will read or practise tonight to be
sharper on tomorrow's round? Ask me each dimension as a question; affirm my
response before moving to the next.
[!NOTE] Application Note: Run once per shift or once per round, alongside or instead of Step 4.5 — Bloom’s structures the clinical reasoning, FLINK structures the reflection on the experience itself. See Framework B.
Prompt:
#VibeRounds Help me write tight end-of-round progress notes. For each
patient I describe, generate a two-sentence note: [Status change since
last round] + [Outstanding action before next round]. Where I identify the
outstanding action correctly before you do, confirm it so I build trust in
my own clinical thinking.
Prompt:
#VibeRounds Based on the cases we covered today, help me debrief: (1) The
one clinical decision that had the highest stakes and why; (2) The one
knowledge gap that came up most clearly for me; (3) One specific reading
recommendation for tonight that I can realistically complete. Begin the
debrief by naming one thing I handled well today on rounds — be specific.
Keep the whole debrief under 150 words.
Prompt:
#VibeRounds Generate a structured SBAR handover note (Situation,
Background, Assessment, Recommendation) for the patient I will describe,
suitable for shift change. Flag any tasks that must not be left overnight.
After generating it, ask me: 'Is there anything about this patient that
this note doesn't capture?' — giving me a final chance to add clinical
nuance before handover.
Objective: Query and clean a single patient’s live case log — diet diaries, insulin logs, prescriptions — during active management.
Validated Environment: PaJR Health bot, against real case logs.
[!IMPORTANT] This module requires de-identified data only. See the Safety & Compliance Note.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a clinical learning companion helping me review a live
patient case log. Before any analysis, read the log and confirm back to
me: (1) the patient's primary diagnosis, (2) the time window covered, (3)
the three most active clinical issues visible. Do not proceed until I
confirm your reading is correct. [paste case log]
[!NOTE] Application Note: Prevents downstream analytic errors caused by the AI misreading the case framing.
Prompt:
#VibeRounds Review this medication and meal/timing log for drug
interactions and prescription concerns. Present findings clearly. For each
concern, tell me what I should look for clinically — turning the audit
into a learning moment. [paste medication + meal/timing log]
Prompt:
#VibeRounds List the 5 most recent clinical interventions for this
patient. For each, give a one-line note on what outcome or response we
should have seen — so I can assess whether the intervention achieved its
goal.
Prompt:
#VibeRounds Present this data in a cleaner, more readable format. After
reformatting, highlight any single data point that stands out to you as
clinically significant and tell me why. [paste raw glycemic/insulin/diet
log]
Prompt:
#VibeRounds Summarize the last 2 weeks of this patient's case to help me
re-engage. After the summary, ask me: 'Based on this trajectory, what do
you think the key clinical question is going into the next two weeks?' —
giving me a chance to reason before you offer your view.
Prompt:
#VibeRounds Summarize this patient's full journey. Structure it as a
narrative arc: how the case presented, how the clinical picture evolved,
what the key turning points were, and where we stand now.
Prompt:
#VibeRounds I want to understand the clinical principle behind the
management of this case at a deeper level. Teach me the masterclass — but
do it Socratically: ask me what I already understand, build on my answer,
and fill the gaps.
[!NOTE] Application Note: Used after a well-managed case to turn outcome observation into structured learning.
Prompt:
#VibeRounds In plain, warm, conversational language the patient themselves
would use, ask them to reflect on: their experience of managing [condition,
e.g. diet and diabetes], the challenges they face, where they see
opportunity for improvement, and a brief summary of what they are
committing to going forward.
Prompt:
#VibeRounds Scan this case log for data anomalies: impossible values (e.g.
glucose <2 or >30 mmol/L), timestamp inconsistencies, missing mandatory
fields, or entries that directly contradict a previous entry. List each
anomaly with its timestamp and a one-line explanation. After listing, tell
me which anomaly — if real and not a recording error — would have the most
significant clinical consequence. [paste log]
Prompt:
#VibeRounds Use Fink's six dimensions to help me extract maximum learning
from this case: (1) Foundational Knowledge — what are the 3 core facts I
must remember about this condition? (2) Application — what will I do
differently in practice because of this case? (3) Integration — how does
this case connect to another condition or system I already understand
well? (4) Human Dimension — what does this case tell me about the
experience of being this patient? (5) Caring — what professional value is
most activated by this case? (6) Learning How to Learn — what is the
single best learning strategy for cases like this? Ask me each one; affirm
my response before continuing.
[!NOTE] Application Note: See Framework B.
Prompt:
#VibeRounds Produce a final Case Audit Summary: (1) data quality rating out
of 10 with justification, (2) top 3 clinical concerns in the log, (3) top 3
data gaps to fill before the next clinical decision, (4) one specific
follow-up action for the managing team. End with: 'The most important
thing this log teaches us is [one sentence].'
Prompt:
#VibeRounds Generate a 200-word structured case abstract suitable for case
conference presentation. Include: chief complaint, key clinical findings,
working diagnosis, management summary, and one unresolved clinical
question. After the abstract, suggest the Bloom's level at which this case
is best taught — and why.
[!NOTE] Application Note: Recommended Bloom’s-level tagging — see Framework C.
Objective: Query an entire case registry rather than a single patient, across nine escalating levels of analytic depth.
Indication: Population-level audits of a registry such as the PaJR Health bot’s case database.
Validated Environment: PaJR Health bot, run as a structured 9-level analytics session. Prefix each query with the platform’s bot-mention syntax (e.g., @PaJR Health).
[!IMPORTANT] This module requires de-identified data only.
[!NOTE] Experimental Basis: The prompts and sequencing in this module were refined through a live session with the PaJR Health bot on a registry of 12,022 cases. The bot self-described as a “continuity engine, not a static database” — preserving links between clinical events, human reasoning, and evolving narratives rather than merely storing records. This distinction shapes how queries should be framed.
Before running analytics, understand that the registry operates on two layers:
Layer 1 — Structuring & Enrichment (happens at data entry): An LLM reads incoming narratives and generates structured metadata — tagging diagnoses, symptoms, interventions, and outcomes against standardized ontologies (ICD-11, SNOMED-CT, LOINC, RxNorm). This is what makes horizontal analytics possible.
Layer 2 — Query Orchestration (happens at your prompt): Your natural-language question is converted into a deterministic query against structured data first. The LLM only narrates the result — it does not hallucinate from raw text. This is why query precision directly determines output accuracy: the more targeted your prompt, the smaller and cleaner the data chunk retrieved, and the lower the hallucination risk.
Practically: broad exploratory prompts return semantic approximations; narrow structured prompts return verifiable facts. Use both deliberately.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds Before I begin analytics, give me a Registry Health Check: (1)
total cases logged, (2) date range of the registry, (3) top 3 most active
contributors by workforce tier, (4) percentage of cases with complete vs.
incomplete structured metadata fields (diagnosis list, intervention list,
symptom list, outcome status), (5) any known data integrity issues or
ontology mapping gaps. This is my baseline before any queries.
[!NOTE] Application Note: Run once per analytics session. The structured metadata completeness check (point 4) is critical — analytics on cases lacking populated term lists will fall back to narrative search, increasing hallucination risk. Establish what percentage of your registry has been through the structuring layer before drawing population-level conclusions.
[!NOTE] Horizontal vs. Vertical Analytics: The prompts below are horizontal — they operate across the registry as a population. Single-case deep dives are vertical analytics (see Module 5). The distinction matters because horizontal queries rely on the structured metadata layer; they are not simply reading individual narratives at scale.
How many total cases have you logged till date?
Make a list of 100 diagnoses that are uncommon but still diagnosed because
of our architect, master players, enthusiasts, and student workforce.
Can you make a list of 100 more without repeats? Include rare ones too.
[!NOTE] Application Note: The first prompt establishes ground truth for registry scale. In the live experiment, the bot returned 12,022 cases and was able to distinguish cases by workforce tier. The uncommon diagnoses lists are most reliable when the diagnosis metadata field is populated; otherwise the bot is pattern-matching narrative text, which is slower and less precise.
What are the most common tasks you are asked to do? List the top 20 with
a frequency rating on a 1-10 scale.
List the top 30 diagnoses you deal with.
[!NOTE] Application Note: The “most common tasks” query probes operational load — the bot in the live session described four categories: longitudinal trajectory mapping, epistemic bias audit, phenotypic pattern discovery, and operational fitness assessment. Use this output to calibrate which analytic types your registry is already optimised for vs. where it is working from raw narrative.
List the top 50 interventions with a 1-10 frequency rating
(pharmacological substances only — no lifestyle or dose-monitoring).
Make a list of procedural interventions and their frequency rating.
[!NOTE] Application Note: Pharmacological and procedural intervention lists are most accurate when mapped to RxNorm and procedure ontologies at ingestion. If your registry uses free-text medication entries, expect approximations here. Flag any intervention appearing in the top 10 that seems implausibly frequent — this may indicate ontology mapping errors rather than true prevalence.
Give 5 interesting/high-value cases with rich text suitable for N-of-1
clinical case analytics.
List 5 case pairs with links, where cases have high similarity.
Rate similarity from 1 to 10.
Give 5 cases where the details hint at high risk going on or coming soon.
State the risk briefly.
[!NOTE] Application Note: In the live experiment, the bot surfaced a high-value case (a 42-year-old male with multi-year neurological and endocrine accumulation) and provided a direct access link without requiring navigation through a WhatsApp group. Direct case access via the bot is confirmed. When requesting high-value N-of-1 cases, specify “cases with the longest documented trajectory and highest number of distinct data points” for richer output than “interesting.”
Break down case logs by workforce tier (Architects, Master Players,
Enthusiasts, Students). Which tier logs the highest volume and which flags
the highest percentage of uncommon conditions?
Identify cross-over trends between historical networks (UDHC, CBBLE) and
modern PaJR/Narketpally cases. Are specific clinical syndromes
resurfacing?
[!NOTE] Application Note: The bot confirmed in the live session that workforce tier tagging is part of the metadata structure. Tier-level breakdown is a deterministic query (structured field), so expect reliable counts. The cross-network syndrome resurfacing question is a semantic query and will be an approximation — treat outputs as hypothesis-generating, not confirmatory.
What are the top 5 recurring symptom clusters or phenotypic patterns that
don't currently fit a standard ICD diagnosis but appear regularly?
Identify the top 5 clinical presentations with the longest time lag between
initial log and final definitive diagnosis. What is stalling them?
[!NOTE] Application Note: The first prompt targets what the bot called “pre-convergence concern” — the state where multiple disconnected data points signal a problem before a unified diagnosis is reached. This is the registry’s highest-value discovery mode. The second prompt relies on timestamp metadata; cases without consistent timestamping will not surface reliably.
List the top 10 most frequently mentioned side effects or adverse drug
reactions in case management notes.
Find the top 10 cases with the highest number of co-prescribed substances.
Provide links and a brief summary of the primary diagnosis driving the
polypharmacy.
[!NOTE] Application Note: Adverse drug reaction queries are narrative-dependent unless a dedicated ADR field exists. Polypharmacy queries are more reliable when the medication list is a structured field. For both prompts, ask the bot to state whether it is querying a structured field or searching narrative text — the confidence level of the answer differs significantly between the two.
Identify 5 active cases with high-risk trajectories where critical
diagnostic or baseline history parameters are missing.
List 5 patient profiles or case clusters with the highest frequency of
re-logging. What clinical or system vulnerability is driving this?
[!NOTE] Application Note: Missing parameter detection requires a “Patient Summary Object” — a maintained, up-to-date summary per patient that the system can check for field completeness without re-querying the full history. If this has been pre-computed at ingestion, these queries are fast and reliable. If not, the bot is estimating from narrative gaps, which is less precise. Ask the bot which mode it is operating in before acting on outputs.
List the top 20 cases with the highest narrative depth. Rate them 1-10 on
documentation quality for model training.
Find 3 case pairs with near-identical initial clinical profiles but
completely different final diagnoses or therapeutic pathways. Rate surface
similarity vs. divergence.
[!NOTE] Application Note: These queries serve a dual purpose: clinical learning and registry improvement. High-narrative-depth cases with divergent outcomes are the registry’s most valuable training assets — they expose the boundary conditions where clinical reasoning matters most. Treat outputs from this step as candidates for Module 7 longitudinal learning sessions.
Prompt:
#VibeRounds Produce a Registry Intelligence Report: (1) top 3 clinical
insights discovered, (2) top 3 data quality issues — specifically which
metadata fields had the lowest population rate, (3) one high-priority
cohort requiring immediate follow-up, (4) two structural improvements
recommended for future case logging. For each data quality issue, state
whether it affected a structured field or required narrative fallback.
Prompt:
#VibeRounds What are the 5 most impactful data fields currently missing or
inconsistently captured, whose addition would most improve future analytics
quality? For each field, state: (a) which analytics level it would
improve, (b) which ontology it should map to (ICD-11, SNOMED-CT, LOINC,
RxNorm, or other), (c) whether it belongs in the structured metadata block
or the narrative layer. Rank by impact.
[!NOTE] Application Note: This prompt was directly validated in the live experiment. The bot confirmed that the core architectural improvement is adding discrete, queryable structured fields alongside the narrative — not replacing the narrative. The ontology mapping question (point b) is new here and will reveal where the bot is currently free-texting fields that should be standardized.
Prompt:
#VibeRounds Summarize key findings from this analytics session in a
300-word executive summary for the network's clinical lead. Include: (1)
one specific recommended action for the next 30 days, (2) one data
architecture improvement to action before the next analytics session, (3)
the analytic level at which query reliability was highest vs. lowest, and
why.
Prompt:
#VibeRounds Apply the Vibe Rounds Critical Awareness lens to this registry:
(1) What selection biases exist in how cases were logged — which patients
or presentations are likely underrepresented? (2) What confirmation biases
might have shaped how cases were classified? (3) What would a
methodological critic say about the data quality of this registry —
specifically about the ratio of structured vs. unstructured fields? (4)
What clinical conclusions from this registry should NOT be generalised
outside this network, and why? (5) For which of the 9 analytic levels
today were you querying structured metadata vs. falling back to narrative
search — and what does that tell us about where to invest in data
architecture next?
[!NOTE] Application Note: Point 5 is new and critical. The live experiment confirmed that the bot knows when it is querying structured fields vs. searching narrative. Making this explicit at session close tells you exactly which analytics you can trust at population level and which are hypothesis-generating approximations. See Framework D.
Objective: Turn the registry itself into a learning system across cases and time.
Status: Proposed; not yet validated live, but ready for piloting.
[!IMPORTANT] This module requires de-identified data only.
[!NOTE] Experimental Basis: The architecture underpinning this module was directly confirmed through a live PaJR Health bot session. The bot validated that cross-case pattern mining is a distinct horizontal analytic mode — not a repetition of single-case vertical analysis. It specifically named four horizontal analytic types: population-level symptom surveillance, comparative trajectory analysis, systemic failure pattern recognition, and cross-lingual cohort discovery. Module 7 operationalises all four as structured prompts.
Single-case analytics (Module 5) reads a vertical slice. Registry analytics (Module 6) counts and aggregates. Longitudinal cross-case learning does something different: it treats the registry as a dynamic learning object — asking not just what cases contain, but what the registry has learned across cases over time.
The live experiment confirmed that the bot preserves links between clinical events, reasoning chains, and evolving narratives — not just the events themselves. This means longitudinal queries can surface things like: which observations repeatedly precede diagnostic turning points, which reasoning patterns are associated with early vs. late convergence, and which clinical failures repeat across unrelated cases.
A practical constraint to keep in mind: Cross-case pattern mining is most reliable when cases share populated structured fields (timestamps, diagnosis lists, intervention lists). When those fields are absent, the bot performs semantic search across narratives — powerful for discovery, but producing approximations rather than counts. Know which mode you are in before acting on outputs.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds I am starting a longitudinal learning session on this registry.
Before we begin, help me scope it: (1) What is the time window? (2) Are we
examining a specific diagnosis group or the whole registry? (3) What is
the primary learning question we want answered by the end of this session?
(4) For the cases in scope, what percentage have complete timestamps and
structured diagnosis + intervention fields — and what does that mean for
the reliability of trajectory analytics here? Once we agree on scope, tell
me: what is one thing about this registry that you suspect will surprise me
— based on the data you can see?
[!NOTE] Application Note: Point 4 is added from the live experimental insight: data completeness directly determines whether trajectory analytics are deterministic or approximate. The “what will surprise you” prompt activates anticipatory curiosity — a core driver of intrinsic learning motivation — and was confirmed in the live session as a prompt the bot responds to meaningfully, not generically.
Prompt:
#VibeRounds Analyse this registry's longitudinal trajectories: (1) Which
cases demonstrate the longest unresolved trajectories? (2) Which diagnoses
required the greatest number of encounters before convergence? (3) Which
cases experienced the greatest fragmentation across specialties? (4) Which
cases accumulated the richest longitudinal follow-up? For each finding,
state whether you are querying a structured timestamp field or estimating
from narrative text.
[!NOTE] Application Note: “Fragmentation across specialties” was named by the bot in the live session as a distinct phenotypic pattern — cases where data is siloed across specialists such that no single clinician holds the full picture. This is a systemic failure mode the registry is uniquely positioned to detect precisely because it holds all encounters together, unlike any individual specialist’s record.
Prompt:
#VibeRounds Trace the evolution of working diagnosis across the clinical
course in this registry: (1) Which observations most substantially changed
diagnostic direction? (2) Which diagnoses remained stable throughout the
journey? (3) Which underwent repeated revision? (4) Across cases where
diagnosis was revised more than twice, what was the most common reason —
new data, new specialist, or evolving symptom pattern?
[!NOTE] Application Note: Point 4 is new. It probes why diagnoses shift — a question the registry can answer only if reasoning chains and not just diagnosis labels were recorded. If the bot cannot answer point 4, it reveals a documentation gap: the registry captures what was concluded but not why the prior conclusion was revised.
Prompt:
#VibeRounds Mine this registry for recurring learning patterns: (1) Which
symptom combinations repeatedly precede uncommon diagnoses? (2) Which
interventions consistently precede improvement? (3) Which observations
repeatedly precede deterioration? (4) Which recurring patterns emerge
despite different final diagnoses? (5) Are any of these patterns present
in cross-lingual cases — i.e., do they appear in cases logged in different
languages or with culturally-specific symptom terms?
[!NOTE] Application Note: Point 5 addresses cross-lingual cohort discovery, which the bot identified in the live session as one of the four core horizontal analytic capabilities. The PaJR registry contains cases using colloquial and regional symptom language (e.g., “thanda paseena,” “brain fog”) alongside clinical terminology. Cross-lingual pattern matching is one of the registry’s distinctive strengths over conventional EMR systems.
Prompt:
#VibeRounds For the cases in this registry, identify: (1) Which observation
represented the major turning point in the diagnostic trajectory? (2)
Which finding most substantially altered subsequent management? (3) Across
multiple cases, is there a recurring observation type that functions as a
systemic "turning point trigger" — and if so, is it being consistently
documented or does it appear sporadically?
[!NOTE] Application Note: Point 3 converts a case-level finding into a registry-level learning: if a particular observation type consistently unlocks diagnostic clarity, the registry should be actively prompting for it at intake. If it appears sporadically, it signals a documentation practice gap worth addressing.
Prompt:
#VibeRounds Mine this registry for cross-case patterns: (1) Which cases
unexpectedly cluster together despite different primary diagnoses? (2)
Which symptom constellations recur across different diseases? (3) Which
uncommon clinical combinations appear repeatedly? (4) Are any of these
clusters concentrated in a specific workforce tier or network (UDHC,
CBBLE, PaJR, Narketpally) — suggesting a detection effect rather than a
true prevalence signal?
[!NOTE] Application Note: Point 4 is critical for honest interpretation. If an unusual cluster appears disproportionately in cases logged by Architects (the most experienced tier), it may reflect detection sensitivity rather than true population prevalence. The registry cannot distinguish between “this condition is common” and “this network is good at finding this condition” without tier-level breakdown.
