Download the Full Version of textbook for Fast Typing at textbookfull.

com

LLMs and Generative AI for Healthcare (Early

Release) Kerrie Holley

https://textbookfull.com/product/llms-and-generative-ai-for-
healthcare-early-release-kerrie-holley/

OR CLICK BUTTON

DOWNLOAD NOW

Download More textbook Instantly Today - Get Yours Now at textbookfull.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

AI First Healthcare 1st Edition Kerrie Becker M D Holley

Siupo Kerrie L Holley Siupo Becker M D

https://textbookfull.com/product/ai-first-healthcare-1st-edition-
kerrie-becker-m-d-holley-siupo-kerrie-l-holley-siupo-becker-m-d/

textboxfull.com

AI-First Healthcare: AI Applications in the Business and

Clinical Management of Health 1st Edition Kerrie L. Holley

https://textbookfull.com/product/ai-first-healthcare-ai-applications-
in-the-business-and-clinical-management-of-health-1st-edition-kerrie-
l-holley/
textboxfull.com

Building Generative AI Services with FastAPI (Early

Release) 1st Edition Ali Parandeh

https://textbookfull.com/product/building-generative-ai-services-with-
fastapi-early-release-1st-edition-ali-parandeh/

textboxfull.com

Introduction to Generative AI Numa Dhamani

https://textbookfull.com/product/introduction-to-generative-ai-numa-
dhamani/

textboxfull.com
Generative AI with LangChain Ben Auffarth

https://textbookfull.com/product/generative-ai-with-langchain-ben-
auffarth/

textboxfull.com

Building Generative AI Powered Apps A Hands on Guide for

Developers 1st Edition Kansal

https://textbookfull.com/product/building-generative-ai-powered-apps-
a-hands-on-guide-for-developers-1st-edition-kansal/

textboxfull.com

Introduction to Generative AI: An ethical, societal, and

legal overview 1st Edition Numa Dhamani

https://textbookfull.com/product/introduction-to-generative-ai-an-
ethical-societal-and-legal-overview-1st-edition-numa-dhamani/

textboxfull.com

Machine Learning and AI for Healthcare: Big Data for

Improved Health Outcomes Arjun Panesar

https://textbookfull.com/product/machine-learning-and-ai-for-
healthcare-big-data-for-improved-health-outcomes-arjun-panesar/

textboxfull.com

Programming Large Language Models With Azure Open Ai:

Conversational Programming and Prompt Engineering With
Llms (Developer Reference) 1st Edition Esposito
https://textbookfull.com/product/programming-large-language-models-
with-azure-open-ai-conversational-programming-and-prompt-engineering-
with-llms-developer-reference-1st-edition-esposito/
textboxfull.com
 LLMs and Generative AI for
Healthcare
The Next Frontier

With Early Release ebooks, you get books in their earliest form—the
author’s raw and unedited content as they write—so you can take
advantage of these technologies long before the official release of these
titles.

Kerrie Holley and Manish Mathur

LLMs and Generative AI for Healthcare
by Kerrie Holley and Manish Mathur
Copyright © 2024 Kerrie Holley and Manish Mathur. All rights reserved.
Printed in the United States of America.
Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North,
Sebastopol, CA 95472.
O’Reilly books may be purchased for educational, business, or sales
promotional use. Online editions are also available for most titles
(http://oreilly.com). For more information, contact our corporate/institutional
sales department: 800-998-9938 or [email protected].

Editors: Angela Rufino and Michelle Smith

Production Editor: Aleeya Rahman

Interior Designer: David Futato

Cover Designer: Karen Montgomery

Illustrator: Kate Dullea

December 2024: First Edition

Revision History for the Early Release

2024-02-07: First Release
See http://oreilly.com/catalog/errata.csp?isbn=9781098160920 for release
details.
The O’Reilly logo is a registered trademark of O’Reilly Media, Inc. LLMs
and Generative AI for Healthcare, the cover image, and related trade dress
are trademarks of O’Reilly Media, Inc.
The views expressed in this work are those of the authors and do not
represent the publisher’s views. While the publisher and the authors have
used good faith efforts to ensure that the information and instructions
contained in this work are accurate, the publisher and the authors disclaim all
responsibility for errors or omissions, including without limitation
responsibility for damages resulting from the use of or reliance on this work.
Use of the information and instructions contained in this work is at your own
risk. If any code samples or other technology this work contains or describes
is subject to open source licenses or the intellectual property rights of others,
it is your responsibility to ensure that your use thereof complies with such
licenses and/or rights. This book is not intended as medical advice. Please
consult a qualified professional if you require medical advice.
978-1-098-16086-9
Chapter 1. Doctor’s Black Bag

A NOTE FOR EARLY RELEASE READERS

With Early Release ebooks, you get books in their earliest form—the
author’s raw and unedited content as they write—so you can take
advantage of these technologies long before the official release of these
titles.
This will be the 1st chapter of the final book. Please note that the GitHub
repo will be made active later on.
If you have comments about how we might improve the content and/or
examples in this book, or if you notice missing material within this
chapter, please reach out to the editor at [email protected].

Dr. Full, a fictional character in C.M. Kornbluth’s book “The Little Black
Bag,” stumbled upon a mysterious doctor’s bag from the future. The bag
filled with advanced medical instruments and medications transformed Dr.
Full into a more competent physician. As he delved deeper into the bag’s
contents, he discovered a trove of knowledge and capabilities that improved
his medical practice and ignited his curiosity about the future’s possibilities
for healthcare. With each patient he treated and every diagnosis he made, Dr.
Full found himself on the cusp of medical innovation, forever changed by this
glimpse into a future where the boundaries of medical science seemed
boundless. Today, clinicians are metaphorically discovering a futuristic
‘black bag'—specifically, large language models designed for medical
purposes. This chapter explores the possibilities of improving clinical care
using medical-specific large language models in applications and apps.

Potential of Large Language Models and

Generative AI
A large language model (LLM) is a machine learning model trained on
massive text datasets that predicts the probability of a token or sequence of
tokens, modeling complex language representations and knowledge at scale.
LLMs explicitly model the relationships between words and meaning across
extended contexts, enabling more fluent textual understanding and
generation. Their substantial dataset scale and model parameterization
distinguish them from more basic language models. Chapter 2 explores the
workings of LLMs in more detail, addressing tokens, parameters, and more.
While existing medical large language models (LLMs) are already
impressive, they remain in the early stages of development but are poised to
revolutionize healthcare delivery. Currently, a heavy focus is on reducing
clinician administrative and documentation burdens, but this is just the
beginning of how LLMs can transform healthcare. The current generation of
LLMs will likely appear rudimentary in a few years. Expect more impressive
and transformative applications as LLMs and AI technologies evolve.
As of this writing, several medical-specific LLMs have been made available,
but they are not making a significant impact on clinical care for several
reasons:
Data Availability and Quality: LLMs are trained on massive amounts
of data, and the quality of this data is crucial for their performance. In
the medical field, data is often scattered across multiple sources, such as
electronic health records, medical journals, and clinical trials.
Additionally, the data may be incomplete, inaccurate, or inconsistent,
which can negatively impact the performance of LLMs.
Bias and Fairness: LLMs are trained on data that reflects the biases of
the real world. This means that they can perpetuate and amplify existing
biases in healthcare, leading to unfair treatment and outcomes for
specific patient groups. For example, an LLM trained on a dataset of
medical records that is biased against certain racial or ethnic groups may
generate recommendations that are also biased.
 Interpretability and Explainability: LLMs are often considered “black
boxes” because their internal workings are complex and difficult to
understand. This lack of interpretability and explainability makes it
challenging for healthcare professionals to trust and adopt LLMs in their
decision-making processes. Without understanding how LLMs arrive at
their recommendations, there is a risk of making decisions based on
flawed or biased reasoning.
Regulatory Landscape: The regulatory landscape for AI in healthcare
is still evolving, and there are currently no clear guidelines or standards
for developing and deploying LLMs in this domain. This uncertainty can
make it difficult for healthcare organizations to adopt LLMs, as they
may be concerned about potential legal and regulatory risks.
Ethical Considerations: Using LLMs in healthcare raises several
ethical considerations, such as the potential for misuse, the erosion of
human autonomy, and the impact on patient privacy. These ethical
considerations must be carefully addressed before LLMs can be widely
adopted in healthcare.
Despite these challenges, there is growing interest in the potential of LLMs to
improve healthcare delivery. Researchers are working to develop more robust
and reliable LLMs, and healthcare organizations are exploring ways to
integrate LLMs into their workflows. Some examples of emerging LLMs
include:

Med-PaLM: A large language model (LLM) from Google AI, trained

on a massive medical text and code dataset. Med-PaLM can perform
various medical tasks, including answering questions, interpreting
images, generating radiology reports, and calling genomic variants.
Med-PaLM2 is one of the research models that power MedLM1, a
family of foundational models fine-tuned for healthcare.
Flan-PaLM: A 540B parameter LLM from Google AI, trained on a
significant amount of code and medical data. Flan-PaLM has achieved
state-of-the-art performance on various medical benchmarks, including
 the Medical Language Understanding Evaluation (MLUE) and the
MultiMedBench.
BioGPT-JSL: A medical LLM from John Snow Labs, trained on a
massive dataset of biomedical literature, scientific papers, and clinical
data. BioGPT-JSL can perform various medical tasks, including
generating clinical reports, summarizing medical literature, and
answering medical questions.
ClinicalBERT: A medical LLM from Google AI, trained on a massive
dataset of clinical text, including electronic health records (EHRs),
clinical notes, and medical reports. ClinicalBERT can perform various
medical tasks, such as identifying medical entities, classifying medical
text, and generating medical reports.
PubMedBERT: A medical LLM from the National Library of Medicine
(NLM) trained on a massive dataset of biomedical literature from
PubMed. PubMedBERT can perform various medical tasks, such as
answering questions, summarizing texts, and extracting relations.
They still have much ground to cover to surpass the expertise of the most
skilled medical professionals consistently. Still, there is enormous potential
for integrating LLMs as a third element in the doctor-patient relationship.
This potential spans from assisting in diagnosis, documentation, and patient
communication.
Every clinical and administrative healthcare process that requires humans to
create original work from medical coding, patient education, diagnosis,
patient intake, treatment planning, medication management, etc. —is up for
reinvention.
Bard and ChatGPT, generative AI products, don’t represent the universe of
large language models but show consumer-facing examples demonstrating
the capabilities of LLMs. Imagine that LLMs are a vast ocean, and Bard and
ChatGPT are two islands. The islands are beautiful and exciting to explore,
but they only represent a small fraction of the diversity and richness of the
ocean. In short, Bard and ChatGPT are excellent examples of the capabilities
of LLMs, but they represent a small fraction of the potential of this
technology.
Generative AI applications are starting to take off, thanks to the maturation of
the platform layer, the continuous improvement of models, and the increasing
availability of free and open-source models. This gives developers, startups,
and enterprises the tools they need to build innovative applications. As
mobile devices spawned new types of applications with new capabilities like
sensors, cameras, and on-the-go connectivity, large language models are
poised to usher in a new wave of generative AI applications and devices for
healthcare.
These days, there is no shortage of healthcare gadgets to help us optimize our
lives, from fitness trackers to blood pressure monitors to smart insulin pumps.
We like to attach these devices to ourselves to make staying healthy and
working our conditions easier.
Internet search remains a favorite of consumers using Google, as some
describe it as Dr. Google, to understand their symptoms and diagnosis.
However, the evidence is clear that Internet searches correlate to small
increases in diagnostic accuracy and almost none in triage accuracy.2 LLMs
will change this equation as Internet search and LLMs integrate over the next
few years. With the arrival of LLM-based apps and gadgets, we are poised
for an even more significant impact on diagnostic accuracy and use in triage
situations.
LLM-powered healthcare gadgets could revolutionize the way we manage
our health and receive care. LLM-powered chatbots will answer our
questions about our health, and LLM-powered diagnostic tools will help
doctors diagnose diseases more accurately. Clinicals can use medical LLMs
to develop personalized treatment plans and monitor patient progress.
Large language models will revolutionize how consumers and patients
navigate their health and healthcare systems. By providing personalized
insights, recommendations, and support, LLMs can enable patients and
consumers to assume greater responsibility for their well-being and make
well-informed choices regarding their healthcare.
For instance, LLMs will be used to power co-pilots for consumers and
patients:

Personalized health education: LLMs provide consumers and patients

with personalized education about their health conditions, treatment
options, and prevention strategies. Clinicians can tailor to the
individual’s specific needs and preferences and update as the
individual’s health changes.
Medical decision support: LLMs help consumers and patients make
informed decisions about their care. LLMs assist patients and consumers
in making educated choices concerning their healthcare. To illustrate,
they can conduct comparisons of various treatment alternatives, evaluate
the advantages and disadvantages of each option, and forecast the
probabilities associated with various results.
Navigation assistance: LLMs help consumers and patients navigate the
complex healthcare system to find qualified providers, schedule
appointments, and understand insurance coverage.
Emotional support: LLMs support the emotional health of consumers
and patients. LLMs can listen to concerns, offer encouragement, and
connect patients with others facing similar challenges.
LLMs will transform how consumers and patients interact with the healthcare
system. By providing personalized insights, recommendations, and support,
large language models can enable patients and consumers to assume greater
responsibility for their well-being and make well-informed choices regarding
their healthcare.
A few examples of LLM-powered co-pilots in practice include:
A consumer with chronic health conditions could use an LLM-powered
co-pilot to track their symptoms, manage their medications, and get
personalized advice on how to stay healthy.
A patient facing a difficult medical decision, such as whether or not to
have surgery, could use an LLM-powered co-pilot to weigh the risks and
 benefits of each option and get personalized advice from a trusted
source.
A caregiver for a loved one with a complex health condition could use
an LLM-powered co-pilot to manage appointments, coordinate care with
different providers, and get support and guidance.
LLMs are poised to level the playing field, granting more expansive access to
top-tier healthcare information and assistance. This increased accessibility to
health insights will not only empower patients and consumers to comprehend
their well-being but also make informed choices regarding their healthcare. It
will also aid healthcare professionals and the broader healthcare system in
enhancing the overall health and wellness of the populace.
The question isn’t just about what LLMs can do for us, but rather how we can
use LLM and generative AI to enhance our health and well-being. The
question is different from what LLMs can do to us but how we can use LLMs
and generative AI to improve our health and well-being. The following
section describes some of these future apps or applications.