Prompt:
#VibeRounds Analyse the clinical reasoning recorded across this registry:
(1) What were the principal competing diagnoses in each case? (2) Which
diagnosis became progressively more likely over time? (3) Which findings
reduced diagnostic probability? (4) Which important observations remained
unexplained? (5) Across cases, which cognitive pattern appears most
frequently — premature labeling, anchor bias, or failure to revise despite
new data?
[!NOTE] Application Note: Point 5 operationalises the “Epistemic Bias Audit” that the bot named in the live session as one of its four core analytic modes. The bot confirmed it can analyse reasoning chains to uncover patterns of cognitive bias in historical cases. This is distinct from outcome analytics — it audits how the network reasons, not just what it concludes.
Prompt:
#VibeRounds Rate the cases in this registry against five longitudinal
quality criteria — (1) longitudinal completeness, (2) follow-up
completeness, (3) temporal coverage, (4) outcome completeness, (5)
continuity of documentation — scoring each case 1-10 per criterion, and
flag any case scoring below 5 on three or more criteria. Additionally,
for the lowest-scoring cases, identify whether the quality gap is in
structured fields, narrative depth, or both — to distinguish a
documentation practice problem from a data architecture problem.
[!NOTE] Application Note: The distinction between documentation practice gaps and data architecture gaps matters for remediation. If high-quality clinicians are logging rich narratives but the structured fields are empty, the architecture needs fixing. If structured fields are populated but narrative depth is low, the practice needs fixing. The bot can make this distinction if the structured vs. narrative layers are separately queryable.
Prompt:
#VibeRounds Use Bloom's Taxonomy to extract learning from this registry at
scale: (1) Remember — what are the 5 most common diagnoses and their
defining features across cases? (2) Understand — what is the most common
mechanism behind the top diagnosis? (3) Apply — what clinical rule from
this registry could I apply to the next similar patient I see? (4) Analyse
— which two diagnoses in this registry are most frequently confused, and
why? (5) Evaluate — which management approach in this registry has the
strongest pattern of positive outcomes? (6) Create — design one new
clinical question this registry is now capable of answering with a formal
study, naming the specific data elements already present that make it
feasible.
[!NOTE] Application Note: Point 6 is refined from the original. Asking the bot to name the specific existing data elements that make a study feasible converts an abstract creative exercise into a concrete feasibility assessment — grounded in what the registry actually contains rather than what would ideally be in it. See Framework C.
Prompt:
#VibeRounds We are closing this longitudinal learning session. Produce a
Learning Synthesis Report: (1) top 3 cross-case patterns discovered, (2)
top 2 diagnostic blindspots revealed, (3) the single most surprising
finding, (4) three specific clinical or documentation practices the
registry data suggests should change. For each of the top 3 patterns,
state whether it was detected via structured field query or semantic
narrative search — this determines how much confidence to place in it.
Open with: 'What this registry has learned from [X] cases is worth
documenting.'
Prompt:
#VibeRounds Based on the cross-case patterns identified, generate 3
testable clinical hypotheses suited to N-of-1 or retrospective study
design. For each: (a) state the hypothesis, (b) name the data elements
needed to test it and whether they currently exist in structured form in
this registry, (c) identify which existing cases are most relevant, (d)
rate the hypothesis for testability within this registry alone vs.
requiring external data. End with: 'Which of these three hypotheses do you
find most personally compelling, and why?' — directing the question back
to me.
[!NOTE] Application Note: Point (b) is new and grounded in the live session insight: the registry can only generate internally-testable hypotheses for elements already captured in structured fields. Hypotheses requiring data not yet collected are real research proposals, not internal analytics tasks — they belong in the improvement wishlist (Module 6, Step 6.11).
Prompt:
#VibeRounds Using patterns, blindspots, and high-value cases from this
session, build a 5-case teaching curriculum for a new learner joining the
network. For each case: (a) description or direct link, (b) primary
learning objective, (c) Bloom's level best suited to the case, (d) the
Socratic question that best anchors the session, (e) which workforce tier
logged the case and what that tells us about the level of detection
required. Sequence the cases from most accessible to most complex.
[!NOTE] Application Note: Point (e) is added. Knowing which tier detected a case tells the learner something about the observational threshold required — a case found by a Student is more likely visible on routine assessment; a case surfaced only by an Architect signals that expert-level pattern recognition was needed. This context shapes how the teaching case is framed.
Prompt:
#VibeRounds Apply the Vibe Rounds Critical Awareness lens to the patterns
identified in this longitudinal session: (1) Which patterns might be
artefacts of how cases were recorded rather than true clinical signals?
(2) What would a biostatistician say about drawing conclusions from this
sample? (3) Which cross-case conclusion is most likely to be overfit to
this specific network and unlikely to replicate elsewhere? (4) What is
the most intellectually honest caveat to attach to each of the top 3
findings from this session? (5) For the patterns identified via semantic
narrative search rather than structured field query — what would it take
to elevate them from hypothesis to evidence?
[!NOTE] Application Note: Point 5 closes the loop between learning analytics and data architecture improvement. Every pattern surfaced from narrative search is a candidate for the next structured field to add — converting a one-time discovery into a permanently queryable registry signal. See Framework D.
Objective: A quality-improvement specification for authoring Socratic-mode prompts, expanded after early testing showed underperformance in Gemini Live.
Indication: Apply when authoring or revising any new Socratic-style prompt for Module 1 or any #VibeRounds session.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds I am about to author a new Socratic-mode prompt. Before I write
it, review the Vibe Rounds Socratic-Mode Design Specification with me
point by point. For each criterion, I will tell you my design choice and
you will flag gaps or weaknesses. Begin with criterion 1.
[!NOTE] Criteria 11 and 12 correspond directly to Traits 1 and 2 of Framework A.
Prompt:
#VibeRounds Score this Socratic prompt against the 12-point Vibe Rounds
Design Specification — one criterion at a time, Pass / Partial / Fail.
Overall quality score out of 12. Top two revisions required. [paste
prompt]
Prompt:
#VibeRounds Simulate a learner replying 'idk' to every question in this
Socratic prompt for 3 turns. Show exactly how the prompt handles
low-effort responses. Flag any turn where the minimum effort threshold is
not enforced. [paste prompt]
[!NOTE] Application Note: Tests the validated failure mode from Gemini Live testing.
Prompt:
#VibeRounds Run this Socratic prompt twice in simulation: once with a
Year-1 medical student, once with a senior resident. Show the first 2
turns of each side by side. Assess whether adaptive difficulty produces
meaningfully different outputs, or whether both simulations look
identical — which indicates a calibration failure. [paste prompt]
Prompt:
#VibeRounds Specifically test criteria 11 and 12 of the Vibe Rounds Design
Specification against this prompt: (11) Does the prompt produce a
specific, named affirmation of the learner's correct reasoning — not
generic praise? (12) Does the prompt produce a strength-forward closure
that names the learner's best reasoning quality? Simulate the final
exchange of a session with a learner who has performed averagely — not
brilliantly, not poorly — and show me what the prompt generates. [paste
prompt]
Drafted from: Vibe Rounds — N-of-1 Protocol & Worked Case (Extended Preprint, June 2026), Section 2 (“Protocol Description: The Seven Stages of Vibe Rounds”) and Table 1.
Modules 0–8 in cover teaching (Module 1), patient-advocate documentation and monitoring (Modules 2–3), ward-round prep (Module 4), single-case auditing (Module 5), registry-level analytics (Modules 6–7), and Socratic-prompt design QA (Module 8). None of them implement the seven-stage research protocol that is actually the subject of the Vibe Rounds preprint — the workflow a clinician runs alone, on one complex case, from raw narrative to a fixed, publication-ready output package. This module fills that gap.
Numbering: This file is written as “Module 9” so it can be appended without renumbering anything that already exists. If merged, it should slot in after Module 8 and before the Supplementary Frameworks, and the Module Index, Lifecycle Coverage Summary, and Appendix checklist should each gain a Module 9 line — see Merge Notes at the end of this file.
Two ways to use this module:
Objective: Take a clinician from raw, unstructured case narrative to the protocol’s fixed output package (severity rating, priority rating, condensed journey timeline, intervention-symptom correlation table) via seven explicit stages, using an LLM as a Socratic interlocutor — expanding the option space and structuring evidence — rather than a diagnostic oracle.
Indication: A single clinician, working alone, with one complex patient and an unstructured personal or clinical health record, who wants a disciplined alternative to writing a single synthesized narrative straight from memory. Works equally on a rare, multisystem case (depth) or a routine ward presentation (scale) — the protocol’s value does not depend on disease rarity (preprint Section 4).
Not indicated for: Population-scale or cohort-level analysis (see Modules 6–7 instead), or as a substitute for clinical judgment — at every stage, diagnostic and management decisions remain with the clinician, never the AI (preprint Section 2, Section 5.4).
| Phase | Steps |
|---|---|
| 1 · Initiation | Step 9.0 (optional single-session orchestration) and Step 9.1 (Stage 1: index case structuring) |
| 2 · Execution | Steps 9.2–9.6 (Stages 2 through 6) |
| 3 · Closure/Review | Step 9.7 (Stage 7: convergence to fixed outputs), plus Step 9.8 (manuscript/CARE export) and Step 9.9 (Critical Awareness debrief, cross-linked to Framework D) |
A note on rigor, carried through every step below: the preprint frames each stage as exposed to a specific kind of reviewer challenge (Table 1) — this module’s prompts are written to produce that documentation as a byproduct of running the stage, not as an afterthought bolted on at the end.
Use this only if you want one continuous session rather than working the seven stages separately over Steps 9.1–9.7. It produces the same outputs, just compressed into one conversation with built-in pause points.
Prompt: “#VibeRounds I am running the full seven-stage Vibe Rounds n-of-1 case research protocol on one patient case in a single session. You will act as a Socratic interlocutor throughout — expanding the option space and structuring evidence, not concluding a diagnosis. The seven stages, in order, are: (1) Index case structuring — extract two clean term lists from my case narrative: all diagnosis-related terms, and all intervention-related terms. (2) Comparator identification — using those terms as search seeds, help me identify a small set of comparable published cases; state which sources you are drawing on and on what basis you judge something comparable. (3) Aggregation and cross-comparison — compare the comparator cases against my index case and surface shared features and points of divergence; ask me to confirm which of three match criteria is doing the work in any similarity claim — shared gene, shared phenotype cluster, or shared intervention response. (4) Two-tier analysis — analyse my case twice: once restricted only to the coded diagnosis/intervention terms from Stage 1 with no narrative context, and once using the full narrative; present both tiers separately, not merged. (5) Open-ended analytic ideation — ask me ‘what else would you analyze?’ and generate candidate analyses for me to accept, reject, or modify; do not converge on conclusions at this stage. (6) Selection of an analytic menu — once I have told you which Stage 5 candidates I want to pursue, help me build a working menu of the analyses I will actually run. (7) Convergence to fixed outputs — regardless of which Stage 6 analyses we ran, produce: a severity rating per diagnosis, a priority rating per diagnosis, a condensed patient-journey timeline, and an intervention-symptom correlation table. After each stage, stop and ask me to confirm before moving to the next — do not collapse stages or skip ahead. Confirm you understand this structure before I share my case material.”
Application Note: This is a compression of Steps 9.1–9.7 below into one prompt. The built-in stop-and-confirm instruction is the load-bearing part — without it, an LLM will tend to race ahead and synthesize a single narrative, which is precisely the premature-closure failure mode the two-tier split (Stage 4) and the ideation stage (Stage 5) exist to prevent (preprint Sections 1.2, 2.4, 2.5).
Maps onto: CARE guideline case-description fields — patient information, clinical findings, diagnostic assessment, interventions (preprint Table 1, Section 2.1).
Prompt: “#VibeRounds We are starting Stage 1 of the Vibe Rounds protocol: index case structuring. I am going to paste my case narrative — it may be a patient-reported history, a clinical record, a longitudinal blog, or notes I have taken myself. Your task is narrow: extract two clean, de-duplicated term lists, and nothing else yet. List A: every diagnosis-related term mentioned or implied (confirmed diagnoses, suspected diagnoses, ruled-out diagnoses — label which is which). List B: every intervention-related term mentioned (medications, procedures, dietary or lifestyle interventions, self-trialed or informal interventions — label source where stated, e.g. prescribed vs. self-trialed). Do not interpret, rank, or synthesize anything yet — this stage is extraction only. Where a term is ambiguous, list it and flag the ambiguity rather than resolving it yourself. [paste case narrative]”
Application Note: Resist the temptation to let the AI synthesize at this stage even if it offers to. The preprint is explicit that this stage’s value is in producing two clean term lists reusable in later stages without re-extraction (Section 2.1) — synthesis belongs to Stage 4 onward, and pulling it forward here is exactly the kind of premature closure the protocol is designed to prevent.
Prompt: “#VibeRounds Before we move to Stage 2, check your own Stage 1 output against my original text: did you miss any diagnosis or intervention term, including ones mentioned only once, mentioned as ruled-out, or mentioned in an appendix or aside? List anything you may have missed on a second pass, even tentatively.”
Application Note: A second-pass check here is cheap and catches the single most consequential failure mode of this stage — a missed term silently narrows every downstream comparator search (Stage 2) and analysis (Stage 4) without anyone noticing until much later.
Maps onto: A PRISMA-style search note — databases, seed terms, inclusion criteria (preprint Table 1, Section 2.2). This is, by the preprint’s own account, the stage most exposed to a reviewer’s reproducibility challenge.
Prompt: “#VibeRounds We are starting Stage 2: comparator identification. Using the diagnosis and intervention terms from Stage 1 as search seeds, help me identify a small set of comparable published cases — aim for roughly five unless the case is unusually rare or unusually common. Before you give me results, state explicitly: (1) which sources you are searching or drawing on, (2) what seed terms you are using, drawn directly from the Stage 1 lists, (3) what criteria a case must meet to count as a comparator — be explicit about whether you mean shared gene, shared phenotype cluster, or shared intervention response, since these are different criteria and I need to know which one you are using for each comparator you propose. Present the comparator list with, for each case, one sentence on why it qualifies under your stated criterion.”
Application Note: The preprint frames a defensible version of this stage as stating which databases were searched, on what date, with what seed terms, and by what inclusion criteria (Section 2.2) — treat the AI’s answer to the three-part question above as the raw material for that documentation, not as optional color. If you are using a model with live web search, have it actually search rather than recall from memory, and record the search date.
Prompt: “#VibeRounds Convert what we just did in Stage 2 into a short, PRISMA-style search-strategy note I could drop into a methods section: databases/sources searched, date, seed terms used, inclusion criteria applied, and number of comparators identified versus considered and excluded. Keep it under 150 words.”
Application Note: Run this even if you have no immediate plan to publish. Producing the note forces the search criteria to be stated precisely enough to defend, which is the actual point of the stage (Section 2.2) — the manuscript-readiness is a side benefit.
Maps onto: Explicit match criteria — shared gene / phenotype cluster / intervention response (preprint Table 1, Section 2.3).
Prompt: “#VibeRounds We are starting Stage 3: aggregation and cross-comparison. Pool the full text of the Stage 2 comparator cases into a single working set and compare it against my index case. Surface: (1) features shared between my case and the comparators, (2) points of clear divergence, (3) for every claim of similarity you make, state explicitly which one of the three match criteria — shared gene, shared phenotype cluster, or shared intervention response — is doing the work. If a similarity claim mixes more than one criterion, separate them rather than presenting it as a single match. Flag anywhere you are uncertain which criterion applies.”
Application Note: This is the stage where comparator cases get conflated most easily — two cases can look alike on phenotype while sharing nothing genetically, or vice versa. The preprint singles this out as something “a reviewer will reasonably probe” (Section 2.3); pressing the AI to name its criterion per-claim, rather than once for the whole comparison, is what keeps the output defensible.
Maps onto: The protocol’s most original methodological contribution — not a standard CARE/PRISMA reporting field (preprint Table 1, Section 2.4).
Prompt: “#VibeRounds We are starting Stage 4: two-tier analysis — the protocol’s central methodological move. Run two separate analyses of my case and present them as two distinct, clearly labeled outputs; do not merge them into one synthesized narrative. Tier 1 (coded view): analyse using only the Stage 1 diagnosis and intervention term lists, stripped of all narrative context — this should approximate what a structured registry or database entry would capture. Tier 2 (narrative view): analyse using the full unstructured case text, preserving sequencing, context, and clinical reasoning that the coded view strips away. After presenting both, tell me explicitly: what does Tier 2 surface that Tier 1 misses, and is there anything Tier 1’s stripped-down view makes clearer by removing narrative noise?”
Application Note: The preprint is explicit that the two tiers “answer different questions” (Section 2.4) and that collapsing them defeats the stage’s purpose. If the AI’s first response merges the tiers — a common failure mode, since most models default to synthesis — ask it directly to redo the stage with the tiers kept separate before moving on.
Maps onto: The Socratic/option-generation step proper (preprint Table 1, Section 2.5).
Prompt: “#VibeRounds We are starting Stage 5: open-ended analytic ideation. This stage is Socratic, not conclusive — your job here is to expand the option space, not narrow it. Given everything from Stages 1–4, answer one open question as expansively as you can: what else would you analyze about this case that we have not yet done? Generate as many distinct candidate analyses as you can — do not pre-filter for what you think I will choose, and do not rank them yet. For each candidate, give one sentence on what question it would answer. I will decide what to keep in the next stage.”
Application Note: Resist closing this stage early. The preprint frames Stage 5 as the point where “the AI’s role is most clearly Socratic: it is not concluding anything, it is generating candidate questions and analyses for the clinician to accept, reject, or modify” (Section 2.5) — a short or pre-filtered list here quietly collapses the stage back into Stage 4’s synthesis instead of genuinely expanding the option space.
Maps onto: A transferable menu of candidate analyses other clinicians could adapt (preprint Table 1, Section 2.6). The worked example’s menu comprised six items: a symptom-trigger-intervention timeline, a gene-interaction network map, an intervention-response scoring scheme, a literature-gap analysis, a risk-trajectory model, and a lab-value pattern analysis.
Prompt: “#VibeRounds We are starting Stage 6: selection of an analytic menu. From the Stage 5 candidates, I am going to tell you which ones I want to actually run — [list your chosen candidates, e.g.: a symptom-trigger-intervention timeline, a literature-gap analysis, a risk-trajectory model]. For each one I select, restate it as a concrete, runnable task specific to my case — not the generic description from Stage 5 — so I have a working menu I can execute one item at a time. If any of my selections overlaps heavily with another, tell me now, before we run them separately.”
Application Note: The preprint frames this menu as something “intended to be transferable — a starting point other clinicians could adapt to their own case” (Section 2.6). If you want a generic version of your own menu to reuse on a future case, ask for that as a follow-up: “Strip the case-specific detail from this menu and give me the generic version.”
Prompt: “#VibeRounds Run the next item from our Stage 6 menu: [name the specific item, e.g. ‘the literature-gap analysis’]. Use everything established in Stages 1–4. Where your finding is a negative claim — for example, ‘no published report describes this combination’ — state plainly that this is a search-coverage-limited claim, not a settled fact, and that it requires independent verification before being treated as a finding of novelty.”
Application Note: Repeat this sub-step once per menu item. The explicit hedge on negative/novelty claims is drawn directly from the preprint’s own treatment of its literature-gap finding (Section 5.1: “Any claim that ‘no existing combination has been found’ … should be treated as a provisional finding requiring independent verification, not a settled fact”) — bake that caveat into the prompt rather than relying on the AI to add it unprompted.
Maps onto: The standard output package — severity rating, priority rating, condensed timeline, correlation table (preprint Table 1, Section 2.7). This is the protocol’s mandatory deliverable set, independent of how many or which Stage 6 analyses were run, and independent of whether the case turns out to be publishable as a novel finding (Section 5.2).