Promise and Possibilities of LLMs in

Healthcare
Eight million people who could have lived with better access to healthcare
die worldwide every year.3 Large language models and generative AI present
a fantastic opportunity to save people’s lives and improve healthcare
globally. Emergent capabilities in large language models are new abilities
that appear as LLMs, trained on larger and larger healthcare and medical-
related text and code datasets become available. As growing volumes of
healthcare and medical data are used to train ever-larger large language
models, new clinical capabilities are arising from these models that were not
originally programmed but rather are emerging from the expanded training.
These new unforeseen abilities related to medical reasoning, treatment
recommendations, epidemiological analysis, and personalized care are
organically developing as the models are exposed to more diverse medical
concepts and interactions at scale. The emergent capabilities are a testament
to the expanding potential of large language models in the healthcare domain
as more medical knowledge across modalities gets incorporated into
advanced model architectures.
These abilities will be challenging to predict or explain but will benefit
healthcare applications. Their emergent capabilities could revolutionize
healthcare unimagined and on a scale previously unimaginable, impacting the
entire planet. Let’s explore a few future apps: a Medical Swiss Army Knife
and a Medical Sherpa. The Medical Swiss Army knife is a consumer tool,
and the Medical Sherpa is for clinicians.

Medical Swiss Army Knife App

An AI startup company introduces a medical-specific large language model
powering a healthcare app called Medical Swiss Army Knife. The Medical
Swiss Army Knife app orchestrates versatility and multifunctional
capabilities for consumers or patients in healthcare contexts. David, a 75-
year-old man, is in love with his Fitbit wearable. He receives a signal
detecting atrial fibrillation (AFib) and contacts his doctor, who refers him to a
cardiologist. David takes medication for high blood pressure and statins to
control his cholesterol. David recently had a calcium scoring test showing
him in a high-risk category. His cardiologist recommends and performs an
atrial fibrillation ablation, but it does not fix the problem. David is rechecked
into the hospital to give the heart a controlled electric shock and
cardioversion to restore a normal rhythm but to no avail.
David’s wife was unsatisfied with the answers from his primary care
physician and local cardiologist, who stated he had no symptoms of
cardiovascular disease, such as any reported shortness of breath or fatigue.
She knew he had minimal physical activity, was a bit overweight, had
persistent AFib, and before the medications, he had high cholesterol and high
blood pressure. She also knew his calcium scoring test showed he was in the
high-risk category. She convinces David to use the Medical Swiss Army
Knife app. The Medical Swiss Army Knife app uses a medical-specific LLM
and an LLM with David’s medical records, medical history, and health
information. The app recommends visiting a preeminent research hospital
and identifying a physician with expertise in Afib to treat his AFib.
David agrees, and the Medical Swiss Army Knife app makes the
appointment, flight reservations, and hotel arrangements. David then engages
in a conversation with the Medical Swiss Army Knife app as he wants to
understand what questions he should be asking. The app suggests:
What is the best treatment plan for me, given my circumstances?
What are the different treatment options available, and what are the risks
and benefits of each?
How does my AFib affect my heart?
What is my risk of stroke?
What should I do if I have an AFib episode?
What are the long-term implications of living with AFib?
The app development, by a reputable company, employs state-of-the-art
security measures to protect patient privacy. The app design attempts to avoid
misinterpretation of the conversation or providing inaccurate information by:
Using a large and diverse dataset to train the LLM: The LLM trained
on a large and diverse dataset of medical conversations. This helps the
LLM learn the nuances of medical language and avoid making mistakes.
Using state-of-the-art natural language processing techniques: The
application employs cutting-edge natural language processing (NLP)
methods to comprehend the conversation effectively. This, in turn,
assists the LLM in pinpointing the essential aspects of the discourse and
refraining from drawing unsupported inferences.
Incorporating feedback from doctors and patients: The app
incorporates feedback from doctors and patients to improve the accuracy
of the LLM. This continuous feedback loop helps identify areas where
the LLM is struggling and make necessary adjustments.
 Providing transparency to users: The app provides transparency to
users about how it works and uses their data, helping users understand
the limitations of the app and to use it responsibly.