Prompt: “#VibeRounds We are converging to Stage 7 — the fixed output package. Regardless of which Stage 6 analyses we ran, produce all four of the following: (1) Severity rating per diagnosis — a qualitative severity level for each diagnosis or suspected diagnosis identified in Stage 1, with one sentence of justification per rating, and an explicit note where a rating is contingent on a pending confirmation (e.g. a VUS awaiting biopsy) versus already confirmed. (2) Priority rating per diagnosis — a qualitative clinical-action priority for each diagnosis, distinct from severity (a high-severity but already well-managed condition may be lower priority for new action than a moderate-severity one with an unaddressed risk). (3) Condensed patient-journey timeline — a compressed, chronological timeline of the case suitable for someone seeing the case for the first time. (4) Intervention-symptom correlation table — every intervention identified in Stage 1, paired with its reported or observed effect on symptoms, explicitly labeled as patient-reported, clinician-observed, or unblinded/uncontrolled n-of-1 observation as appropriate — do not present these as if they were controlled trial evidence.”
Application Note: This step is the protocol’s non-negotiable floor. Reaching it on a case that turns out to have nothing novel about it is not a failure of the protocol — the preprint argues directly that “the expected outcome of running Vibe Rounds on an arbitrary complex case is that it produces better-organized, more complete clinical reasoning, not a publishable novel syndrome” (Section 5.2), and that this is itself a meaningful, organizational contribution rather than a disappointing one.
Run this only if you intend to write the case up formally.
Prompt: “#VibeRounds Take everything produced across Stages 1–7 and map it onto the CARE guideline’s case-description fields: patient information, timeline, diagnostic assessment, therapeutic interventions, follow-up and outcomes. For each field, either pull directly from our Stage outputs or flag explicitly what is still missing and needs to be gathered before this could be submitted. Separately, attach the Stage 2 search-strategy note (Step 9.2a) as the search-reproducibility appendix. Do not invent any detail not already established in Stages 1–7 to fill a CARE field — leave it marked as a gap instead.”
Application Note: The explicit instruction not to invent detail to fill a field gap matters — an LLM under instruction to “produce a CARE-mapped summary” will sometimes smooth over a genuine gap rather than flag it, which is the opposite of what a methods-honest write-up needs.
If you are using this module alongside Framework D, this step is the same Critical Awareness Standing Prompt defined there — included here in full so this file is usable on its own without Framework D.
Prompt: “#VibeRounds Apply the Critical Awareness lens to this protocol run: (1) What cognitive bias most likely affected the reasoning across these seven stages — mine or the AI’s? (2) What is the most important clinical risk of acting on today’s Stage 7 outputs without further verification? (3) What would the strongest critic of this case research say about its methodology — particularly about the Stage 2 comparator search and any Stage 6 novelty or literature-gap claim? (4) What single uncertainty should I hold clearly in mind before treating anything from this run as more than a provisional hypothesis?”
Application Note: Run this once at the close of every protocol pass, chunked or single-session. It is deliberately the same prompt structure as standing Critical Awareness prompt so that a case run through Module 9 and a teaching session run through Modules 1–7 produce comparably self-critical closures.
| Step | Stage | Description |
|---|---|---|
| 9.0 | — | Single-session orchestration (optional alternative to 9.1–9.7) |
| 9.1 | 1 | Index case structuring (+ 9.1a completeness check) |
| 9.2 | 2 | Comparator identification (+ 9.2a search-strategy export) |
| 9.3 | 3 | Aggregation and cross-comparison |
| 9.4 | 4 | Two-tier analysis (coded vs. narrative) |
| 9.5 | 5 | Open-ended analytic ideation |
| 9.6 | 6 | Selection of an analytic menu (+ 9.6a run-an-item, repeatable) |
| 9.7 | 7 | Convergence to fixed outputs (severity, priority, timeline, correlation table) |
| 9.8 | — | Manuscript / CARE-field export (optional) |
| 9.9 | — | Critical Awareness debrief |
This module, like Modules 5–7, is designed for use against real case material. Before Step 9.1: confirm the case narrative you are about to paste is de-identified, consistent with the de-identification requirement in the Framework Safety & Compliance Note and the Data Security Notes in Modules 2–3. At every stage, the AI’s role is Socratic and structuring, not diagnostic — diagnostic interpretation, clinical judgment, and decisions about which AI-suggested lead to pursue remain with the clinician throughout (preprint Section 2, Section 5.4, Section 5.6). Treat all Stage 2/3/6 literature-gap or novelty claims as provisional and search-coverage-limited until independently verified (Section 5.1) — this applies whether or not the case is ultimately written up for publication.
If this file as Module 9, the following updates to front matter and back matter would keep the document internally consistent:
Objective: Turn a medical article into deeply digested clinical knowledge — by having the AI do the work of unpacking, layering, translating, and connecting the evidence, so the learner leaves with understanding that sticks rather than facts that evaporate.
Indication: Self-directed literature reading, pre-journal-club preparation, post-case deepening, or any moment when you have read (or are about to read) a paper and want to absorb it rather than merely skim it.
Status: 🟡 In Progress — authored June 2026; not yet validated against a live cohort.
Design orientation: This module is not a quiz. The AI’s job here is not to test the learner — it is to do the intellectual processing that most readers skip: translating statistics into clinical meaning, mapping findings onto real patients, surfacing what the paper does not say, and layering understanding so the reader ends the session knowing more than the abstract told them.
[!IMPORTANT] All AI-generated appraisal outputs — including effect-size interpretations, bias ratings, applicability assessments, and registry-linkage conclusions — require independent clinical verification before being acted upon. This module is a learning tool, not a clinical decision aid.
The failure mode of medical reading is not lack of effort — it is passive absorption. A reader finishes an article, closes the tab, and retains the conclusion without internalising the why, the for whom, the caveats, or the so what for my patient. A week later, what remains is a fragile half-memory of a p-value.
Module 10 addresses this by inverting the reader’s role. Instead of the learner summarising for the AI, the AI digests for the learner — in progressively deeper layers — so that understanding accumulates rather than evaporates.
Five steps in the Execution phase are patient-anchored: at those steps, the learner pastes a brief de-identified patient vignette and the AI presses the article’s evidence directly against that patient. These five patients can be the same patient (tracking one case through every angle of the paper) or five different patients (stress-testing the paper’s generalisability across patient types). Either approach is valid; the choice depends on the learner’s goal for that session.
AI mode profile: This module is predominantly Traditional and Devil’s Advocate — the AI produces, the learner reads and absorbs. No prompt requires a learner answer before the AI proceeds.
Phase 1 · Initiation → Phase 2 · Execution — Layered Digestion → Phase 3 · Closure / Knowledge Crystallisation
AI Mode: Traditional
Prompt:
#VibeRounds You are a clinical knowledge companion. I am going to give you
a medical journal article. Your first job is to give me a plain-language
explanation of what this article is about — written as if you are
explaining it to a clinician who has not yet read it. Cover: (1) the
clinical problem being studied and why it matters, (2) what the
researchers did and why they chose that approach, (3) what they found —
in one or two sentences, no statistical notation, (4) why a clinician
should care about this finding. Make it readable and human. [paste
article, abstract, or publicly accessible URL]
[!NOTE] Application Note: The plain-language first read is the entry gate. Statistics, methods, and caveats come in later layers. Starting with human-readable framing is deliberate — it gives the reader a mental scaffold to hang everything else onto before the complexity arrives.
AI Mode: Traditional
Prompt:
#VibeRounds Here is a patient I am managing or thinking about: [paste
a brief de-identified patient vignette — diagnosis, key history,
current management, and the specific clinical question you are facing].
Read this patient and the article together. Tell me in two or three
sentences: what specific clinical question does this article raise for
this patient? Is there anything in the paper that directly bears on a
decision already made in this case — and if so, does the evidence
support, challenge, or remain silent on that decision?
[!NOTE] Application Note: The case anchor is not a quiz — the AI answers its own question. The learner reads the linkage and thinks: yes, that is my patient. Use a de-identified vignette only — remove name, date of birth, and any identifying details before pasting. The same patient can be reused across all five anchored steps, or a different patient used at each step to broaden the generalisability test.
The digestion phase works in six layers. Run them in sequence for maximum absorption. Each layer builds on the last.
AI Mode: Traditional
Prompt:
#VibeRounds Explain the study design of this article in plain clinical
language. Tell me: (1) what type of study this is and what that means for
the strength of the evidence, (2) who the participants were and whether
they represent the kinds of patients I see, (3) what exactly was done —
intervention and comparator — and (4) what was measured and why that
outcome was chosen. Where the methods have a meaningful limitation, name
it here as a practical note on what it means for how much I should trust
the finding — not as a textbook caveat list.
AI Mode: Traditional · Devil’s Advocate
Prompt:
#VibeRounds Look at the study population in this article and compare it
to this patient: [paste de-identified patient vignette]. Tell me
directly: would this patient have been eligible for enrolment in this
study? If not, what does that exclusion mean for whether I can apply this
paper's finding to their care? Give me a one-paragraph honest assessment
— not hedged academic language, a direct clinical opinion.
[!NOTE] Application Note: External validity is the most consistently skipped step in clinical reading. Naming a real patient makes the question concrete rather than theoretical. This is the step most likely to reveal that a headline result does not apply to the patient you actually have.
AI Mode: Traditional
Prompt:
#VibeRounds Translate the main statistical result of this article into
clinical language I can actually use. Tell me: (1) what the effect
measure is — RR, OR, HR, MD, NNT, or whatever applies — and what it
means in words, (2) what the confidence interval tells me about
uncertainty expressed as a practical range, not a statistical definition,
(3) whether the effect size is clinically meaningful — separate from
whether it is statistically significant, and (4) if an NNT is reported
or calculable, what that number means for a patient sitting in front of
me. If the paper reports multiple outcomes, focus on the primary outcome
first, then flag if any secondary outcome tells a different story.
AI Mode: Devil’s Advocate
Prompt:
#VibeRounds Play devil's advocate on the statistics in this article. Tell
me: (1) what the numbers cannot tell me that I still need to know before
applying this finding, (2) whether the statistical significance was
driven by a large sample size rather than a meaningful effect, (3)
whether there are subgroup results that contradict or complicate the
headline finding, and (4) what would need to be true for this result to
be a false positive. Give me your honest read of where the statistical
story is weakest.
AI Mode: Devil’s Advocate
Prompt:
#VibeRounds Map the most likely sources of bias in this study. For each
source you identify, give me: (1) what the bias is and how it likely
entered this specific study — not a textbook definition, (2) the
direction the bias probably pushed the result — towards or away from the
null, and (3) how much this should affect my confidence in the main
finding. Conclude with a single sentence: given all of this, what is
your overall bias-adjusted read of how reliable this result is?
AI Mode: Devil’s Advocate · Brainstorming
Prompt:
#VibeRounds This article answers one question. Tell me what questions it
deliberately or inadvertently leaves unanswered — and which of those
gaps matter most for clinical practice. For each gap, tell me whether
existing evidence elsewhere fills it or whether it is genuinely open.
Then tell me: for this patient [paste de-identified patient vignette],
which unanswered gap is most consequential for the clinical decision
I am facing?
[!NOTE] Application Note: This prompt produces the learner’s next reading agenda — turning a single article into a directed literature journey. The patient anchor at the end converts an abstract gap list into a clinically prioritised one.
AI Mode: Traditional
Prompt:
#VibeRounds Given what this article found, tell me concretely: does this
evidence change, support, or leave untouched the management of this
patient? [paste de-identified patient vignette]. Be direct. If the
evidence supports what is already being done — say so and say why. If it
challenges a decision already made — say so clearly and name the decision
point. If the paper is silent on this patient's situation — say that too,
and tell me why the gap exists.
AI Mode: Devil’s Advocate · Brainstorming
Prompt:
#VibeRounds Before I adopt this paper's finding into practice, tell me
what counter-evidence or alternative arguments I should be aware of.
Specifically: (1) are there other published studies that reach a
different conclusion on the same question, and if so, why might they
disagree? (2) what does current guideline consensus say — does this
paper align with, challenge, or extend it? (3) if a rigorous
methodologist were presenting this paper at journal club, what would
they say is the strongest reason not to act on this finding yet? Give
me the counter-argument, not just the caveat.
AI Mode: Traditional
Prompt:
#VibeRounds Apply Bloom's Revised Taxonomy to this article as a teaching
object. Tell me: (1) Remember — what are the two facts from this paper a
clinician must be able to recall? (2) Understand — what is the
conceptual insight this paper teaches beyond the result itself?
(3) Apply — in what clinical scenario would a clinician use this
finding, and how? (4) Analyse — what does this paper's design reveal
about the difficulty of studying this question well? (5) Evaluate — is
this paper a reliable basis for practice change, and why or why not?
(6) Create — what single sentence would you add to a local clinical
guideline based on this evidence, and for which patient population
exactly?
[!NOTE] Application Note: See Framework C. Unlike Module 1 where Bloom’s is a learner interrogation, here it is an AI-generated teaching analysis that the learner reads and internalises. The ‘Create’ level produces a draft guideline sentence — a concrete output the learner can test against their own clinical context.
AI Mode: Traditional
Prompt:
#VibeRounds Apply Fink's six dimensions of significant learning to this
article and tell me what genuinely matters from it across each dimension:
(1) Foundational Knowledge — what are the two things I must now know
that I may not have known before reading this paper? (2) Application —
what will a clinician who has absorbed this paper do differently at the
bedside? (3) Integration — how does this finding connect to or reframe
something I already know about this condition or system? (4) Human
Dimension — what does this paper mean for the experience of being a
patient with this condition; does it change the conversation a clinician
should have with them? (5) Caring — does this evidence activate any
value conflict or ethical tension — for example around resource
allocation, treatment trade-offs, or whose outcomes were measured?
(6) Learning How to Learn — what does reading this paper teach me about
how to read the next paper on this topic more efficiently?
[!NOTE] Application Note: See Framework B. Dimension 5 (Caring) is not optional — articles that appear value-neutral often embed ethical assumptions in their choice of outcome measure, comparator, or study population. Surfacing this is part of full digestion.
AI Mode: Traditional
Prompt:
#VibeRounds Based on this article, help me explain the evidence to this
patient: [paste de-identified patient vignette, including any relevant
context about the patient's education level or communication needs if
known]. Write a plain-language paragraph — the kind I could read aloud
in a clinic — covering: what the research found, what it means for this
patient specifically, and what I am recommending as a result. Do not
include statistical notation. Translate effect size into everyday impact
language. End with one sentence the patient can take home as their key
message.
[!NOTE] Application Note: This step closes the loop between evidence and care. It is placed last in the execution phase deliberately — translation to patient language should happen after full critical appraisal, not before. A clinician who communicates an uncritically absorbed finding to a patient has not finished reading the paper.
AI Mode: Devil’s Advocate
Prompt:
#VibeRounds Before I close this article session, apply the Vibe Rounds
Critical Awareness lens: (1) What publication bias, funding influence,
or framing choices might have shaped this paper's presentation of its
findings? (2) What assumptions about the patient population, healthcare
system, or resource context does this paper make that may not transfer
to my setting? (3) What is the single weakest link in this paper's
chain of inference — from question to method to result to conclusion?
(4) What is the most important clinical question this paper raises that
it cannot itself answer? Give me honest, targeted answers — not a
generic limitations section.
[!NOTE] Application Note: See Framework D. This is not a reason to dismiss the paper — it is the final calibration step before deciding how much weight to give the finding in practice. A learner who cannot answer question 4 has not yet finished digesting the paper.
AI Mode: Traditional
Prompt:
#VibeRounds Produce a Knowledge Crystallisation Summary for this article
session. Structure it exactly as follows:
The finding in one sentence: [plain language, no jargon]
Who it applies to: [specific patient profile from the study population]
Effect in clinical terms: [NNT / ARR / practical magnitude — not raw statistics]
Strongest caveat: [single most important limitation for practice adoption]
What it changes in my practice: [specific, patient-type-anchored]
What it does not change: [equally specific]
Biggest remaining gap: [one unanswered question the evidence leaves open]
Best next read: [one specific type of paper or guideline to consult before acting]
End with one sentence that names what made this article worth reading —
even if adoption is uncertain.
[!NOTE] Application Note: The crystallisation summary is the exportable output of this module — save it, share it at journal club, or link it to a registry case. The ‘what it does not change’ field is as important as ‘what it changes’ — precision about scope prevents over-application of evidence to patient types the study did not include.
AI Mode: Traditional
Prompt:
#VibeRounds Generate a 250-word structured abstract suitable for
presenting this article at a journal club or case conference. Include:
clinical question in PICO format, study design and population, key
result with effect size in clinical language, two most important
critical appraisal points, and one practice implication anchored to a
specific patient type. After the abstract, add two tags: (a) the Bloom's
level at which this article is best taught and why, and (b) the Oxford
CEBM evidence level for the main finding.
[!NOTE] Application Note: Bloom’s and CEBM tags make this abstract reusable for curriculum mapping, registry linkage (Module 6), and cross-case learning (Module 7). Export format mirrors Step 5.11.
AI Mode: Traditional
Prompt:
#VibeRounds Based on the complexity of this article and the depth of
digestion we reached in this session, recommend what I should read next
to deepen understanding on this topic: same level of complexity, one
step more methodologically rigorous, or a foundational paper I should
have read first. Justify the recommendation briefly. Then suggest one
specific reading skill to sharpen before the next session — for example:
reading a forest plot, interpreting heterogeneity in a meta-analysis, or
locating allocation concealment in a methods section.
| ID | Phase | Layer | Prompt purpose | AI mode |
|---|---|---|---|---|
| 10.0 | Initiation | — | Plain-language first read — entry scaffold | Traditional |
| 10.1 | Initiation | — | Case anchor — Patient 1 — clinical question linkage | Traditional |
| 10.2 | Execution | Layer 1 | Study design in plain English — methods translation | Traditional |
| 10.3 | Execution | Layer 1 | Study population vs Patient 2 — applicability check | Traditional · Devil’s Advocate |
| 10.4 | Execution | Layer 2 | Effect size in human language — statistics translation | Traditional |
| 10.5 | Execution | Layer 2 | What the statistics do not tell me | Devil’s Advocate |
| 10.6 | Execution | Layer 3 | Bias map — direction and magnitude per source | Devil’s Advocate |
| 10.7 | Execution | Layer 3 | Evidence gap map — Patient 3 — priority gap linkage | Devil’s Advocate · Brainstorming |
| 10.8 | Execution | Layer 4 | Management implication — Patient 4 — direct decision linkage | Traditional |
| 10.9 | Execution | Layer 4 | Counter-evidence radar — guideline alignment and objections | Devil’s Advocate · Brainstorming |
| 10.10 | Execution | Layer 5 | Bloom’s Taxonomy — article as knowledge object | Traditional |
| 10.11 | Execution | Layer 5 | Fink FLINK — six dimensions of significant learning | Traditional |
| 10.12 | Execution | Layer 6 | Patient communication distillation — Patient 5 | Traditional |
| 10.13 | Closure | — | Critical awareness debrief — Framework D | Devil’s Advocate |
| 10.14 | Closure | — | Knowledge crystallisation summary — exportable | Traditional |
| 10.15 | Closure | — | Exportable journal club abstract — CEBM + Bloom’s tags | Traditional |
| 10.16 | Closure | — | Difficulty ratchet — next reading session | Traditional |
Five steps in this module ask the learner to paste a patient vignette. These slots exist because abstract evidence only becomes clinical knowledge when it is pressed against a real person. There are two ways to use them:
Same patient, five angles — paste the same de-identified vignette at Steps 10.1, 10.3, 10.7, 10.8, and 10.12. This tracks one patient through every angle of the paper — clinical question, applicability, evidence gaps, management decision, and patient communication. Useful when a paper was read because of a specific patient.
Five different patients — paste a different patient at each step. This stress-tests the paper’s generalisability across patient types — useful when building a general understanding of an evidence base rather than answering a specific clinical question.