The LLM Medical Swiss Army Knife app reminds David that it is not a
replacement for medical advice from his AFib doctor. It informs David that
he should always consult with his doctor before making decisions about his
care. David and his wife fly 2000 miles and check into the recommended
hotel adjacent to the hospital. Both are immediately impressed as the
cardiologist phones and asks if he can stop by and say hello. This personal
service is beyond their expectations. Before the doctor’s meeting, David
opens the Medical Swiss Army Knife app to check on the questions he wants
to ask. The app prompts David if he would like the app to listen in on the
conversation. David informs the doctor he is using an LLM app that will
listen to their conversation and help David better understand the conversation
afterward. The doctor smiles and says of course and reminds David that he
would be happy to answer any questions he has any time before the surgery.
It is now Monday and time for a pre-procedure CT scan in preparation for the
Isolator Synergy ablation clamp to treat Atrial Fibrillation. The CT scan
shows severe blockage in his main arteries, and the cardiologist cautions
David that he is at high risk of a heart attack, so much so that he’s going to be
having open heart surgery. Given the blockage, the risk of doing the Afib
ablation is too significant.
David begins conversing with his Medical Swiss Army Knife app, asking if
his local doctors should have discovered this blockage. The app informs
David that further tests may not have warranted it because he had no reported
symptoms. It also advises him to ask his treating cardiologist and local doctor
this same question when time permits.
Without using the Medical Swiss Army Knife app, David would have
remained solely engaged with his local cardiologist, unaware of his high risk
of a heart attack. The app’s significant benefit was recommending David see
a well-regarded Afib specialist.
David entered what was expected to be a three to four surgery, but instead, it
took six hours. The doctor completes the surgery and tells David’s wife, Ann,
what occurred. He states the reason David’s surgery took six hours instead of
4 or 3 is that he had a physical abnormality, causing blood to go from his
lungs to his heart in a way that the doctor had never seen or anyone else he
knows.
The doctor emphasizes he has been doing this for decades, even working with
babies with congenital heart disease and birth abnormalities, and has never
seen anything like it. It took them time to try to get to the bottom of it, and
instead of using one pump to recycle the blood, they had three of them going
on, which wasn’t enough.
We would be remiss not to mention why Ann had such confidence in the
Medical Swiss Army Knife LLM medical app. She was diagnosed with CLL
leukemia four years earlier. She had an appointment with an oncologist on a
Monday and received a call from her daughter the previous Thursday. Her
daughter was an active user of the Medical Swiss Army Knife app. The app
suggested her mother would receive the best outcome at a cancer research
hospital versus the local hospital she had planned for treatment. Her mother
was not too keen on rescheduling her appointment as she liked her
oncologist, and the local hospital was a short drive away compared to the
research hospital. But she relented, canceled her appointment, and made an
appointment to see an oncologist at the research hospital.
The research hospital had a slightly different treatment plan, which included a
recently available FDA drug, IMBRUVICA®. Ann was quite pleased with
the results and currently finds her cancer in remission. She credits her
daughter and the app directing her to a care facility that produced better CLL
leukemia outcomes. Ann understood that clinical outcomes could differ
drastically based on the provider, and she was delighted that she got her
husband, David, connected to an expert in treating AFib. She firmly believes
it saved her husband’s life. It’s no secret that Medical facilities that released
research findings achieved elevated patient satisfaction scores and exhibited
reduced patient mortality rates across a variety of medical conditions and
procedures.4
By leveraging expansive data on providers’ clinical outcomes, the Medical
Swiss Army Knife app, powered by a large language model, is able to match
individual patients with the physicians statistically poised to provide the most
effective treatment for the patient’s particular condition profile and risk
factors.

Medical Sherpa App

Henry had been a primary care physician for over 20 years and had seen it
all. But when his patient, John, came in for a routine physical checkup, Henry
noticed something that made him pause. John had a small lump on the side of
his throat. “John,” Henry said, “I’d like to take a closer look at that lump on
your throat.” John nodded, and Henry gently tilted his head back. He used his
fingers to palpate the lump and furrowed his brow. The lump was firm and
fixed and didn’t move when Henry pressed on it. “I’m concerned that this
lump may be cancerous,” Henry said. “I’d like to recommend seeing a
specialist for further evaluation.” John looked worried. “But I feel fine,” he
said. “I don’t have any other symptoms.”
“It’s important to remember that cancer can often be asymptomatic in the
early stages,” Henry said. “That’s why it’s so vital to get regular checkups.”
John reluctantly agreed to see a specialist, and Henry scheduled an
appointment for him the following week. But Henry couldn’t shake the
feeling that something was wrong. He decided to consult with his medical
sherpa, an LLM app that could access and process vast medical knowledge.
Henry described the lump to his medical sherpa, and the LLM app quickly
returned with several suggestions. A proposed course of action included
requesting a fine needle aspiration (FNA) biopsy, a minimally invasive
procedure for extracting a cell sample from the lump. An alternative
recommendation was to direct John to an otolaryngologist, a specialist in the
diagnosis and treatment of ear, nose, and throat conditions.
Henry decided to follow the medical sherpa’s advice. He ordered an FNA
biopsy for John, and he also referred him to an otolaryngologist. A few days
later, Henry received the results of the FNA biopsy. The results were positive
for cancer.
Henry called John and told him the news. John was devastated, but he was
also grateful that Henry had caught the cancer early. Henry confirmed with
John if it was okay if Henry’s Medical Sherpa worked with John to schedule
an appointment with the otolaryngologist to discuss his treatment options.
Thanks to Henry’s quick thinking and his use of a medical sherpa, John was
able to get the care he needed to fight his cancer.
Medical Sherpa is a clinician-facing LLM used by doctors seeking
consultation, a fairly common practice in medicine, described as hallway,
elevator, or curbside consultations. A medical sherpa app is a quick and
informal consultation between a clinician and an LLM. The term “sherpa” is
used because, like the guides who help climbers ascend Mount Everest,
medical sherpas help clinicians navigate complex medical terrain.
The sherpa can provide the requesting provider with expert insights,
recommendations, and resources. Examples of typical consultation subjects
include:
Medication interactions: Medical sherpas can help providers identify
and manage potential medication interactions.
Dosage adjustments: Medical sherpas can help providers determine the
correct medication dosage for a particular patient.
Therapeutic options: Medical sherpas can help providers develop and
evaluate treatment plans for patients with complex medical conditions.
Differential diagnoses: Medical sherpas can help providers narrow the
list of possible diagnoses for a patient with a complex presentation.
Test interpretation: Healthcare guides, aka medical sherpas, can assist
healthcare professionals in deciphering the outcomes of laboratory
examinations and medical imaging.
There are several benefits to using medical sherpas. Primarily, they play a
vital role in enhancing clinicians’ quality of care. They achieve this by
offering expert guidance and support to providers, enabling them to arrive at
more well-informed decisions regarding their patients’ healthcare.
Second, medical sherpas can help providers to save time. By providing
providers with a quick and easy way to get advice and support, medical
sherpas can free up providers’ time to focus on other tasks.
Third, medical sherpas can help to reduce provider burnout. Medical sherpas
can help providers feel more confident and supported in their work by
presenting providers with assistance and guidance.
While LLM-powered apps wow users with their intuitive interfaces and
impressive capabilities, it’s crucial to look beyond the surface. The real
magic happens behind the scenes, in the complex architecture that powers
these tools. This architecture is like a well-oiled machine, seamlessly
orchestrating data, algorithms, and computing resources. But just like any
intricate system, designing and building LLM-powered apps comes with its
own set of challenges. Let’s briefly describe this architectural landscape and
explore the key considerations that shape the foundation of these
transformative tools.