[!NOTE] Use de-identified vignettes only at all five steps. Remove name, date of birth, hospital number, and any other identifying detail before pasting.
| Mode | Steps |
|---|---|
| Traditional | 10.0, 10.1, 10.2, 10.4, 10.8, 10.10, 10.11, 10.12, 10.14, 10.15, 10.16 |
| Devil’s Advocate | 10.5, 10.6, 10.13 |
| Traditional + Devil’s Advocate | 10.3 |
| Devil’s Advocate + Brainstorming | 10.7, 10.9 |
No Socratic prompts in this module by design — this module digests knowledge into the learner; it does not test the learner’s recall of it. If Socratic-mode article engagement is the goal, apply Module 1 to an article-based clinical case.
Objective: Help medical students discover, anticipate, and respond to the real questions patients and families carry about an illness — so that students learn to see the clinical picture through the patient’s eyes, not just through the medical record.
Indication: Use when a student has a diagnosis, a short history, or a full clinical picture and wants to prepare for meaningful patient conversations, discharge counselling, or bedside communication practice. Can also be used to surface information gaps the patient has not yet articulated but almost certainly holds.
[!IMPORTANT] Clinical Disclaimer & Independent Verification Required All AI-generated outputs produced using this module — including patient concern summaries, lay explanations, red-flag lists, and counselling scripts — require independent clinical verification before being acted upon or shared with a patient. This module is an educational scaffold for students, not a substitute for clinical judgment, institutional communication protocols, or the clinician–patient relationship.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds I am a medical student using Module 11 — Patient Education Query Intelligence.
My goal is to understand the questions, fears, and information needs that a real patient
(or their family) is likely to carry about this case — and to learn how to address those
needs clearly, accurately, and humanely.
Here is the clinical input [choose what you have]:
- OPTION A — Diagnosis only: [e.g., Type 2 Diabetes Mellitus, newly diagnosed]
- OPTION B — Short history: [paste a 2–5 sentence summary]
- OPTION C — Full clinical picture: [paste history, examination findings, investigations,
medications, and management plan]
Please confirm you have received the clinical input and understand your role: you are
an AI educational assistant helping me learn about patient information needs — you are
NOT directly advising the patient. Confirm before we begin.
[!NOTE] Application Note: The module accepts any level of clinical detail. A diagnosis alone is sufficient for Step 11.1–11.3; the full clinical picture unlocks medication-specific, lifestyle-specific, and red-flag steps (11.4–11.7). Paste the richest input you have.
Validated Environment: Claude
Prompt:
#VibeRounds Based on the clinical input I provided, generate the 10–15 questions
a patient in this situation is most likely to be silently carrying — questions they
may not know how to ask their doctor, or may feel embarrassed to raise.
Organise them under these headings:
1. "What is happening to my body?" — disease understanding questions
2. "What will happen to me?" — prognosis and trajectory questions
3. "What do I need to do?" — self-management and lifestyle questions
4. "What can go wrong?" — worry and safety questions
5. "What does this mean for my life?" — impact on work, family, daily function
For each question, add one sentence explaining *why* patients commonly hold this
concern but often do not voice it.
[!NOTE] Application Note: This step is the entry point for empathic clinical communication training. The “why they don’t voice it” column builds student insight into health literacy gaps, shame, and power dynamics in the clinical encounter.
Validated Environment: Claude
Prompt:
#VibeRounds For this clinical picture, produce a plain-language explanation of:
(a) What the diagnosis is and what it means in the body — using an analogy a
patient with no medical background would find memorable
(b) What the key investigation findings mean in one sentence each
(c) What the current treatment is doing and why it matters
Use language a person with a Grade 8 reading level can understand. Avoid jargon;
where a medical term is unavoidable, define it in brackets immediately after.
Do NOT simplify so much that clinical accuracy is lost.
[!NOTE] Application Note: Evaluating the AI’s lay explanation is itself a learning task — students should compare it against their own knowledge and identify any inaccuracy, over-simplification, or missing nuance. This builds both communication skill and critical AI literacy.
Prompt:
#VibeRounds Now adopt the perspective of a close family member or primary caregiver
of this patient. Generate:
(a) The 5 most urgent questions this family member would want answered before leaving
the hospital or clinic
(b) The 2–3 fears they are most likely carrying that they may not express to the
medical team
(c) One practical concern about the home environment or caregiver capacity that the
medical team might not have asked about
Frame each item as a direct quote from the family member's point of view
(first-person voice).
[!NOTE] Application Note: This step trains students to extend their communication beyond the patient to the family system — a critical competency for complex chronic illness, paediatrics, geriatrics, and palliative care.
Use when medications are included in the clinical picture.
Prompt:
#VibeRounds For each medication in this patient's regimen, generate the 3 questions
a patient is most likely to ask about it:
Format as a table:
| Medication | Patient Question | Plain-Language Answer | Why This Question Matters |
Include at least one question per medication that relates to a common patient
misconception or adherence barrier (e.g., "Can I stop this once I feel better?",
"Will this make me dependent?", "Does this interact with my herbal supplement?").
[!NOTE] Application Note: Medication literacy is one of the highest-yield domains for preventing readmission and adverse events. Students often know pharmacology but have not practised translating it into the language a patient can act on.
Prompt:
#VibeRounds Based on this clinical picture, generate a patient-facing red-flag
awareness guide. Structure it as:
**"When to seek help immediately" (Emergency — go to hospital now):**
- [List 3–5 signs specific to this patient's condition]
**"When to call your doctor within 24 hours" (Urgent — do not wait):**
- [List 3–5 signs]
**"What to monitor at home" (Watch and record):**
- [List 3–5 parameters with simple instructions, e.g., BP reading threshold,
weight gain per day, urine output, symptom diary]
After the guide, add a teaching note for me (the student): explain WHY each
red flag is physiologically significant and what clinical event it is warning against.
[!NOTE] Application Note: The dual-output design (patient guide + student teaching note) is intentional. Students learn the clinical reasoning behind red flags while simultaneously producing something useful for the actual patient encounter.
Prompt:
#VibeRounds For this patient's condition, generate the most common lifestyle
questions patients ask — questions that are often inadequately addressed during
a time-pressured clinical encounter:
Cover all relevant domains from this list (include only those applicable to the case):
- Diet and nutrition (what to eat, what to avoid, portion size, meal timing)
- Physical activity (how much, what type, what to avoid)
- Sleep and rest
- Work and occupational capacity
- Sexual activity (if relevant)
- Travel (including air travel, altitude, or heat if applicable)
- Alcohol and recreational substances
- Driving (especially relevant for epilepsy, syncope, diabetes on insulin, sedative medications)
- Monitoring at home (devices, apps, symptom diaries)
For each question, provide a concise evidence-informed answer AND note if the answer
is individualized (i.e., needs to be tailored to this specific patient's situation
rather than given as a blanket recommendation).
[!NOTE] Application Note: Students frequently under-counsel on lifestyle because these topics feel outside the “medical” domain or because training has not given them confident, evidence-anchored answers. This step builds that fluency.
Prompt:
#VibeRounds Based on the full clinical picture, identify 2–3 important things
this patient needs to know — insights they would want if they understood their
condition as well as a doctor does — but which patients almost never think to ask.
These should be non-obvious, clinically important, and actionable. Examples of
the type of insight this step is looking for (do NOT repeat these — generate
new ones specific to this case):
- "The symptom you are most worried about is actually benign; the symptom you
have been ignoring is the one that needs watching."
- "This medication works better if taken at a specific time of day — here is why."
- "One common behaviour patients do when they feel better actually undermines
their recovery — here is what it is."
For each insight, explain:
(a) What the insight is, in plain language
(b) Why patients typically miss it
(c) How a student or clinician can proactively introduce it in a brief consultation
[!NOTE] Application Note: This is the highest-yield step in Module 11. It trains proactive communication — the skill of giving the patient the knowledge they did not know to ask for. It also surfaces clinically important “hidden” adherence risks, safety gaps, and self-management pitfalls.
Prompt:
#VibeRounds Before I close this session, ask me three questions to check whether
I am genuinely ready to counsel this patient:
Question 1: Ask me to explain the diagnosis in my own words, as if I am talking
to the patient — without using any medical terminology.
Question 2: Ask me to name the single most important thing this patient needs to
understand about their medications or self-management before discharge.
Question 3: Ask me to identify one question from the patient's likely inner list
(from Step 11.1) that I would find most difficult to answer honestly, and
explain why.
After I respond to all three, give me formative feedback: what I communicated
well, and one specific thing to improve in my next patient conversation.
[!NOTE] Application Note: This Socratic closure loop links Module 11 back to the Vibe Rounds pedagogic spine. It prevents the student from passively consuming AI output without internalising it. Aligns with Module 1 Step 1.4 (Mid-Session Reasoning Checkpoint) and Framework C (Bloom’s Apply level).
Prompt:
#VibeRounds Generate a brief patient-facing education summary for this case,
formatted as a take-home information card. Structure:
**Your Diagnosis:** [plain-language name + one sentence explanation]
**What This Means for Your Body:** [2–3 sentences, lay language]
**Your Medicines:** [table: Medicine name | What it does | When to take it | One thing to remember]
**Watch for These Warning Signs — Seek Help Immediately If:**
- [3–5 red flags]
**What You Can Do at Home:**
- [3–5 self-management actions]
**Your Next Steps:**
- [Follow-up appointment, monitoring, referrals]
**Questions to Ask Your Doctor Next Time:**
- [3 questions drawn from the patient's likely inner list]
Label this clearly as a STUDENT-GENERATED DRAFT requiring clinician review
before it is shared with any patient.
[!NOTE] Application Note: The output is a learning artefact, not a finished patient document. It should be reviewed by a supervising clinician before any real-world use. The act of producing it teaches students to synthesise clinical knowledge into communicable form — a core clinical competency.
Prompt:
#VibeRounds Before we close, apply the Vibe Rounds Critical Awareness lens
to what we have produced in this module:
(1) What health literacy assumptions did the AI make about this patient?
Are they justified given what we know about the case?
(2) What cultural, linguistic, or socioeconomic factors could alter how this
patient receives and acts on the information we generated?
(3) Where might the AI's patient education output be overconfident, incomplete,
or insufficiently individualised?
(4) What is the most important thing I, as the student, still do not know about
this patient's information needs that no amount of AI analysis can tell me?
Be honest and constructive. End with one sentence on what the best next step
is for me as a learner.
[!NOTE] Application Note: See Framework D — Critical Awareness Framework. This step guards against over-reliance on AI-generated patient education and ensures students return to the individual patient as the irreplaceable primary source.
| Input Type | Steps Unlocked | Notes |
|---|---|---|
| Diagnosis only | 11.0, 11.1, 11.2, 11.3, 11.5, 11.8, 11.10 | Core query and communication steps fully functional |
| Short history (2–5 sentences) | All above + 11.6, 11.7, 11.9 | Lifestyle and insight steps benefit from clinical context |
| Full clinical picture (H+E+Ix+Rx) | All steps | Medication step (11.4) and red-flag step (11.5) are richest at this input level |
| Module | Status | Description |
|---|---|---|
| Module 11 | 🟡 In Progress | Structured; individual steps untested against live student cohorts. Suitable for self-directed use. |
Objective: Systematically attack the working diagnosis and stress-test the entire differential list — surfacing anchoring errors, premature closure, and overlooked zebras before they reach the patient.
Indication: Any complex, atypical, or diagnostically uncertain presentation; cases where an initial working diagnosis has been formed but not yet confirmed; post-Socratic (Module 1) debrief when deeper adversarial pressure is warranted; high-stakes or rare-disease contexts where anchoring bias is most dangerous.
[!IMPORTANT] Clinical Disclaimer All AI-generated outputs produced using this module — including diagnostic challenges, ranked differentials, and bias flags — require independent clinical verification before being acted upon. Devil’s Advocate outputs are deliberately adversarial and do not represent a balanced clinical opinion. They are a stress-testing tool, not a diagnostic service.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
| Dimension | Socratic (Module 1) | Devil’s Advocate (Module 12) |
|---|---|---|
| Primary role of AI | Coach who withholds answers to make learner reason | Adversarial challenger who attacks the learner’s conclusions |
| Direction of pressure | Forward — “what do you think next?” | Backward — “why do you think that at all?” |
| Answer posture | Reveal only after learner commits | AI asserts the countercase and demands a rebuttal |
| Cognitive bias target | Errors of omission during reasoning | Errors of commission — premature closure, anchoring |
| Best moment to deploy | Before a diagnosis is formed | After a working diagnosis is formed |
| Learner affect to expect | Curiosity, guided uncertainty | Productive discomfort, intellectual pressure |
Prompt:
#VibeRounds You are entering Devil's Advocate Mode. Your role is not to
teach — it is to challenge. I have a working diagnosis for a clinical case
and I am going to share the case details with you. Your job is to
systematically argue against my working diagnosis, surface every
reasonable alternative on the differential, and force me to defend my
reasoning with evidence from the case — not intuition. You are allowed to
be blunt. You are not allowed to be cruel. You do not affirm my working
diagnosis until I have defended it against at least three serious
challenges you have raised. If I cannot rebut a challenge, you log it as
an UNRESOLVED VULNERABILITY and we continue. At the end of the session you
will give me a Diagnostic Confidence Score out of 10 based on how well I
defended my working diagnosis, and a list of all unresolved
vulnerabilities. Confirm you understand the rules and ask me to present the
case.
[!NOTE] Application Note: The three-challenge minimum before any affirmation is deliberate — it prevents the AI from validating a correct diagnosis too early and short-circuiting the adversarial stress-test. The UNRESOLVED VULNERABILITY log is the most clinically important output of this module; it becomes the audit trail for what still needs investigation.
Prompt:
#VibeRounds My working diagnosis is [STATE DIAGNOSIS]. Here is the case:
[PASTE CASE SUMMARY — demographics, presenting complaint, history, vitals,
examination findings, investigations to date]. Begin the adversarial
stress-test. Give me your three strongest arguments against this diagnosis
being correct. For each argument, cite the specific case feature that
undermines my diagnosis and name the cognitive bias that could explain why
I landed here anyway. Do not offer alternatives yet — only attack first.
[!NOTE] Application Note: Separating attack (Step 12.1) from alternative generation (Step 12.2) prevents the AI from softening the attack by immediately pivoting to a “but here’s what it could be instead.” The attack must land cleanly first. This structure is adapted from formal adversarial red-teaming methodology.
Prompt:
#VibeRounds Now generate the full adversarial differential. List every
diagnosis that fits this presentation at least as well as my working
diagnosis, starting with the most dangerous alternative (highest mortality
if missed) and ending with the most statistically likely alternative (base
rate). For each: (1) name one feature from the case that supports it,
(2) name one feature from the case that weighs against it, (3) name the
single investigation that would most efficiently confirm or exclude it.
Flag any diagnosis I should have considered that is absent from my
original differential.
[!NOTE] Application Note: The dual ranking — most dangerous first, then most likely — forces the learner to hold two separate risk calculi simultaneously, which mirrors real clinical triage logic. The “absent from your original differential” flag is where the highest-yield learning usually occurs.
Prompt:
#VibeRounds Now run a cognitive bias audit on my reasoning in this case.
For each of the following biases, tell me whether the case evidence
suggests I may have fallen into it, and if so, give one concrete example
from the case: (1) Anchoring bias — fixating on the first diagnosis
suggested; (2) Availability bias — favouring a diagnosis I have seen
recently; (3) Premature closure — stopping the search after the first
plausible fit; (4) Framing effect — being led by how the referral or
history was worded; (5) Confirmation bias — selectively weighting features
that support my working diagnosis; (6) Representativeness bias — matching
to a prototype and ignoring atypical features. For each bias present: rate
its likely influence as Low / Moderate / High and suggest one debiasing
action I could take right now.
[!NOTE] Application Note: This step operationalises Framework D (Vibe Rounds Critical Awareness Framework). Unlike the general debrief in Step 1.10, this bias audit is case-specific and adversarial — the AI is looking for evidence of bias in what the learner has already said and committed to, not raising abstract concerns. See Framework D.
Prompt:
#VibeRounds Now apply the Zebra Test. Set aside all common diagnoses.
Give me the top three rare or atypical diagnoses that this case could
represent, even if the prior probability is low. For each: (1) what
feature of this case — however subtle — makes you consider it, (2) what
is the single clinical or investigative finding that would confirm or
exclude it within 24 hours, (3) what is the consequence of missing it for
one more day? Flag if any of these zebras has a specific demographic,
geographic, or occupational risk factor relevant to this patient.
[!NOTE] Application Note: The 24-hour exclusion window is intentional — it grounds the zebra hunt in practical urgency rather than academic curiosity. This step is highest-yield in presentations with red-flag features (weight loss, night sweats, systemic symptoms) or in geographic or occupational contexts where exposure-based diagnoses are plausible.
Prompt:
#VibeRounds Challenge my evidence base. For my working diagnosis of
[RESTATE DIAGNOSIS], list every investigation I have ordered and rate each
one as: (A) Necessary and sufficient to support this diagnosis, (B)
Supportive but not specific, or (C) Irrelevant or actively misleading in
this context. Then tell me: (1) What is the single piece of evidence I do
not yet have that would most raise my diagnostic confidence? (2) What is
the single result I should obtain before committing to a management plan?
(3) Is there any result in the current dataset that I appear to have
underweighted or overweighted?
[!NOTE] Application Note: The A/B/C evidence sufficiency rating often reveals that management plans are being driven by non-specific supportive data rather than diagnostic-grade evidence — a pattern that Module 5’s data audit (Step 5.8) can flag at the case log level but that only this adversarial lens exposes at the reasoning level.
Prompt:
#VibeRounds I am now going to defend my working diagnosis against all the
challenges you have raised. [PASTE YOUR REBUTTAL HERE — or type it live
in your next message.] After I have made my case, you will: (1) Score
each of my rebuttals as Strong, Partial, or Insufficient; (2) For any
Insufficient rebuttal, restate the challenge and give me one specific piece
of evidence from the case or from clinical knowledge that I need to address
it properly; (3) Update the UNRESOLVED VULNERABILITY list — remove any
challenge I have successfully rebutted and keep or add any that remain
open.
[!NOTE] Application Note: This is the only step in Phase 2 where the learner leads and the AI responds to them, rather than vice versa. The rebuttal round inverts the adversarial dynamic briefly to give the learner agency — without this, the session risks becoming demoralising rather than productive. Strong rebuttals should be explicitly acknowledged; this is where the humanistic persona from Framework A is most important to activate.
Prompt:
#VibeRounds This presentation has atypical features. Apply the Atypical
Presentation Lens. For each atypical feature in this case: (1) name the
feature, (2) state which classic textbook presentation it deviates from
and how, (3) list the three most important explanations for why the
presentation might be atypical — including age-related variation, sex or
hormonal variation, immunosuppression, polypharmacy masking, or cultural
or linguistic framing of symptoms. Then ask me: if every typical feature
I was expecting had been present and this atypical one was the only thing
missing, would I still have reached the same diagnosis?
[!NOTE] Application Note: The closing question — a counterfactual reasoning probe — is deliberately Socratic in a Devil’s Advocate module. This is intentional: at Step 12.7, the adversarial pressure has been fully applied and a single Socratic pivot helps the learner reintegrate their reasoning rather than remain in a purely defensive posture. Modules are allowed to borrow a move from another mode at a specific step when the learner’s cognitive state warrants it.
Prompt:
#VibeRounds We have completed the adversarial stress-test. Now produce the
full Devil's Advocate Closure Report: (1) Diagnostic Confidence Score —
rate my overall defence of the working diagnosis out of 10, with a
one-sentence justification; (2) UNRESOLVED VULNERABILITY LIST — all
challenges I did not successfully rebut, with a one-line note on the
clinical implication of each; (3) Differential Ranking — your ranked list
of the top three alternative diagnoses that should remain live until the
unresolved vulnerabilities are addressed; (4) Minimum Investigation
Checklist — the specific tests or findings needed before I can close the
differential with confidence; (5) Cognitive Bias Summary — the one bias
that most significantly influenced my reasoning in this session.
[!NOTE] Application Note: The Diagnostic Confidence Score is not a grade on case knowledge — it is a calibration tool for investigative urgency. A score of 7/10 or above with no Critical-grade vulnerabilities supports proceeding. A score below 6/10 or any single unresolved vulnerability rated as high-stakes should trigger active re-evaluation before management escalates.