Design and Architecture Concerns for LLM-

Powered Apps
The emergence of large language models represents a potent technological
advancement poised to transform healthcare and various aspects of our daily
lives. When creating and implementing LLM-powered applications,
developers must tackle design and architectural considerations.

Performance and Scalability

LLMs can be very computationally expensive because they need to process
large amounts of data with various modalities to generate accurate and
informative outputs in a medical setting. As a result, it is essential to design
LLM-powered apps in an efficient and scalable way. Several design choices
and techniques achieve this outcome:
 Use a cloud computing platform to provide scalable infrastructure for
your app and a platform that facilitates the building of search and
conversation-based LLM AI apps.
Use a cloud computing platform with built-in capabilities for creating
generative AI-powered agents.
LLMs can be computationally expensive, and ensuring smooth performance
for a large user base requires careful optimization of resources. This may
involve distributed computing systems, specialized hardware, and efficient
data pipelines. Moreover, as LLMs continue to evolve and learn, the
architecture needs to be adaptable and able to integrate new functionalities
seamlessly.

Security and Privacy

It is essential to design LLM-powered apps in a way that protects user data
and prevents unauthorized access to the LLM. It is essential to design LLM-
powered apps in a way that protects user data and prevents unauthorized
access to the LLM. Data leaks and malicious manipulation of these powerful
language models could have severe consequences, ranging from personal
privacy breaches to the spread of misinformation.
Therefore, security architecture needs to be embedded from the ground up.
This includes robust encryption for user data, layered access controls for the
LLM itself, and continuous monitoring for potential vulnerabilities.
Additionally, transparency in data handling practices and responsible training
procedures are crucial to building trust with users and ensuring the ethical use
of these technologies.
Ultimately, the success of LLM-powered apps lies in the intersection of
powerful technology, thoughtful design, and responsible use. By carefully
navigating the architectural landscape, developers can create transformative
tools that empower users while safeguarding their data and the integrity of
these intelligent systems.
Ethics and Bias
LLMs can be biased, and their outputs can be used to harm or offend people.
It is, therefore, essential to design LLM-powered apps to mitigate bias and
prevent the LLM from being used for harmful purposes. This requires a
multi-pronged approach. Firstly, data selection and curation are crucial.
Biases often creep in through the training data, so careful selection and
filtering are essential to minimize their presence. Techniques like debiasing
algorithms and diversifying data sources can help address this.
Secondly, algorithmic fairness needs to be prioritized. Design choices during
the development of the LLM itself can impact the fairness of its outputs.
Implementing diverse evaluation metrics, incorporating techniques like
counterfactual reasoning, and regularly auditing the LLM’s behavior for bias
are crucial steps.
Finally, user-facing safeguards are necessary. Transparency about how the
LLM works and potential limitations in its abilities are essential. Building in
features that allow users to flag or report biased outputs, along with providing
channels for feedback and redressal, are key measures in empowering users
and preventing harm.
Tackling bias and potential misuse requires ongoing vigilance and iterative
improvement. Developers must actively engage with diverse stakeholders,
researchers, and ethicists to refine their approach and ensure that LLM-
powered apps are designed to serve not just as technical marvels but as tools
for social good.
By building secure, efficient, and responsible architectures, we can harness
the power of LLMs to create a future where these intelligent systems not only
augment our capabilities but also promote inclusivity, safety, and responsible
AI practices.