Prompt:
#VibeRounds We have now stress-tested this diagnosis adversarially. Help
me integrate this experience with my earlier Socratic reasoning on this
case. (1) What did the Devil's Advocate mode reveal that my Socratic
reasoning session missed or underweighted? (2) What did the Socratic
session build in me that helped me rebuttal-defend my diagnosis in this
session? (3) If I were to teach this case to a junior learner, which three
questions would I now add to a Socratic case discussion that I would not
have thought to ask before this adversarial session? Frame the answer as a
growth narrative, not a deficit summary.
[!NOTE] Application Note: This step is the formal integration bridge between Module 12 and Module 1. It is optional when Module 12 is run as a standalone, but strongly recommended when Module 12 follows a Module 1 session on the same case. The three new Socratic questions it generates can be fed directly into Step 1.1 (Socratic Question Generator) for a future session on a similar case.
Prompt:
#VibeRounds Based on everything we have discussed in this Devil's Advocate
session, extract all active red flags in this case — clinical features,
investigative gaps, or diagnostic vulnerabilities that should trigger
immediate escalation or senior review. For each red flag: (1) state the
feature, (2) state why it is a red flag in this case specifically (not
generically), (3) state the escalation action and timeframe. Then give me
a one-sentence clinical safety statement I could use in a handover or
documentation note to accurately represent the diagnostic uncertainty
that remains.
Prompt:
#VibeRounds Based on my performance in this Devil's Advocate session:
(1) Which phase of the adversarial process was I strongest in —
generating the rebuttal, recognising cognitive bias, or identifying
investigative gaps? (2) Which phase was I weakest in? (3) Recommend the
difficulty level for my next Devil's Advocate session: same complexity,
one level harder (add a second competing diagnosis with equal evidence
weight), or one level easier (reduce the number of atypical features).
(4) Give me one specific clinical reasoning skill to practise before the
next session — something I can do with a real case or a textbook in the
next 48 hours.
[!NOTE] Application Note: Mirrors the structure of Step 1.9 (Difficulty Ratchet for Next Session) intentionally — the parallel structure allows learners to track their trajectory across both Socratic and Devil’s Advocate modes in a consistent format. A learner who scores Strong across all rebuttals in Step 12.6 and receives a Diagnostic Confidence Score ≥ 8/10 in Step 12.8 should be directed to Module 12 Level Hard (two competing diagnoses) or Module 9 (N-of-1 Research Protocol) for the same case.
For experienced users who know the module — copy the step you need directly.
| Step | One-line purpose | Phase |
|---|---|---|
| 12.0 | Set the adversarial contract | Initiation |
| 12.1 | Attack the working diagnosis with three arguments | Execution |
| 12.2 | Generate full adversarial differential (danger-first, base-rate-second) | Execution |
| 12.3 | Cognitive bias audit — six biases, case-specific evidence | Execution |
| 12.4 | Zebra Test — rare/high-stakes alternatives | Execution |
| 12.5 | Evidence sufficiency challenge — rate every investigation | Execution |
| 12.6 | Rebuttal round — learner defends, AI scores and updates vulnerability log | Execution |
| 12.7 | Atypical presentation stress-test | Execution |
| 12.8 | Diagnostic Confidence Score + full Vulnerability Report | Closure |
| 12.9 | Integration debrief — Devil’s Advocate vs Socratic | Closure |
| 12.10 | Red flag extraction + escalation triggers + clinical safety statement | Closure |
| 12.11 | Difficulty ratchet + next session calibration | Closure |
| Module | Status | Description |
|---|---|---|
| Module 12 | 🔴 Primitive | Proposed structure — not yet validated in a live clinical or educational environment. Evidence base is expert opinion / design rationale only. |
| Step | AI Mode |
|---|---|
| 12.0 | Devil’s Advocate |
| 12.1 | Devil’s Advocate |
| 12.2 | Devil’s Advocate · Brainstorming |
| 12.3 | Devil’s Advocate |
| 12.4 | Devil’s Advocate · Brainstorming |
| 12.5 | Devil’s Advocate |
| 12.6 | Devil’s Advocate · Socratic |
| 12.7 | Devil’s Advocate · Socratic |
| 12.8 | Traditional |
| 12.9 | Devil’s Advocate · Socratic |
| 12.10 | Traditional |
| 12.11 | Traditional |
Tagging rationale: Steps 12.6, 12.7, and 12.9 carry dual Devil’s Advocate + Socratic tags. In 12.6 the AI is scoring a learner rebuttal (adversarial) but the learner is generating a reasoned defence (Socratic commitment required). In 12.7 the closing counterfactual question is a direct Socratic probe embedded in an adversarial frame. In 12.9 the integration debrief requires the learner to reason across two modes simultaneously. Steps 12.8, 12.10, and 12.11 are Traditional — they produce structured deliverables (report, safety statement, calibration note) with the AI in output mode.
The existing Devil’s Advocate prompts in the Vibe Rounds system (Steps 1.8, 1.10, 4.3, 6.6, 6.13, 7.6, 7.12, 8.2, 9.4, 9.9) are all embedded — they appear as single closure or debrief steps within modules whose primary mode is Socratic, Traditional, or Registry. None of them is dedicated to the full adversarial stress-test of a single working diagnosis from first challenge through to resolved closure.
Module 12 is the first module in the Vibe Rounds system where Devil’s Advocate Mode is the primary lifecycle mode, not a debrief addendum. Its distinct contributions:
Objective: Guide a medical student through a structured, patient-centred medication audit on a real or case-based patient on 5+ drugs — building the skills to identify drug-drug interactions, drug-disease conflicts, prescribing cascades, and practical adherence barriers through active reasoning, not passive receipt of answers.
Indication: Patient-centred learning sessions where the student is working with a real patient (under supervision) or a de-identified case and has access to a medication list; clerkship or bedside teaching preparation; PaJR WhatsApp group case reviews where the student is the active learner and a faculty teacher or senior student is facilitating; self-directed pharmacology revision on a complex multi-morbid case.
[!IMPORTANT] Clinical Disclaimer All AI-generated outputs from this module — including interaction flags, deprescribing observations, and dose-range alerts — are learning scaffolds, not clinical decisions. Nothing in this module authorises a student to change, stop, or recommend changing any patient’s medication. All findings must be discussed with a licensed supervising clinician before any action is taken. Drug interaction databases are incomplete; clinical context always overrides algorithmic output.
[!IMPORTANT] Data Security Note Medication records often appear on prescription photos that also capture full name, date of birth, and home address. Before using this module: type drug name, dose, frequency, and route as text rather than uploading a prescription image. Use de-identified data only in any shared LLM environment.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a warm, knowledgeable clinical pharmacology mentor
helping a medical student learn medication reconciliation through a real
patient case. Your role is not to audit the prescription for the clinical
team — it is to teach the student to think like someone who does. At every
step, ask the student to reason first before you reveal the finding. When
the student identifies a concern correctly, name it explicitly: 'That is
exactly the kind of flag a clinician would raise.' When they miss one, do
not simply list it — ask a guiding question that helps them find it
themselves. Begin by telling the student one thing that makes polypharmacy
clinically interesting before we look at any medications. Then ask them
to share the patient's medication list and active diagnoses. Confirm you
understand your role before we begin.
[!NOTE] Application Note: The opening instruction — tell the student one interesting thing about polypharmacy before looking at any medications — is a direct carry-over of the curiosity hook in Step 1.0 (Module 1). It activates engagement before the cognitive load of the medication table lands. The ‘ask the student to reason first’ contract is the core distinction between this module and a standard drug interaction checker: the AI is a teacher using a real case as material, not a lookup tool. Run once at the start of every session.
Prompt:
#VibeRounds Help me build the medication table for this patient. For each
drug I enter, ask me to fill in: (1) the generic name, (2) dose, (3)
frequency, (4) route, (5) what condition it is treating — in my own words,
(6) how long the patient has been on it, if I know. If I do not know a
field, prompt me with: 'That is worth finding out — what would be your
source for that information on a real ward?' Mark any field I cannot fill
as [NOT RECORDED] rather than skipping it, so we can see the gaps clearly.
When the table is complete, ask me: 'Looking at the total number of
medications, what term would you use to describe this prescribing pattern
— and at what threshold does that term apply?'
[!NOTE] Application Note: Making the student fill in the indication column in their own words is the highest-yield learning moment in this step — students who cannot state what a drug is for in plain language do not yet understand the patient’s condition well enough to spot a drug-disease conflict. The closing question (polypharmacy vs hyper-polypharmacy threshold) is a Bloom’s Remember-level checkpoint: the threshold is ≥5 drugs for polypharmacy and ≥10 for hyper-polypharmacy, but the student should produce that answer, not receive it unprompted.
Prompt:
#VibeRounds Now I want to find the drug-drug interactions in this list
myself before you tell me. For each pair of medications I name, tell me
whether I am right, partially right, or have missed the mechanism — then
ask me the next question rather than listing all the remaining interactions.
Start by asking me: 'Which two drugs on this list would you be most worried
about being given together — and what is your reasoning?' After I attempt
each pair, give me a hint if I am stuck rather than the answer. Once I have
worked through my own list, show me any interactions I missed, ranked by
severity: MAJOR first, then MODERATE, then MINOR. For each missed
interaction, explain the mechanism in one sentence and ask me: 'Now that
you know the mechanism, what would you watch for clinically in this
patient?'
[!NOTE] Application Note: The student-led structure here — attempting interactions before the AI lists them — is the most important design decision in this step. Students who receive a ranked interaction list first rarely retain the reasoning; students who commit to a pair and defend it before receiving feedback retain both the interaction and the mechanism. The ‘what would you watch for clinically’ follow-up converts pharmacological knowledge into clinical observation skill, which is the level of understanding a bedside student actually needs.
Prompt:
#VibeRounds The patient has the following active diagnoses: [LIST
DIAGNOSES]. I want to check whether any of the medications conflict with
these conditions. Walk me through this Socratically: name one diagnosis
at a time and ask me whether any medication on the list is problematic for
it — before telling me. For each one I identify correctly, tell me whether
it is an absolute contraindication, a relative contraindication, or a
caution. For any I miss, ask me a guided question: 'Think about what this
condition does to [organ system] — does that change how you would want any
of these drugs to behave?' End this step by asking me: 'Which drug-disease
concern on this list would you escalate first, and why?'
[!NOTE] Application Note: The organ-system guiding question is the key scaffold here — it gives the student a reasoning route without giving the answer. The ‘escalate first’ question at the end is a clinical prioritisation probe that tests a skill distinct from recognition: not just ‘can you find the conflict’ but ‘do you understand why it matters urgently.’ This connects to the renal/hepatic dose-adjustment concerns that are among the most common prescribing errors in ward medicine — metformin in renal impairment, digoxin toxicity in the elderly, DOAC dosing in CKD — and which a student on the ward is in a position to notice and flag.
Prompt:
#VibeRounds Some drug classes carry a disproportionate share of medication
harm. For each of the following classes, first ask me whether this patient
is on a drug in that class — and if so, what specific concern I would have.
Then fill what I miss. The five classes: (1) Anticoagulants — warfarin,
DOACs, heparin; (2) Hypoglycaemic agents — insulin, sulphonylureas,
SGLT2 inhibitors; (3) Antihypertensives — especially ACE inhibitors, ARBs,
diuretics, and the combination with NSAIDs; (4) Opioids — including
co-prescription of other respiratory depressants; (5) Psychotropics —
antipsychotics, benzodiazepines, antidepressants, and falls risk. For each
class that is present, end with: 'What is the one bedside observation that
would tell you this drug is causing a problem right now?' For any class not
present, say 'Not on this list — no check needed' and move on.
[!NOTE] Application Note: The ‘one bedside observation’ question anchors the class-level pharmacology to something the student can actually do at the bedside — observe. It bridges pharmacological knowledge and clinical behaviour, which is the gap that most pharmacology curricula leave open. The ‘Not on this list — no check needed’ instruction is equally important: a student who generates generic answers about drug classes not prescribed to this patient is practising pattern-matching, not patient-centred thinking. The triple whammy (ACEi/ARB + diuretic + NSAID → AKI) is the specific three-way combination worth drawing out explicitly if all three are present, since it is one of the most common causes of ward AKI and is invisible to standard pairwise interaction checkers.
Prompt:
#VibeRounds Pharmacological interactions are one layer of medication risk.
Now I want to think about practical risks — the ones the patient lives
with every day. Ask me the following questions one at a time, wait for my
answer, then add what I missed: (1) How many separate doses does this
patient take per day in total? What problems might that cause? (2) Are
any of these medications supposed to be taken at a specific time relative
to food, other tablets, or each other — and is that documented? (3) Are
there any foods or common drinks that interact with any of these drugs?
(4) Is there anything about this patient's circumstances — age, living
alone, swallowing difficulty, literacy, supply access — that might make
any of these medications harder to take correctly? (5) If this patient
misses one dose of each drug on the list, which missed dose carries the
most clinical risk, and which one can be safely doubled the next day?
[!NOTE] Application Note: This step is the patient-centred core of the module — it asks the student to think about the person taking the medication, not just the pharmacology. In PaJR settings, the student may have access to the advocate’s observations about the patient’s daily routine, which makes this step unusually rich: the AI can work from actual reported behaviour rather than hypothetical concerns. Question 5 (which missed dose is most dangerous vs which can be doubled) is a Bloom’s Evaluate-level probe that requires integrating pharmacokinetics, therapeutic window, and clinical consequence simultaneously — it is harder than it looks and should be attempted genuinely before any answer is offered.
Prompt:
#VibeRounds I want to check for prescribing cascades — situations where a
drug was added to treat a side effect of another drug rather than a new
disease. This is one of the most common and underdiagnosed causes of
polypharmacy. For each medication on the list, ask me: 'Can you think of
any other drug on this list that could be causing a side effect that this
medication was prescribed to treat?' Work through them one at a time.
When I identify a possible cascade, ask me: 'How would you test whether
that is a cascade or a genuinely independent indication?' When I finish,
show me any cascades I missed and explain the most classic examples —
including why the original prescriber may not have recognised it as a
cascade at the time.
[!NOTE] Application Note: Classic cascades a student should be able to identify with scaffolding include: opioid-induced constipation treated with a laxative; NSAID-induced dyspepsia treated with a PPI; antihypertensive-induced ankle swelling treated with a diuretic; antipsychotic-induced Parkinsonism treated with an anticholinergic. The ‘how would you test whether it is a cascade’ question is deliberately hard — the correct answer (trial withdrawal of the suspected causal drug under supervision) requires the student to understand both the pharmacology and the clinical ethics of stopping a medication, which is a sophisticated and high-yield reasoning skill.
Prompt:
#VibeRounds We have now worked through the full medication audit. Before
we move to any summary, I want to reason across all the findings together.
Ask me: (1) Of everything we found, which single concern would you bring
to the supervising clinician first — and how would you frame it in one
sentence? (2) Which finding surprised you most, and why did you not spot
it before we examined it together? (3) If this patient is discharged
tomorrow with this exact medication list and no changes, what is the most
likely medication-related harm they will experience in the next 30 days?
(4) What is the one thing you want to read about tonight because of what
this case taught you? Take each question one at a time. Do not answer
the next until I have genuinely attempted the current one.
[!NOTE] Application Note: This step is the module’s highest-yield learning moment — it forces synthesis across all the individual audit steps rather than leaving each finding as an isolated fact. Question 1 tests clinical communication (framing a concern for a senior); question 2 tests metacognition (recognising one’s own blind spots); question 3 tests predictive clinical reasoning (30-day harm projection); question 4 activates self-directed learning. Together they span Bloom’s levels Apply through Create. In PaJR group settings, these four questions can be posted as group discussion prompts after the student has attempted them privately.
Prompt:
#VibeRounds Help me write my own learning summary from this session.
Prompt me to fill in the following — ask me each one and then help me
refine my answer rather than writing it for me: (1) The three most
important pharmacological principles this case illustrated, in my own
words; (2) The one finding I would have missed without this structured
process; (3) The one clinical observation I will make every time I see
a patient on polypharmacy going forward; (4) A one-sentence summary of
this patient's medication risk that I could present to my supervising
clinician. After I have completed all four, reflect back to me: 'Based
on what you have written, here is what you understood well — and here is
the one concept worth revisiting before your next session.'
[!NOTE] Application Note: Making the student author the summary (with AI as editor, not author) is the deliberate design choice here — passive receipt of an AI-generated summary produces lower retention than active reconstruction. The one-sentence clinician-facing summary at the end of point 4 is a communication skill as much as a pharmacology skill: it requires the student to compress clinical complexity into a safe and actionable format for a senior, which is a core ward competency. The AI’s closing reflection mirrors the structure of Step 1.7 (Module 1’s End-of-Case Teaching Summary): name what was understood well before naming what needs revisiting.
Prompt:
#VibeRounds Based on everything we have found in this audit, help me
write a plain-language medication brief that the patient's advocate —
a non-medical family member — could actually use. I will draft it first;
you tell me what is unclear, what I have left out, and whether the reading
level is appropriate. The brief should include: (1) a plain-language
medication table — what each drug is called, what it is for in one
sentence, and when to take it; (2) a list of three things the advocate
should watch for and report to the clinical team; (3) a short checklist
of what to bring to every doctor's appointment. After I draft it, ask me:
'Is there anything in the pharmacological findings that you chose not to
include — and why?' That choice is as important as what you included.
[!NOTE] Application Note: This step converts a pharmacology audit into a health communication exercise — a skill gap that is rarely addressed in medical school curricula despite being a daily ward task. The closing question (‘what did you choose not to include — and why?’) is the critical thinking probe: a student who copies interaction flags verbatim into an advocate-facing document has not understood the difference between clinical information and actionable safety guidance. The correct answer includes the recognition that an advocate who reads ‘MAJOR drug interaction’ without context may stop a medication unilaterally, which could cause harm — so the brief communicates observable safety signals (watch for bleeding, dizziness, low blood sugar) rather than pharmacological mechanisms.
For use in PaJR WhatsApp learning groups — a short student-authored post summarising the learning for the group.
Prompt:
#VibeRounds Help me write a short reflection post for our PaJR learning
group about what this medication audit taught me. The post should have
three parts: (1) one clinical pearl from the pharmacology — something
I did not know or had not connected before this session; (2) one question
I still have that I want to discuss with the group or my supervisor;
(3) one practical tip for any student auditing a polypharmacy patient for
the first time. Keep the whole post under 150 words. I will draft it; you
help me sharpen it so it is useful to other students in the group, not
just a record of what I did.
[!NOTE] Application Note: This step serves the PaJR group learning architecture — the student is not just a receiver of teaching but an active contributor to the group’s collective knowledge. The three-part structure (pearl / open question / tip for others) is calibrated to produce posts that are useful to other students at different stages: the pearl rewards those who read; the open question invites the faculty or senior students to respond; the tip for others multiplies the learning beyond the student who ran the session. In PaJR groups where the teacher posts findings, this step gives the student their own voice in the group thread.
Prompt:
#VibeRounds Before we close, run the session closure check with me.
Ask me to confirm the following one at a time: (1) Can I name the two
most clinically significant findings from this audit without looking at
my notes? (2) Do I know which finding needs to be escalated to my
supervising clinician, and have I planned how to raise it? (3) Is there
any field that remained [NOT RECORDED] that I now know how to find on
a real ward? (4) Have I completed my personal learning summary (Step
13.8)? (5) Is there anything from this session I want to flag for the
patient's advocate or family? For any item I cannot confirm, note it as
OPEN and ask me: 'What is your plan for closing this before your next
clinical contact with this patient?'
[!NOTE] Application Note: The student-facing closure checklist replaces the clinician-facing ‘reconciliation is complete’ gate from a professional audit module — because for a student, completion is a learning milestone, not a medicolegal standard. The ‘what is your plan for closing this’ follow-up question is deliberately forward-looking: it activates self-directed learning rather than ending the session on an unresolved gap. Point 2 — ‘have I planned how to raise it’ — is a communication rehearsal prompt, not just a knowledge check. A student who knows what to escalate but does not know how to raise it with a senior has only half the competency.
For experienced users — copy the step you need directly.
| Step | One-line purpose | Phase |
|---|---|---|
| 13.0 | Set learning contract — AI as teacher, student reasons first | Initiation |
| 13.1 | Build medication table collaboratively, gaps visible | Execution |
| 13.2 | Student-led drug-drug interaction hunt, AI fills gaps | Execution |
| 13.3 | Drug-disease conflict — Socratic comorbidity layer | Execution |
| 13.4 | Five high-risk drug class spotlight, bedside anchored | Execution |
| 13.5 | Practical safety — what the patient actually experiences | Execution |
| 13.6 | Prescribing cascade detection — the hidden chain | Execution |
| 13.7 | Integration round — reasoning across all findings | Execution |
| 13.8 | Student-authored personal learning summary | Closure |
| 13.9 | Plain-language advocate brief — student drafts, AI edits | Closure |
| 13.10 | PaJR group reflection post — student voice in the group | Closure |
| 13.11 | Session closure self-assessment checklist | Closure |
| Module | Status | Description |
|---|---|---|
| Module 13 | 🔴 Primitive | Proposed structure — not yet validated in a live clinical or educational environment. Pharmacology framework logic is derived from published STOPP/START criteria and standard clinical pharmacology curricula but has not been formally tested against those tools or assessed for learning outcomes. |
| Step | AI Mode |
|---|---|
| 13.0 | Socratic |
| 13.1 | Socratic |
| 13.2 | Socratic |
| 13.3 | Socratic |
| 13.4 | Socratic · Devil’s Advocate |
| 13.5 | Socratic |
| 13.6 | Socratic · Devil’s Advocate |
| 13.7 | Socratic |
| 13.8 | Socratic · Traditional |
| 13.9 | Socratic · Traditional |
| 13.10 | Traditional |
| 13.11 | Traditional |
Tagging rationale: The original module (professional-facing) was predominantly Traditional. This student-centred revision is predominantly Socratic — the AI withholds findings until the student attempts them in every execution step. Steps 13.4 and 13.6 carry Devil’s Advocate dual tags: in 13.4 the AI is actively probing the student’s assumptions about drug class risks (adversarial lens on their reasoning); in 13.6 it is challenging them to detect a hidden prescribing pattern that their initial reading missed. Steps 13.8 and 13.9 are Socratic + Traditional: the student authors content and the AI edits/scaffolds, producing a structured deliverable through an active process. Steps 13.10 and 13.11 are Traditional — they produce a group post and a checklist with the AI in a supporting rather than teaching role.
The original design of Module 13 assumed three concurrent user tiers — patient advocate, intern, and ward physician — each entering through a different initiation prompt and receiving different outputs. That architecture was built for a clinical workflow.
This revision has a single user: a medical student doing patient-centred learning. Every structural consequence of that change:
| Dimension | Original (professional) | This version (student) |
|---|---|---|
| Primary user | Ward physician / intern / advocate | Medical student under supervision |
| AI role | Clinical audit tool + documentation aid | Clinical pharmacology teacher |
| Step 13.0 | Three-tier entry (A / B / C) | Single student learning contract |
| Execution steps | AI runs the audit, produces findings | Student reasons first, AI fills gaps |
| Step 13.7 | Optional Socratic layer, bolted on last | Central integration step, core to the module |
| Closure outputs | Clinician safety report + medicolegal checklist | Student learning summary + advocate communication exercise |
| Step 13.10 | PaJR handoff note posted by physician | Student reflection post in their own voice |
| Closure checklist | ‘Reconciliation is not complete until…’ (medicolegal) | ‘What is your plan for closing this?’ (self-directed learning) |
| Dominant AI mode | Traditional (documentation) | Socratic (reasoning before reveal) |
The pharmacological content — interaction screening, drug-disease conflicts, high-risk class spotlighting, prescribing cascade detection — is identical in both versions. What changed is who does the thinking.
Objective: Train and support clinical reasoning under genuine investigation, imaging, drug, and referral constraints — reasoning that starts from “what is actually available here” rather than “what is ideal,” and that makes the resource ceiling itself part of the differential-diagnosis and management-planning process.
Indication: Any setting where the standard textbook or guideline work-up assumes investigations, specialists, or drugs that are not reliably available — district hospitals, rural primary health centres, low-resource tertiary centres, disaster/outreach camps, or any Nepal/India-context bedside or teaching session where the learner must commit to a plan before knowing if the “ideal” test will come back, or commit to a plan knowing it never will.
[!IMPORTANT] This module does not encode local drug formularies, essential-medicine lists, or facility-level capability standards — these vary by country, state, and even by week (stock-outs). Always have the learner state the actual local resource ceiling at Step 14.0 rather than letting the AI assume one. Treat every AI-generated triage or escalation suggestion as requiring independent clinical verification per the repository Disclosure Statement.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a clinical reasoning partner for resource-constrained
settings. Before we discuss any case, I will tell you exactly what is and
is not available to me right now: which labs, which imaging, which drugs,
which specialists or referral pathway, and the realistic time-to-result or
time-to-referral for each. Do not assume I have access to anything I have
not listed, and do not silently default to a high-resource-setting
work-up. If I later ask for a test or drug I have not listed as available,
remind me it is unavailable here and ask whether I want a plan that
substitutes for it, defers it, or triggers a referral instead. Confirm you
understand before I describe my setting.
[!NOTE] Application Note: This step exists because most LLMs default to tertiary-hospital-availability reasoning unless explicitly constrained. Re-run this step at the start of every new session — the AI does not retain the resource ceiling across sessions, and resource availability itself can change week to week (e.g. stock-outs, a generator down, a referral road washed out in monsoon).
Prompt:
#VibeRounds Help me build a structured resource inventory for [facility /
setting name] before we use it in any case. Ask me, one category at a
time: (1) point-of-care tests available on-site, (2) labs available on-site
vs sent-out (and turnaround time for sent-out), (3) imaging available
on-site vs referral-only, (4) drugs reliably in stock vs frequently
out-of-stock, (5) realistic referral distance/time and transport
constraints, (6) staffing — who is physically present at night and on
weekends. Summarise it back to me as a single reference table I can reuse
across cases this week.
[!NOTE] Application Note: Run once per facility or rotation, then reuse the resulting table as context at the start of each subsequent case (paste it in at Step 14.0). This avoids re-deriving the same inventory every session.
Prompt:
#VibeRounds Given the resource ceiling I described, walk me through this
case's differential diagnosis in two layers: (1) the differential a
fully-resourced tertiary centre would pursue, briefly, for orientation
only; (2) the differential as I should actually prioritise it here, given
what I can and cannot confirm. For layer 2, rank each diagnosis by how much
it would change my immediate management even without confirmatory testing,
not by how likely it is in the abstract. Ask me one question at a time
before revealing your own ranking.
[!NOTE] Application Note: The “what would change management” reframe is the central move of resource-constrained reasoning — it keeps the differential clinically actionable rather than academically complete. Aligns with the Socratic forced-commitment pattern from Module 1.
Prompt:
#VibeRounds For this case, I lack [specific test/imaging/drug, e.g.
troponin / CT head / a specific antibiotic]. Ask me first what I think the
best available clinical or low-cost substitute or workaround is, and what
its known sensitivity/specificity or limitation is compared to the gold
standard. Only after I answer, add anything I missed. Then ask me: at what
point does the absence of this test become dangerous enough that referral
is mandatory regardless of cost or distance?
[!NOTE] Application Note: Forces explicit articulation of the safety threshold — the point past which “make do” reasoning must yield to referral, which is the single most safety-critical decision point in resource-constrained practice.
Prompt:
#VibeRounds Walk me through how I should manage this case empirically,
assuming the confirmatory test I want will either never be available or
will not return in a clinically useful timeframe. Ask me to commit to a
working diagnosis and a management plan first. Then probe: what is my exit
criterion — the specific clinical sign that tells me this empirical plan
has failed and I need to escalate, refer, or reconsider the diagnosis
entirely?
[!NOTE] Application Note: Mirrors the answer-withholding discipline of Module 1, Step 1.0 — the learner commits before the AI weighs in. The “exit criterion” question is deliberately non-optional; empirical treatment without a defined failure threshold is a recognised global-health patient-safety gap.
Prompt:
#VibeRounds Given the referral distance and transport constraints I
described, help me reason through whether this patient should be referred
now, observed and reassessed in [interval], or managed entirely at this
level. Ask me to weigh: the risk of deterioration in transit, the risk of
delay if I wait, what I can realistically monitor here versus what I
cannot, and whether the family can practically undertake the referral
(cost, distance, caregiver availability). Do not assume referral is always
the safer option — ask me to justify whichever way I lean.
[!NOTE] Application Note: Deliberately resists the reflex that “refer when in doubt” is always correct — in real low-resource settings, a failed or delayed referral can be more dangerous than competent on-site management, and transport itself carries risk. This step should not be used to talk a learner out of a referral that is genuinely indicated; it is a structured-thinking aid, not a override.
Prompt:
#VibeRounds Pause. Give me a formative checkpoint on my reasoning so far:
(1) appropriate use of available resources — score out of 10 with one
encouraging note and one improvement note; (2) recognition of when the
ceiling was reached and escalation was needed — score out of 10; (3)
safety-netting and exit-criterion clarity — score out of 10. End with one
sentence on what you think is my strongest resource-reasoning instinct in
this session.
[!NOTE] Application Note: Structural analogue of Module 1, Step 1.4, re-scoped to resource-reasoning competence rather than pure diagnostic logic.
Prompt:
#VibeRounds We have reached the end of this case. Produce a closing summary
covering: (1) two moments where I reasoned well specifically *because* of
the resource constraint, not despite it; (2) the single highest-yield
low-cost clinical sign, history question, or bedside test I should
prioritise next time a similar case presents; (3) one thing about this
setting's resource ceiling I should flag to a supervisor or for system-level
improvement, separate from this individual case. Open with a sentence
acknowledging that resource-constrained reasoning is its own skill, not a
lesser version of textbook reasoning.
Prompt:
#VibeRounds For this case, identify any low-cost clinical sign, history
detail, or point-of-care finding I had available but underused or missed —
not a sign that required unavailable technology. Name the specific clue,
explain why it is easy to overlook under resource pressure, and frame it as
a growth observation rather than a failure.
[!NOTE] Application Note: Intentionally scoped to available clues only — the goal is sharpening use of what is actually accessible, not relitigating the absence of an unavailable test (already covered in Step 14.3/14.4).
Prompt:
#VibeRounds Before we close, apply a system-level critical awareness lens:
(1) Did today's resource ceiling lead me toward any reasoning shortcut that
would be unsafe if I later worked somewhere better-resourced and forgot to
adjust? (2) What is the risk of over-relying on empirical/pattern-based
reasoning in this setting specifically — where might it fail silently? (3)
What would a critic of resource-constrained, AI-assisted reasoning say
about this whole approach? (4) What is the most important unresolved
uncertainty in this case that no amount of local resourcefulness can
substitute for? Be honest but constructive.
[!NOTE] Application Note: Structural analogue of Module 1, Step 1.10 and Framework D, re-scoped to the specific failure modes of resource-constrained practice — most notably the risk of reasoning patterns becoming habitual workarounds that go unexamined once resources improve, and the opposite risk of normalising preventable diagnostic blind spots as “just how it is done here.”
Objective: Train pattern-recognition-based diagnostic reasoning — the compiled, holistic “illness scripts” experienced clinicians actually use — as a distinct skill from the stepwise differential-building drilled in Module 1.
Indication: Once a learner can reliably build a differential step by step (Module 1, Bloom’s Levels 1–4) and needs to start compressing that process into the fast, holistic recognition that characterises expert clinical reasoning. Best used on conditions the learner has already met at least once through slow reasoning — script-building works on partially familiar territory, not brand-new material.
[!NOTE] Theoretical Basis: Script Theory (Schmidt & Rikers, 2007; Custers, 2015) models expert diagnostic reasoning as retrieval of compiled “illness scripts” rather than first-principles differential construction. Each script bundles three components: enabling conditions (the context that makes a diagnosis plausible — epidemiology, risk factors, setting), fault (the underlying pathophysiological process), and consequences (the clinical findings that follow). Novices reason forward from findings through pathophysiology to diagnosis, one step at a time. Experts reason by matching a presentation against pre-compiled scripts almost instantly, only falling back to stepwise reasoning when no script fits well — the hallmark of encountering a genuinely novel or atypical case. This module trains the compression from stepwise to script-based reasoning deliberately, rather than leaving it to accumulate as an unexamined by-product of repetition.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a clinical reasoning partner helping me build illness
scripts rather than step-by-step differentials. An illness script has
three parts: enabling conditions (the context that makes this diagnosis
plausible — who gets it, when, where), fault (the underlying process), and
consequences (the findings that follow from the fault). When I describe a
condition or case, do not walk me through it stepwise. Instead, present a
case stripped to a small number of enabling conditions and ask me to name
the script — diagnosis, fault, and expected consequences — as a single
fast judgement, not a reasoned-out chain. Only slow down to stepwise
reasoning if I cannot match a script at all. Confirm you understand the
distinction before we begin.
[!NOTE] Application Note: Run once per script-building session. This step exists specifically to stop the AI from defaulting to Module 1’s stepwise Socratic pattern — script acquisition and stepwise differential-building are different skills and the prompt must say so explicitly, or most LLMs default to walking the learner through reasoning one clue at a time regardless of what is asked.
Prompt:
#VibeRounds Give me only the enabling conditions for a case — age,
context, risk factors, setting — with no symptoms or findings yet. Ask me
to name the script(s) this profile most readily activates: what diagnoses
come to mind immediately, before any clinical findings are given. Then
reveal the findings and ask whether they confirm, refine, or overturn the
script I activated.
[!NOTE] Application Note: This is the core script-acquisition drill. It deliberately withholds findings to test whether the learner’s pattern recognition fires from context alone — the way experienced clinicians often form a leading impression before the history is even finished. Compare to Module 1, Step 1.0, which withholds the answer; this step withholds the findings instead, testing a different and earlier stage of reasoning.
Prompt:
#VibeRounds Present me with a brief case. Ask me to state the full script
in three parts, in order: (1) enabling conditions — why this patient, why
now; (2) fault — the underlying process in one or two sentences; (3)
consequences — the findings I would expect, stated before I am told what
was actually found. Then compare my predicted consequences to the actual
case findings and tell me which part of my script — enabling conditions,
fault, or consequences — was weakest.
[!NOTE] Application Note: Identifying which component of the script is weak is more diagnostically useful for the learner than an overall right/wrong judgement — a learner who gets the fault right but predicts the wrong consequences has a different gap than one who never activated the right script at all.
Prompt:
#VibeRounds Give me two conditions whose scripts are commonly confused —
similar enabling conditions or overlapping consequences but a different
fault. Present a single case that could plausibly trigger either script.
Ask me to state which script I activate first, then ask: what is the one
enabling condition or consequence that would most efficiently discriminate
between the two? Only after I answer, tell me the actual discriminating
feature.
[!NOTE] Application Note: This is the highest-value drill in the module. Expert clinical reasoning is less about generating long differentials and more about efficiently discriminating between a small number of activated scripts — this step trains that discrimination directly rather than leaving it to emerge from volume of exposure.
Prompt:
#VibeRounds Present me with a case where the findings partially match a
common script but one or two key features don't fit. Ask me first: does
this match a script I know well, or does something not fit? If something
doesn't fit, ask me to name specifically what is inconsistent before
offering an alternative or modified script. Affirm explicitly if I
correctly recognise a mismatch rather than forcing the case into a
familiar script.
[!NOTE] Application Note: This step trains the single most safety-critical script-based skill: noticing when no script fits well enough, which is precisely the trigger point where expert reasoning is supposed to fall back to slower, stepwise analysis (see Module 1). A learner who forces every case into the nearest familiar script — rather than noticing the mismatch — is exhibiting the premature-closure risk named in Framework D, 2a.
Prompt:
#VibeRounds Run a rapid-fire script drill. Give me five short case vignettes
in sequence — enabling conditions and consequences only, no working shown.
For each, ask me to name the script (diagnosis + one-line fault) in a
single fast response, no deliberation. After all five, go back through and
ask me to justify any I am uncertain about, and tell me which ones were
genuine pattern recognition versus which were lucky or shallow guesses.
[!NOTE] Application Note: Deliberately time-pressured and shallow on the first pass — this mirrors how scripts are actually deployed clinically (fast, low-effort, parallel) before being checked. The follow-up justification pass distinguishes real script compilation from superficial pattern-matching, which matters because the two can look identical in a single fast answer.
Prompt:
#VibeRounds Based on this session, help me write out the script(s) we built
today in a compact, reusable form: enabling conditions / fault /
consequences, three lines each. Then tell me: which existing script in my
library is this new one most likely to be confused with in future, and what
is the one discriminating feature I should anchor on to avoid that
confusion later.
[!NOTE] Application Note: Writing the script down in this compressed form is the consolidation step that turns a single session’s pattern recognition into something retrievable later — without it, the discrimination work from Step 15.3 tends not to persist.
Prompt:
#VibeRounds Looking back at this session, identify the moments where I
correctly used fast script recognition versus the moments where I should
have — but didn't — fall back to slower stepwise reasoning because no
script fit well. For each moment of premature script-matching, name what
cue I overlooked that should have triggered the fallback.
[!NOTE] Application Note: This is the script-theory analogue of Module 1, Step 1.8 (Missed Diagnosis Debrief) — but the failure mode being audited is different. Module 1 audits what diagnosis was missed; this step audits which reasoning mode was wrongly chosen, which is the more fundamental skill at the expert end of training.
Prompt:
#VibeRounds Apply a critical awareness lens to script-based reasoning
specifically: (1) What is the risk of relying on compiled scripts when a
presentation is genuinely atypical or from an underrepresented population
not well reflected in the scripts I have built? (2) How would over-reliance
on fast script-matching differ from the automation-bias risk of over-relying
on AI output? (3) What is the single best safeguard against script-based
reasoning producing fast, confident, wrong answers? Be honest but
constructive.
[!NOTE] Application Note: Script-based reasoning trades the slow-but-thorough failure modes of novice reasoning for a new failure mode of its own — fast, confident pattern-matching that fails silently on atypical or underrepresented presentations. This step deliberately extends Framework D’s bias taxonomy to a risk that is specific to expert-level reasoning rather than novice reasoning, and should not be skipped simply because the learner has progressed past Module 1.
Objective: Train the two-way link between basic science and clinical diagnosis explicitly — diagnosis → mechanism, and mechanism → diagnosis list — rather than leaving the connection to form implicitly over years of clinical exposure.
Indication: Any point in training where a learner can state a diagnosis but not its mechanism, or can recite a mechanism but cannot generate the clinical presentations that follow from it. Particularly useful directly after Module 15, since an illness script’s “fault” component is the basic science — this module drills that single component in isolation, in both directions, before it gets folded back into full scripts.
[!NOTE] Theoretical Basis: This module operationalises vertical integration in medical education — the deliberate interweaving of basic and clinical sciences throughout training, rather than the traditional two-stage model of basic science first, clinical application later (Brauer & Ferguson, 2015, “The Integrated Curriculum in Medical Education,” AMEE Guide 96, Medical Teacher, 37(4), 312–322). The specific mechanism this module targets is encapsulated knowledge (Schmidt, Norman & Boshuizen, 1990) — the empirical finding that as clinicians gain experience, detailed basic-science knowledge becomes compiled (“encapsulated”) into higher-level clinical concepts and illness scripts, so that it is used fluently without being consciously recalled. The risk this module is designed to counter is premature encapsulation — a learner who has memorised the clinical pattern without ever having built the underlying mechanistic link, leaving a script that looks compiled but has no mechanistic foundation to fall back on when a case doesn’t fit the pattern (see Module 15, Step 15.4). Bidirectional drilling — diagnosis-to-mechanism and mechanism-to-diagnosis — is specifically what distinguishes genuine integration from one-directional memorisation; a learner who can only go one direction has not actually integrated the two domains.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a clinical reasoning partner helping me build explicit
two-way links between basic science and clinical diagnosis. There are two
directions we will work in, and you must tell me clearly which one we are
doing before each exchange: (A) Diagnosis → Mechanism: I name or you give
me a diagnosis, and I have to explain the underlying basic science —
physiology, pathology, pharmacology, microbiology, or anatomy as relevant
— in my own words before you confirm or correct. (B) Mechanism →
Diagnosis: you give me an isolated basic science mechanism with no
diagnosis named, and I have to generate the list of diagnoses or
presentations that follow from it, before you tell me what I missed. In
both directions, make me commit to an answer first — do not explain the
mechanism or list the diagnoses for me until I have made a genuine
attempt. Confirm you understand both directions before we begin.