Summary
The road ahead for LLM-powered applications is paved with immense
potential and significant challenges. We stand at the precipice of a future
where these intelligent systems can solve complex problems, unlock new
creative avenues, and reshape our interactions with the world. Yet, navigating
the technical hurdles of performance, stability, and security is just the first
step.
Beyond the technological infrastructure lies the human dimension. Privacy
concerns cast a shadow over data-hungry LLMs, demanding transparency
and ethical stewardship of user information. Biases woven into the fabric of
training data threaten to perpetuate inequalities and injustices, necessitating
constant vigilance and proactive mitigation strategies.
As we embark on this journey, we must remember that technology alone
cannot build a brighter future. Responsible design, informed by diverse
perspectives and ethical considerations, is the key to unlocking the true
potential of LLMs. By carefully addressing the architectural concerns and
prioritizing inclusivity, transparency, and fairness, we can ensure that these
intelligent systems serve as tools for human empowerment and progress, not
instruments of harm or exclusion.

Enterprise Architecture for LLM Medical Apps

Figure 1-1 illustrates an architecture for realizing the Medical Swiss Army
Knife and Medical Sherpa Apps. We use Google Cloud products to provide
an implementable architecture in this example. It is essential to build large
language models and generative AI applications using a cloud platform for
several reasons:
Scalability: Cloud platforms offer the ability to scale your application up
or down as needed. This is important for LLM and generative AI
applications, which can be computationally expensive.
Reliability: Cloud platforms are designed to be highly reliable. Your
application will be less likely to experience downtime or performance
issues.
Security: Cloud platforms offer a variety of security features that can
 help to protect your application from cyberattacks.
Cost: Cloud platforms can be more cost-effective than building and
maintaining your own infrastructure. This is because cloud platforms
allow you to pay for the resources you use as needed.

Figure 1-1. : LLM App Enterprise Architecture

In addition to these general benefits, cloud platforms also offer several

features that are specifically designed for building and deploying LLM and
generative AI applications. For example, cloud platforms offer:
Pre-trained LLMs: Many cloud platforms offer pre-trained LLMs that
you can use in your applications. This can save you significant time and
 effort, as you do not need to train your own LLM.
Specialized hardware: Some cloud platforms offer specialized hardware
for running LLM and generative AI applications. This hardware can
significantly improve the performance and efficiency of your
applications.
Managed services: Many cloud platforms offer managed services for
running LLM and generative AI applications. This means you do not
need to worry about managing the underlying infrastructure.

Here are some specific examples of how cloud platforms can be used to build
and deploy LLM and generative AI applications. Google, Amazon, and
Microsoft, three major cloud providers, offer a variety of services for
building and deploying LLM and generative applications:
Google Cloud Platform: Google Cloud Platform (GCP):
Vertex AI: A managed service for building and deploying machine
learning models, including LLMs.
Cloud TPUs: Specialized hardware for running machine learning
models which can significantly improve the performance and
efficiency of LLM and generative AI applications.
AutoML: A set of tools that make it easy to build machine learning
models even without expertise in machine learning.
Amazon Web Services (AWS):
SageMaker: A managed service for building and deploying
machine learning models.
EC2 F1 instances: Specialized hardware for running machine
learning models.
SageMaker Canvas: A drag-and-drop tool for building machine
learning models without writing code.
 Microsoft Azure: Azure:
Machine Learning Studio: Another drag-and-drop tool for building
machine learning models without writing code.
Azure Databricks: A managed service for running Apache Spark
and Apache Hadoop workloads.
Azure Batch: A managed service for running large-scale batch
processing jobs.

These are just a few examples of how cloud platforms can build and deploy
LLM and generative AI applications. Cloud platforms offer a wide range of
features and services that can help you build and deploy powerful LLM and
generative AI applications to production.
A fundamental component in this enterprise architecture (Figure 1-1) is
Google Cloud’s Adapter Layers, part of Vertex AI. These layers are
strategically positioned within the customer’s Google Cloud (GCP) tenant to
effectively segregate the customer’s data, preventing it from being utilized in
the training of the Large Base Model.
This architectural design is particularly noteworthy because these adapter
layers are intricately integrated into your GCP tenant, establishing a close
connection with the centrally hosted Large Base Model by Google Cloud.
The adapter layer is pivotal when integrating customer-specific data and fine-
tuning your model. This segmentation ensures that a customer’s data remains
isolated from the Large Base Model’s training, guaranteeing data privacy and
security. Remarkably, this design doesn’t compromise the advantages of
leveraging the expansive central models, as it maintains a robust data
interface for optimal performance.
Moreover, besides the adapter layers, this enterprise architecture offers the
distinct advantage of harnessing conversational AI and enterprise search
capabilities directly within the GCP tenant. This empowers you to retrieve
information from a spectrum of sources, including private corporate
databases and public websites while maintaining your enterprise’s stringent
security protocols.
Another feature here is the ability to perform database embeddings, enabling
the generation of precise and factual answers directly from your database,
seamlessly integrated within the Large Language Model. This approach
allows one to engage with your content in a fresh and secure manner.
This architecture leverages an organization’s security measures and
infrastructure within the GCP tenant, seamlessly integrating data with
cutting-edge Generative AI (Gen AI) capabilities. It ensures that your data
remains secure, enabling you to optimize your data interaction and retrieval
processes effectively. It’s noteworthy that alternative cloud data can be used
with the LLMs. It can also be used for apps designed as consumer or business
LLMs.