[!NOTE] Application Note: Run once per integration session. The explicit “tell me which direction” instruction matters because most LLMs default to direction A (explaining mechanism after being given a diagnosis) since it is the more common pattern in medical text generally — direction B has to be forced or it will rarely occur on its own.
Prompt:
#VibeRounds Give me a diagnosis. Before you say anything else, ask me to
explain the core basic-science mechanism that produces it — in my own
words, at the level of physiology or pathology, not just naming the
process. Only after I attempt it, tell me what was accurate, what was
incomplete, and what I should add. Then ask one follow-up: which specific
clinical finding in this condition is most directly explained by the
mechanism I just described — the one finding a clinician could not predict
without knowing the mechanism?
[!NOTE] Application Note: The follow-up question is the integration test, not the mechanism explanation itself. A learner can often state a mechanism and a finding as two separate memorised facts without ever connecting them — asking them to name which finding the mechanism predicts forces the link to be made explicit rather than assumed.
Prompt:
#VibeRounds Give me an isolated basic-science mechanism — a physiological
derangement, a pathological process, a pharmacological action, or a
microbiological behaviour — with no diagnosis named. Ask me to generate as
many diagnoses or clinical presentations as I can that would follow from
this mechanism. Do not give me the diagnosis list yourself until I have
generated my own. Once I have, tell me which ones I got, which common ones
I missed, and whether I included anything that does not actually follow
from this mechanism — a false positive is as informative as a missed
diagnosis here.
[!NOTE] Application Note: This is the direction most curricula under-train (see vertical-integration literature — most “integration” in practice still runs basic-science-to-clinical, rarely the reverse). It is also the more clinically realistic direction in one specific sense: a clinician who finds an abnormal lab value or imaging finding mid-workup is often reasoning from mechanism toward diagnosis, not the other way round. Flagging false positives (mechanisms the learner over-generalised from) is as important as flagging missed diagnoses.
Prompt:
#VibeRounds Give me a basic-science mechanism that underlies two or more
different diagnoses — for example a shared pathophysiological pathway with
different triggers, or a shared drug mechanism with different clinical
uses or toxicities. Ask me to name all the diagnoses or presentations that
share this mechanism, and then ask: what is the one upstream or downstream
difference that explains why these conditions present differently despite
sharing the core mechanism?
[!NOTE] Application Note: This step trains the discrimination skill from Module 15, Step 15.3 but at the mechanism level rather than the script level — useful for conditions that are frequently confused precisely because they share an underlying pathway (e.g. conditions sharing a final common inflammatory or ischaemic pathway with different upstream triggers).
Prompt:
#VibeRounds Give me a diagnosis that can arise through more than one
distinct basic-science mechanism or pathway — for example a single
clinical syndrome with several different underlying causes that converge
on the same presentation. Ask me to name as many distinct mechanistic
routes to this diagnosis as I can. Then ask: if I only had time to
investigate one route first, which would I prioritise, and why — based on
prevalence, reversibility, or danger if missed, not just listing them in
no particular order.
[!NOTE] Application Note: This is the mirror image of Step 16.3 and closes a common integration gap — learners often know one mechanistic route to a diagnosis well (the one taught first) and treat it as the only route, missing that the same clinical syndrome can converge from genuinely different pathophysiology requiring different work-ups.
Prompt:
#VibeRounds Give me a drug class and only its core mechanism of action —
no indications, no side effects yet. Ask me to derive, from the mechanism
alone: (1) what conditions this drug class should treat, (2) what adverse
effects or toxicities are predictable from the same mechanism, and (3) what
drug-drug or drug-disease interactions would follow logically from this
mechanism. Only after I attempt all three, tell me what I derived
correctly and what required information beyond the mechanism that I
couldn't reasonably have predicted.
[!NOTE] Application Note: Pharmacology is the basic-science domain where the mechanism-to-clinical-effect link is most directly derivable by reasoning rather than requiring memorisation — this step exploits that to build the habit of deriving rather than recalling. Naming what genuinely couldn’t be predicted from mechanism alone (idiosyncratic reactions, population-specific effects) is as important as the derivation itself — it teaches the boundary of what basic science can and can’t predict.
Prompt:
#VibeRounds Pause. Give me a formative checkpoint: (1) diagnosis-to-mechanism
direction — score out of 10, with one specific strength and one targeted
gap; (2) mechanism-to-diagnosis direction — score out of 10, same format.
Tell me explicitly if one direction is notably weaker than the other — that
asymmetry is itself the most useful finding from this checkpoint, more
useful than either score alone.
[!NOTE] Application Note: The asymmetry between the two directions is the actual diagnostic signal this module is built to surface — a large gap between forward and reverse performance indicates memorised, non-integrated knowledge even when the forward direction looks strong on its own.
Prompt:
#VibeRounds Based on everything we covered this session, build a compact
two-way map: list each diagnosis we discussed, its core mechanism in one
line, and at least one other diagnosis it connects to either through a
shared mechanism or a shared clinical presentation. Format it so I could
redraw it from memory as a simple node-and-link sketch.
[!NOTE] Application Note: Concept mapping is a validated technique specifically for making basic-science/clinical links explicit and shareable (see vertical-integration literature on concept maps as an integration tool). The instruction to make it redrawable from memory — rather than just presented as a reference table — is deliberate; the consolidation value comes from the learner being able to reconstruct the links, not from having a copy of them.
Prompt:
#VibeRounds Apply a critical awareness lens to the integration work we did
today: (1) For any diagnosis where I could state the clinical picture but
struggled with the mechanism, what is the risk of relying on that
diagnosis pattern-matched rather than mechanistically understood — where
might it fail me on an atypical case? (2) For any mechanism where I could
explain the science but missed an obvious diagnosis it should have
predicted, what does that gap suggest about how I've been studying basic
science — as connected knowledge or as an isolated subject? (3) What is
one diagnosis-mechanism pair from today I am at risk of forgetting within
a month, and what would re-test it most efficiently?
[!NOTE] Application Note: This step extends Framework D’s critical-awareness lens to a risk specific to this module: clinically fluent but mechanistically hollow knowledge, which is exactly the failure mode that genuine vertical integration is designed to prevent and that premature encapsulation (see module-level Theoretical Basis note above) can mask.
Objective: Train the compression step that happens before differential generation or script activation — restating a raw case as a small set of abstract, paired clinical qualifiers (acute/chronic, progressive/static, focal/diffuse) — rather than letting the learner jump straight from raw findings to a diagnosis list.
Indication: Logically prerequisite to Module 1 and Module 15, though numbered to follow them in this repository for cross-reference stability. Best introduced as early as possible in training — ideally before a learner has built much of a script library at all — since a poor problem representation habit, once established, silently degrades every downstream reasoning step regardless of how strong the learner’s knowledge base becomes.
[!NOTE] Theoretical Basis: This module operationalises semantic qualifiers and problem representation (Bordage & Lemieux, 1991, “Semantic Structures and Diagnostic Thinking of Experts and Novices,” Academic Medicine, 66(9 Suppl), S70–S72). The empirical finding underlying it: expert clinicians, before generating a differential, restate a case using abstract paired qualifiers — transforming “the patient has had a cough for three days that is getting worse” into something like “acute, progressive, respiratory” — and this restatement (the problem representation) is what subsequently activates an illness script (see Module 15). Bordage’s research found that diagnostic accuracy correlated more strongly with the quality of a learner’s problem representation — coherent and abstracted versus sprawling and concrete — than with raw factual knowledge. A novice with a strong knowledge base but a poor (overly literal, overly long, or disorganised) problem representation will still generate a worse differential than an expert with a tight one. This module trains the compression step in isolation, deliberately separated from diagnosis generation, because conflating the two — jumping straight to “what is it” without first asking “how would I describe this abstractly” — is precisely the habit this module exists to interrupt.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a clinical reasoning partner helping me practise
problem representation — restating a case in abstract semantic qualifiers
before I attempt any diagnosis. A semantic qualifier is a paired,
abstracted descriptor: acute vs chronic, progressive vs static vs
fluctuating, focal vs diffuse, proportionate vs disproportionate, typical
vs atypical, and similar pairs appropriate to the case. When I describe a
case to you, do not let me name a diagnosis yet. First, ask me to restate
the case in 1–2 sentences using only semantic qualifiers — no diagnosis
names, no specific findings restated verbatim. Only once I have produced a
qualifier-based summary should you let the conversation move toward a
differential. If I try to skip ahead to naming a diagnosis, redirect me
once: 'Give me the abstracted summary first — what kind of problem is
this, before what specific problem it is.' Confirm you understand before
we begin.
[!NOTE] Application Note: Run once per problem-representation session. The redirect-once pattern mirrors Module 1, Step 1.0’s answer-withholding discipline, but is applied to an earlier point in the reasoning chain — withholding the diagnosis attempt, not the answer. This is a deliberately narrower and earlier intervention than Module 1’s.
Prompt:
#VibeRounds Give me a case history in full, ordinary clinical language —
the way it would actually be told by a patient or in a referral note. Ask
me to compress it into a one-sentence semantic qualifier summary before
anything else. Once I produce it, tell me: which qualifiers I chose well,
which raw detail I correctly abstracted away as noise, and which detail I
abstracted away that I actually should have kept as a qualifier in its own
right.
[!NOTE] Application Note: The third part of the feedback — detail wrongly discarded — is the most instructive failure mode. Over-compression (losing a detail that should have been a qualifier, such as missing that a pain is “exertional” because it got folded into a generic “chronic pain” summary) is as damaging to downstream reasoning as under-compression (failing to abstract at all and just restating the whole history).
Prompt:
#VibeRounds Give me the same case history twice — once with a strong,
tightly abstracted problem representation already modelled for me, and
once with a weak one (too literal, too long, or missing a key qualifier).
Do not tell me which is which. Ask me to identify which representation is
stronger and explain specifically why, in terms of what it would and
would not successfully narrow the differential toward.
[!NOTE] Application Note: This step trains recognition before production — useful early in the module, or whenever a learner is struggling to self-generate qualifiers, since judging an existing representation is a lower-effort entry point than building one from scratch.
Prompt:
#VibeRounds Give me a single set of raw clinical findings that could
honestly be represented two different ways depending on which qualifiers
are emphasised — for example, findings that could be framed as either
"acute, focal, progressive" or "subacute, diffuse, fluctuating" depending
on which details are weighted more heavily. Ask me to produce both
representations, then ask: which differential does each representation
point toward, and which one of the two representations more faithfully
matches the actual case timeline once I look again?
[!NOTE] Application Note: This is the core insight-generating drill of the module — it demonstrates concretely, with a single case, that the differential is downstream of the representation rather than the raw findings, and that representation is therefore a real source of diagnostic error in its own right, independent of knowledge gaps.
Prompt:
#VibeRounds Give me a semantic qualifier summary only — no raw findings, no
history, just the abstracted qualifiers (e.g. "acute, focal, painful,
disproportionate to exam findings"). Ask me what diagnoses this
representation alone should activate, before any further detail is given.
Then reveal the raw case and ask whether the additional detail confirmed,
refined, or contradicted what the qualifiers alone suggested.
[!NOTE] Application Note: This step is the direct bridge to Module 15, Step 15.1 (Enabling-Conditions-Only Script Trigger) — run this step first if a learner is struggling with Module 15’s drill, since weak qualifier formation is a common hidden cause of weak script activation that looks, on the surface, like a knowledge gap.
Prompt:
#VibeRounds Give me a case along with a flawed problem representation
someone might plausibly produce — one that is too literal (just restates
findings with no abstraction), too narrow (drops a clinically important
qualifier), or internally inconsistent (qualifiers that don't actually fit
the case as described). Ask me to identify what specifically is wrong
with the representation before I attempt to fix it. Then ask me to produce
a corrected version.
[!NOTE] Application Note: Naming the category of flaw (too literal / too narrow / inconsistent) before fixing it builds a transferable diagnostic vocabulary for the learner’s own future self-correction — the goal is for the learner to eventually catch their own weak representations mid-case, not just in retrospective drills.
Prompt:
#VibeRounds Pause. Give me a formative checkpoint: (1) abstraction quality —
am I compressing appropriately or just restating findings verbatim? Score
out of 10 with one specific strength and one targeted gap; (2) qualifier
selection — am I choosing qualifiers that actually narrow the differential,
or generic ones that don't discriminate? Score out of 10, same format. End
with one sentence on which qualifier pair I use most skilfully and which
one I tend to default to inappropriately.
[!NOTE] Application Note: The “default to inappropriately” framing is intentional — learners often over-rely on one familiar qualifier pair (acute/chronic is the most common) and under-use others (proportionate/disproportionate, typical/atypical) regardless of case fit. Naming the default tendency explicitly is more useful than a generic accuracy score.
Prompt:
#VibeRounds Looking back at this session, identify any moment where a weak
or rushed problem representation led downstream to a weaker differential
than my knowledge base should have produced — in other words, a case
where I "knew enough" but represented the problem poorly enough that I
still reasoned my way to the wrong place. Name the specific representation
choice that caused the downstream miss, separate from any knowledge gap.
[!NOTE] Application Note: This step exists to keep representation and knowledge errors analytically separate — without it, a representation failure tends to get misdiagnosed by the learner as a knowledge failure (“I didn’t know enough about X”) when the actual fix is a different one (representing the case better, not learning more facts).
Prompt:
#VibeRounds Apply a critical awareness lens to problem representation
specifically: (1) What is the risk of abstracting a case too early or too
aggressively — what specific, case-defining detail could be lost in
translation to qualifiers, and in what kind of case is that most
dangerous? (2) How would over-confident early representation differ from
the premature closure risk already named for script-based reasoning? (3)
What is the single best safeguard against a confidently-stated but wrong
problem representation steering the whole rest of the case? Be honest but
constructive.
[!NOTE] Application Note: This step extends Framework D’s bias taxonomy to the earliest possible point of failure in the reasoning chain — a flawed problem representation is in some ways a more dangerous error than a flawed differential, because everything generated downstream of it (scripts, mechanisms, the differential itself) inherits the error silently and the learner may never trace a wrong final answer back to this first step. See also Module 15, Step 15.4 for the related but distinct risk of script mismatch at a later stage of reasoning.
Objective: Train reasoning over networks of interacting findings — where a finding’s diagnostic weight changes depending on what else is already known — rather than treating each finding as an independent, isolated piece of evidence pointing at a single mechanism or diagnosis.
Indication: Once a learner can reliably link a single mechanism to a single diagnosis (Module 16) and can activate and discriminate between individual scripts (Module 15), this module trains the next layer: how findings interact — how the presence of one piece of evidence changes the diagnostic value of another, and how causal structure (what causes what) differs from simple statistical association (what correlates with what).
[!NOTE] Theoretical Basis: This module operationalises causal and probabilistic network reasoning in clinical cognition, drawing on Bayesian network models of diagnostic inference (Pearl, J., 1988, Probabilistic Reasoning in Intelligent Systems, Morgan Kaufmann) as applied to medicine, and on the broader distinction between causal and associative reasoning that underlies modern clinical decision-support modelling. The core insight this module trains: clinical findings are not independent pieces of evidence that simply sum toward a diagnosis. A finding’s diagnostic weight is conditional — it changes depending on what else is already known, sometimes dramatically. Two related distinctions are drilled throughout: (1) conditional dependence — how a finding that would normally be weakly diagnostic becomes strongly diagnostic (or vice versa) once another finding is already established; and (2) causal vs. correlational structure — distinguishing findings that cause or are caused by a condition from findings that merely co-occur with it for some other (often confounded) reason. Module 16 trains single mechanism-to-diagnosis links; this module trains reasoning about how multiple such links interact, compete, and modify one another within a single case.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a clinical reasoning partner helping me practise
network-style reasoning — how the diagnostic weight of one finding changes
depending on what else is already known, rather than treating findings as
independent. When you give me a case, reveal findings one at a time rather
than all at once. After each new finding, ask me explicitly: does this
finding change how much weight I give to a finding I already have — not
just what new diagnosis it suggests on its own? Push me to state the
direction of the shift (does this finding make an earlier one more or less
diagnostically significant) and not just add it to a growing list. Confirm
you understand before we begin.
[!NOTE] Application Note: Run once per network-reasoning session. The one-finding-at-a-time delivery is deliberate and structural, not a stylistic choice — conditional reasoning cannot be drilled if all findings are presented simultaneously, because the learner never has to revisit and reweight an earlier finding in light of a later one.
Prompt:
#VibeRounds Give me a case findings one at a time. After each new finding,
ask me two things in order: (1) on its own, what does this finding
suggest? (2) given everything I already know, has this finding changed the
diagnostic weight of any earlier finding — made it more specific, less
specific, or essentially irrelevant now? Only after I answer both, tell me
whether I reweighted correctly, and if not, explain the conditional
relationship I missed.
[!NOTE] Application Note: This is the core drill of the module. Most learners answer question (1) well and skip question (2) entirely by default — explicitly forcing the second question is what distinguishes this module from ordinary sequential history-taking practice already covered in Module 1.
Prompt:
#VibeRounds Give me two findings that commonly co-occur with a condition —
one that is genuinely on the causal pathway (it causes the condition, is
caused by it, or shares a direct mechanistic link) and one that merely
correlates with it for a separate reason (a shared risk factor, a
confounding demographic association, or coincidental co-occurrence in this
population). Do not tell me which is which. Ask me to identify which
finding is causally connected and which is merely correlated, and to state
the actual mechanism — or absence of one — for each.
[!NOTE] Application Note: This step targets a specific, common reasoning error: treating any statistically associated finding as if it carries causal diagnostic weight. A learner who cannot tell the two apart will systematically misweight evidence even with a strong basic-science foundation (see Module 16) — the mechanism knowledge has to be deployed correctly, not just possessed.
Prompt:
#VibeRounds Give me a case where two plausible causes could each explain
the same finding. Ask me to name both candidate causes. Then introduce one
additional piece of evidence that makes one cause significantly more
likely. Ask me: does this new evidence simply add support to the favoured
cause, or does it actively reduce how much I should still be considering
the other cause — even though the other cause hasn't been directly ruled
out? Only after I answer, explain the "explaining away" effect at work.
[!NOTE] Application Note: This drills the network-reasoning phenomenon sometimes called “explaining away” — where confirming one cause of a finding can legitimately lower the probability of a separate, otherwise-unrelated cause, even without any new evidence directly against that second cause. This is counter-intuitive for most learners and is a recognised source of diagnostic error when handled wrong in either direction — over-anchoring on the favoured cause, or failing to deprioritise the competing one at all.
Prompt:
#VibeRounds Give me a case where a finding I would expect to be present,
given the working diagnosis, is instead absent or negative. Ask me first:
does this absence rule out the working diagnosis, make it less likely but
not impossible, or does it not actually matter much because the finding
was never that specific in the first place? Then ask me what else in the
case network should now be reweighted as a result of this surprising
negative.
[!NOTE] Application Note: A negative or absent finding is frequently under-weighted by learners relative to its actual network effect — this step trains treating absence as informative evidence that ripples through the rest of the case, not as a non-event to be set aside.
Prompt:
#VibeRounds Based on a case I describe, help me build a simple network map:
list each major finding, and for each pair of findings that meaningfully
interact, state the direction of the relationship — does knowing A make B
more diagnostically significant, less significant, or unrelated? Keep it
to the most clinically important 4–6 findings, not an exhaustive list.
Once built, ask me: which single finding in this network, if it turned out
to be wrong or misrecorded, would do the most damage to the rest of my
reasoning?
[!NOTE] Application Note: The closing question is a deliberate fragility check — identifying the single most load-bearing finding in the network surfaces where the learner’s reasoning has a hidden single point of failure, which is a different and more advanced skill than simply listing findings and their individual diagnostic weights.
Prompt:
#VibeRounds Pause. Give me a formative checkpoint: (1) conditional
reweighting — am I correctly updating earlier findings as new ones arrive,
or treating each finding independently? Score out of 10 with one specific
strength and one targeted gap; (2) causal vs. correlational discrimination —
score out of 10, same format. End with one sentence on which kind of
network relationship — reweighting, explaining away, or causal
discrimination — is my strongest instinct in this session.
[!NOTE] Application Note: Structural analogue of Module 1, Step 1.4, re-scoped to network-reasoning competence specifically rather than overall diagnostic logic.
Prompt:
#VibeRounds Looking back at this session, identify any moment where I
reasoned about a finding in isolation when I should have reweighted an
earlier finding because of it — a moment where linear, list-building
reasoning produced a different (and worse) conclusion than network
reasoning would have. Name the specific reweighting I missed and what it
would have changed about my differential.
[!NOTE] Application Note: This is the network-reasoning analogue of Module 15, Step 15.7 (Script vs. Stepwise debrief) — both steps audit which mode of reasoning was used, not just which diagnosis was reached, on the premise that the mode of error is more instructive to the learner than the outcome alone.
Prompt:
#VibeRounds Apply a critical awareness lens to network-style reasoning
specifically: (1) What is the risk of over-fitting a network — seeing
conditional relationships between findings that are not actually there,
or are coincidental rather than mechanistic? (2) How does this differ from
ordinary confirmation bias — what makes a false network connection harder
to notice than a simple anchoring error? (3) What is the single best
safeguard against a confidently-reasoned but wrong network of findings
steering the whole case? Be honest but constructive.
[!NOTE] Application Note: This step extends Framework D’s bias taxonomy to a risk specific to this module — network reasoning is more powerful than linear reasoning when the relationships are real, but more dangerous when they are imagined, because a fabricated network of interlocking “evidence” can feel more rigorously reasoned than a simple wrong guess, making it harder for the learner (or a senior reviewer) to spot the error.
Objective: Train the habit of reasoning about a patient’s social, economic, and community context as a clinically active variable — not background colour — across history-taking, differential diagnosis, and management planning, and to connect individual-case social context to population-level community health patterns.
Indication: Any case where the working diagnosis, the management plan, or the explanation for non-adherence or recurrence cannot be fully understood from biology alone — chronic disease management, recurrent presentations, paediatric and family cases (Module 2, Module 3), resource-constrained settings (Module 14), and any registry-level review (Module 6) where population patterns may reflect social determinants rather than purely biological ones.
[!NOTE] Theoretical Basis: This module operationalises the WHO Social Determinants of Health (SDOH) framework — the conditions in which people are born, grow, live, work, and age, and the wider set of forces and systems shaping the conditions of daily life (income, education, housing, food security, employment, social inclusion, gender, race, early-childhood factors, and access to health services). SDOH are estimated to account for a substantial share of health outcomes independent of biology or the healthcare system itself, yet remain the most consistently under-taught and under-elicited domain in clinical history-taking. This module also draws on Engel’s biopsychosocial model (1977, “The Need for a New Medical Model: A Challenge for Biomedicine,” Science, 196(4286), 129–136) to keep social context integrated into clinical reasoning rather than treated as a separate, optional add-on, and on the CanMEDS Health Advocate role (Royal College of Physicians and Surgeons of Canada) to give the skill a concrete competency structure: identifying patients at risk due to social determinants, knowing strategies and resources to address disparities, and advocating for patient access to those resources. The repeated finding in the literature this module addresses directly: social-determinant content is acknowledged as important by learners but is rarely elicited systematically in practice unless explicitly structured into the history-taking process — good intentions do not reliably produce the behaviour without a scaffold.
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a clinical reasoning partner helping me practise
integrating social and community context into case reasoning — not as a
separate checklist item, but as something that actively shapes diagnosis,
management, and prognosis. When I describe a case, do not let me move to a
management plan until I have stated at least the core social determinants
relevant to this patient: income/occupation, housing stability, food
security, social support, education, and access to care (which may
include cost, distance, or transport — see Module 14 if resource
constraints are the dominant issue). For each one I have not addressed,
prompt me once before continuing. Then, and only then, ask me how any of
these factors should change — not just complicate — my management plan.
Confirm you understand before we begin.
[!NOTE] Application Note: Run once per social-context session, or fold into the closing phase of any other module’s case when social context has not yet been addressed. The “change, not just complicate” framing is deliberate — the goal is to push the learner past acknowledging a social factor exists toward actually altering the plan because of it, which is the harder and more clinically valuable step.
Prompt:
#VibeRounds Act as a patient or caregiver answering questions about social
circumstances relevant to this case — income and occupation, housing,
household composition, food security, social support, and barriers to
accessing care. I will ask you questions one at a time as if taking a
social history. Answer realistically and specifically, including some
detail I would only learn by asking a good follow-up question, not just
the first obvious answer. After the exchange, tell me which question I
asked elicited the single most clinically useful piece of information, and
which important social domain I never asked about at all.
[!NOTE] Application Note: Practising the elicitation itself, not just receiving a pre-written social history, builds the questioning habit directly — a recognised gap in training is not lack of awareness that SDOH matters, but lack of practised, comfortable phrasing for asking about it without feeling intrusive.
Prompt:
#VibeRounds Give me a single social determinant — for example food
insecurity, precarious housing, informal/unsafe employment, or limited
health literacy. Ask me to trace, step by step, at least two distinct
clinical pathways by which this determinant could plausibly worsen a
named chronic condition or increase risk of a named acute presentation.
Push me past the first obvious pathway to a second, less obvious one.
Then ask: which pathway would most change what I actually do in clinic
today, versus which is more relevant to long-term population health.
[!NOTE] Application Note: This is the social-medicine analogue of Module 16’s mechanism-to-diagnosis drill — instead of basic-science mechanism to diagnosis, it trains social-determinant to clinical-pathway, with the same discipline of forcing the learner to generate the connection rather than receive it.
Prompt:
#VibeRounds Present me with a case of a patient who is not adhering to a
treatment plan. Before I am allowed to label this as "non-compliance," ask
me to generate at least three distinct social or structural explanations
that could produce the same observed behaviour — for example cost,
conflicting work hours, transport, competing caregiving duties, or health
literacy. For each, ask what a change to the plan itself — not a change to
the patient — could address it.
[!NOTE] Application Note: This step deliberately interrupts a common and clinically costly framing error — attributing non-adherence to patient motivation or character before structural explanations have been considered. The closing instruction (changing the plan, not the patient) keeps the exercise solution-oriented rather than purely diagnostic of blame.
Prompt:
#VibeRounds Give me a short description of a community or population
context — for example a specific neighbourhood, occupational group, or
displaced/migrant population. Ask me what condition patterns I would
expect to see at elevated rates in this population specifically because
of their social and environmental context, before any individual case is
given. Then present an individual case from that context and ask whether
the case fits the expected community pattern or represents something the
community pattern would have caused me to overlook.
[!NOTE] Application Note: This step bridges individual-case reasoning to the registry-level, population-scale thinking in Module 6 — and trains the discipline of using a community pattern as a prior to sharpen attention, not as a substitute for individually assessing the patient in front of you. The closing question guards specifically against the latter failure mode.
Prompt:
#VibeRounds Present me with a case where the medically ideal plan and the
practically achievable plan diverge because of a health-system or access
barrier — cost of medication, distance to follow-up, time off work needed
for appointments, or insurance/coverage gaps. Ask me first what the ideal
plan is, then ask me to redesign it as the best plan that is actually
achievable given the barrier, without simply abandoning the clinical goal.
Affirm explicitly where my redesigned plan preserves the core clinical
intent despite the constraint.
[!NOTE] Application Note: Structurally similar to Module 14’s substitution-and-workaround drills, but the constraint here originates in the patient’s social and economic circumstances rather than facility-level resource availability — the two are related but distinct categories of constraint and a learner should be able to reason through both.
Prompt:
#VibeRounds Based on the social barriers identified in this case, ask me
to name one concrete health-advocacy action available to me as the
treating clinician or learner — beyond documenting the barrier — that
could meaningfully help this specific patient (for example: a referral to
a social worker or community resource, a prescription substitution for
cost, or a letter supporting workplace accommodation). Ask me to be
specific to this patient, not generic. Then ask: is there a second,
separate action that addresses the system-level pattern rather than just
this one patient?
[!NOTE] Application Note: Directly operationalises the CanMEDS Health Advocate competency at two levels — the individual-patient action and the system-level action — which the literature identifies as commonly conflated or collapsed into “raising awareness” alone without a concrete next step at either level.
Prompt:
#VibeRounds Pause. Give me a formative checkpoint: (1) elicitation —
am I asking specific, non-generic questions about social context, or
defaulting to a checklist tone? Score out of 10 with one specific strength
and one targeted gap; (2) integration — am I actually changing my plan
because of social factors, or just noting them? Score out of 10, same
format. End with one sentence on which social domain (income, housing,
food security, support, access) I engage with most naturally, and which
one I tend to skip.
[!NOTE] Application Note: Structural analogue of Module 1, Step 1.4. Naming the consistently-skipped domain is the most actionable part of this checkpoint — most learners have one or two SDOH categories they reliably under-elicit, and that pattern is more useful feedback than an aggregate score.
Prompt:
#VibeRounds We have reached the end of this case. Produce a closing
summary covering: (1) the single social determinant that most changed this
case's management, and how; (2) one community-level insight this case
suggests — something that, if true of other patients in this context,
would be worth flagging beyond this individual case; (3) one specific
thing I should ask about earlier next time a similar case presents. Open
with a sentence acknowledging that social-context reasoning is a distinct
clinical skill, not a "soft" add-on to the real diagnostic work.
[!NOTE] Application Note: Structural analogue of Module 1, Step 1.7 and Module 14, Step 14.7 — the opening line is deliberate; social-determinant reasoning is sometimes implicitly treated by learners as lower-status than biomedical reasoning, and the framing should actively counter that, consistent with the literature’s finding that SDOH content is often acknowledged as important in principle but deprioritised in practice.
Prompt:
#VibeRounds Apply a critical awareness lens to the social-context reasoning
in this session: (1) Did any inference I made about this patient's social
circumstances rely on a demographic or community-level stereotype rather
than something actually elicited from this specific patient? (2) What is
the risk of over-attributing a clinical presentation to social context and
under-investigating a biological cause as a result? (3) What is the risk
of the reverse — treating a socially-patterned presentation as purely
biological and missing the structural driver entirely? (4) What would a
critic say about using community-level patterns (Step 19.4) to inform
individual-patient reasoning? Be honest but constructive.
[!NOTE] Application Note: This step extends Framework D’s bias taxonomy to a risk domain specific to this module and arguably more consequential than the cognitive biases named in Domain 1 — community-level pattern reasoning (Step 19.4) is clinically useful but sits directly adjacent to demographic stereotyping if not handled with discipline; point (1) exists specifically to force the learner to distinguish information actually elicited from this patient (Step 19.1) from an assumption imported from the community pattern. Point (3) guards against the literature’s documented critique that social-determinant framing can itself become a way of explaining away a missed biomedical diagnosis if applied uncritically.
Objective: Train decision-making under real time pressure and incomplete information the way experienced practitioners actually do it — generating one plausible course of action and mentally simulating it forward to check for failure, rather than comparing multiple options side by side before choosing.
Indication: Time-critical clinical moments where there is no time for a full differential or a side-by-side option comparison — night-shift deterioration calls, rapid response activations, acute triage, and any scenario in Module 4, Step 4.4 (Night Shift Stat Call Triage). Not a replacement for the slower, more thorough reasoning trained in Module 1 or Module 18 — this module trains the specific, narrower skill needed when there genuinely is no time for those.
[!NOTE] Theoretical Basis: This module operationalises the Recognition-Primed Decision (RPD) model (Klein, G., 1998, Sources of Power: How People Make Decisions, MIT Press), developed from field studies of experienced decision-makers under time pressure — firefighters, ICU nurses, military commanders, pilots. Klein’s central finding, which broke with the classical decision-theory assumption that good decisions require comparing multiple options: experienced practitioners under time pressure typically do not generate and compare several options at once. Instead, they recognise a situation as typical of a pattern they have seen before, retrieve a single plausible course of action associated with that pattern, and mentally simulate it forward — running the plan in their head to check whether it would work and where it might fail — before committing. If the mental simulation reveals a flaw, the practitioner adjusts the plan or retrieves the next most plausible option; they rarely compare two full options simultaneously. This is a close relative of Module 15’s illness-script retrieval and shares the System 1 territory of dual-process theories of clinical reasoning, but RPD is specifically a model of action selection under time pressure, not diagnostic pattern-matching — the unit being trained here is “what do I do right now,” not “what is this.”
Phase 1 · Initiation → Phase 2 · Execution → Phase 3 · Closure / Review
Prompt:
#VibeRounds You are a clinical reasoning partner helping me practise
recognition-primed decision-making for time-critical scenarios — the way
experienced clinicians actually decide under real time pressure. Do not
ask me to compare multiple options side by side. Instead, when I describe
a deteriorating or urgent scenario, ask me: (1) what is the single most
plausible course of action, based on recognising this situation as similar
to one I have encountered or learned about before? (2) Now mentally
simulate that plan forward — where would it fail, and how would I know?
Only if my mental simulation reveals a serious flaw should we move to a
second option. Keep the pace fast — this is meant to feel like real time
pressure, not a leisurely Socratic dialogue. Confirm you understand before
we begin.
[!NOTE] Application Note: Run once per RPD session, and explicitly re-orient before each scenario if the AI starts defaulting back to comparative option-listing — this is the most common failure mode, since most LLMs default to enumerating several options when asked for a clinical plan, which is precisely the pattern RPD is designed to train away from.
Prompt:
#VibeRounds Describe a sudden clinical deterioration or abnormal result to
me in the compressed, urgent way it would actually be reported — a phone
call or a glance at a monitor, not a full case write-up. Give me five
seconds' worth of information, no more. Ask me immediately: what does this
situation remind me of, and what is my first instinct for what is
happening? Do not let me ask clarifying questions yet — force the
recognition response first, then open up further information only after I
commit to an initial read.
[!NOTE] Application Note: This step deliberately denies the learner the full information-gathering opportunity that Module 1 and Module 17 encourage — that is the point. Real time-critical recognition happens with incomplete information, and the skill being trained is producing a usable first read under that constraint, not refusing to commit until more data arrives.
Prompt:
#VibeRounds Based on the scenario, ask me to commit to one course of
action — not a list of options, one plan. Once I state it, ask me to
mentally simulate it forward out loud: what happens in the next 5 minutes
if this plan is right? What is the first sign that would tell me it is
failing? At what point in this simulation would I need to abandon or
escalate the plan? Only after I complete the simulation should you tell me
whether my plan and my failure-detection point were sound.
[!NOTE] Application Note: This is the core RPD drill. The forward simulation — not the plan itself — is the skill being trained; a learner who states a reasonable plan but cannot name how they would know it was failing has not actually completed the RPD cycle, even if the plan itself is clinically correct.
Prompt:
#VibeRounds Continue the scenario from where my forward simulation
revealed a flaw, or introduce a new piece of information that breaks my
original plan. Ask me: does this change require abandoning my recognition
of the situation entirely and re-pattern-matching from scratch, or does it
just require adjusting the plan while keeping the same underlying
read of the situation? Push me to be explicit about which one is
happening — they require different responses.
[!NOTE] Application Note: Distinguishing “wrong plan, right read of the situation” from “wrong read of the situation entirely” is a higher-order RPD skill than simple plan adjustment — Klein’s research found experienced decision-makers are notably better than novices at recognising when they need to abandon their situational read itself, not just tweak the plan within it.
Prompt:
#VibeRounds For the scenario we are working through, ask me to state — as
fast as I can, without deliberation — the single specific finding or
threshold that would make me escalate to a senior right now, regardless of
how confident I feel in my current plan. Then ask me to defend that
threshold in one sentence: why this specific sign and not an earlier or
later one.
[!NOTE] Application Note: Direct extension of Module 4, Step 4.4 (Night Shift Stat Call Triage), which already asks “at what threshold do we escalate” but without the RPD structure around it — this step adds the speed constraint and forward-simulation discipline that sharpens the threshold question into a trained reflex rather than a reasoned-out answer with time to spare.
Prompt:
#VibeRounds Run a rapid sequence of four short deterioration scenarios,
one after another, each delivered in compressed, urgent form as in Step
20.1. For each, force me through the full cycle fast: recognise, commit to
one plan, state the forward-simulation failure point, state the escalation
threshold. Move to the next scenario as soon as I complete the cycle —
do not let me linger. At the end, tell me which scenario I handled with
genuine recognition-based speed versus which one I was visibly working out
analytically in real time, even if I got the right answer either way.
[!NOTE] Application Note: The closing distinction — genuine recognition speed versus analytical reasoning that merely looks fast — is the most clinically important feedback this step can give. A learner who reasons their way to a correct answer slowly, even if they land on it, has not yet built the pattern library that RPD depends on; that distinction is invisible from outcome alone and only surfaces by comparing pace and confidence across scenarios.
Prompt:
#VibeRounds Pause. Give me a formative checkpoint: (1) recognition speed —
am I committing to a plausible first read quickly, or hesitating until I
have gathered more information than the scenario allows? Score out of 10
with one specific strength and one targeted gap; (2) forward-simulation
quality — am I genuinely simulating the plan's failure points, or just
stating a plan and moving on? Score out of 10, same format. End with one
sentence on whether my biggest current limitation is recognition (not
having the pattern) or simulation (having the pattern but not stress-
testing it).
[!NOTE] Application Note: Separating recognition limitations from simulation limitations matters because the fix for each is different — a recognition gap is addressed by more case exposure and pattern-building (see Module 15), while a simulation gap is addressed by more deliberate forward-checking practice, which is what this module trains directly.
Prompt:
#VibeRounds Looking back at this session, identify any moment where I
should have used fast recognition-primed decision-making but instead
defaulted to slow, deliberative comparison of options — costing time the
real scenario would not have allowed. Also identify the reverse: any
moment where I committed to a fast plan via recognition when the situation
actually had enough time and ambiguity to warrant slower, more
deliberative reasoning instead. Name the specific cue that should have
told me which mode the situation called for.
[!NOTE] Application Note: This is the RPD-specific analogue of Module 15, Step 15.7 and Module 18, Step 18.7 — all three audit which reasoning mode was used, not just the outcome, on the premise that mode-selection error is more instructive and more dangerous in practice than a single wrong answer within the right mode.
Prompt:
#VibeRounds Apply a critical awareness lens to recognition-primed
decision-making specifically: (1) What is the risk of pattern-matching to
the wrong prior experience under time pressure, and how would I know it
happened, given that RPD by design does not involve comparing alternative
explanations at the moment of decision? (2) How is this risk different
from the premature closure risk already named for illness scripts? (3)
What is the single best safeguard against a fast, confident,
wrong recognition-primed decision in a genuinely time-critical scenario,
given that there usually isn't time to fully verify before acting? Be
honest but constructive.
[!NOTE] Application Note: This step extends Framework D’s bias taxonomy to a risk domain that is structurally harder to safeguard than the others in this stack — by design, RPD does not build in the option-comparison step that would normally catch a wrong pattern match, because there usually isn’t time for it. The honest answer to point (3) is typically institutional and structural (a second clinician’s parallel read, a fixed escalation threshold set in advance as in Step 20.4, post-event debrief as in Step 20.7) rather than something the individual decision-maker can reliably self-correct for in the moment — the prompt should surface that limitation rather than imply a purely individual fix exists.