Consumer and Business LLMs

Consumer and business-oriented Large Language Models are designed to
cater to their respective audiences’ distinct needs and use cases. Let’s dive
into a brief overview of consumer and business LLMs.

Consumer LLMs and Generative AI

We design consumer LLMs/generative AI to be accessible to the general
public. In the context of healthcare, consumer Patients use LLMs/generative
AI to:

Find and compare healthcare providers,

Get advice on how to stay healthy
Manage their health records
Connect with other patients and caregivers
Consumer LLMs are designed for individual users, offering various
applications and functionalities tailored to personal needs and interests. These
LLMs include models like chatbots, virtual assistants, and content generators.
Key characteristics of consumer LLMs include
 Conversational Assistants: Consumer LLMs like virtual assistants (e.g.,
Siri, Google Assistant) are developed to assist users in setting reminders,
answering general knowledge questions, sending messages, and playing
music. They are designed for everyday convenience.
Content Generation: Some consumer LLMs can generate text, which can
be helpful for tasks like drafting emails, writing creative content, or even
coding assistance. These models focus on enhancing personal
productivity and creativity.
Personalization: Consumer LLMs often prioritize personalization,
learning from user interactions to provide tailored recommendations,
content, and responses.
Accessibility: These models are designed with user-friendly interfaces,
accessible to a broad range of users, and are often available on mobile
devices and personal computers.

Business LLMs and Generative AI

Businesses and organizations design business LLMs and generative AI for
their use. They often use them to automate tasks, provide insights into data,
and generate creative content such as:
Improve clinical decision-making
Develop new drugs and therapies
Improve the efficiency of clinical trials
Improve health equity
Improve operations such as customer service
Business-focused LLMs are engineered to address the specific needs of
organizations and enterprises. They are equipped to handle more complex,
industry-specific tasks and processes. Key characteristics of business LLMs
include:
 Data Integration: Business LLMs are designed to integrate with an
organization’s data sources, such as Electronic Health Records ( EHRs),
clinical data, claims, pharmacy, or eligibility databases. They can offer
insights and generate reports based on this integrated healthcare data.
Industry-Specific Use Cases: Business LLMs are tailored to serve
specific industries, such as healthcare. They can provide domain-
specific solutions like medical diagnosis, claims processing, or clinical
decision-making.
Collaboration and Productivity: These LLMs often include features that
enable team collaboration, document sharing, and workflow automation
to improve organizational productivity.
Data Security and Compliance: Business LLMs prioritize data security,
privacy, and regulatory compliance. They often provide features to
ensure that sensitive business data remains secure.
Enterprise-Level Support: Business LLMs come with dedicated
customer support and service-level agreements, ensuring the reliability
and consistency of their services for businesses.
In summary, consumer and business LLMs serve distinct purposes, with
consumer LLMs focusing on individual convenience and personal
productivity. In contrast, business LLMs are designed to meet the specific
needs of organizations and industries, offering tailored solutions, data
integration, and robust support for enterprise-level operations.

Why the difference is significant:

The difference between consumer and business LLMs/generative AI is
crucial because it determines usage and intended audience. We design
consumer LLMs/generative AI to be accessible to the general public and to
provide them with information and resources to manage their health.
Another critical difference between consumer and business LLMs/generative
AI is the level of regulation. Consumer LLMs/generative AI are generally
less regulated than business LLMs/generative AI. This is because consumer
LLMs/generative AI are not used to make medical decisions or to provide
medical advice. Business LLMs/generative AI, on the other hand, are used in
more sensitive settings where the accuracy and reliability of the information
are critical.
It’s important to distinguish between business and consumer LLMs for
several reasons:
Purpose and goals
Data training
Security and Privacy
Regulatory Landscape
Ethics and Bias
Purposes and goals:
Business LLMs: Primarily focused on solving specific business
problems or improving business processes. Examples include
automating health plan member customer service interactions or
providing insights from business data.
Consumer LLMs: Primarily focused on personal use and education.
Examples include translating languages or answering questions in an
informative way.
Data training:
Business LLMs: Trained on specialized datasets relevant to the specific
business domain. This ensures the LLM understands the business
context and jargon, leading to more accurate and relevant outputs.
Consumer LLMs: Trained on massive, general-purpose datasets
containing text and code from various sources. This allows for
versatility and broad knowledge, but may also introduce biases or lack
specific domain expertise.
Security and Privacy:
Business LLMs: Often involve handling sensitive business data,
necessitating strict security measures to protect against unauthorized
access or data breaches.
Consumer LLMs: May or may not handle sensitive data, but privacy
considerations are still important to ensure user trust and protect
personal information.
Regulatory Landscape:

Business LLMs: Subject to specific regulations depending on the

industry and data handled, such as financial sector regulations or
healthcare data privacy laws.
Consumer LLMs: Subject to consumer protection laws and regulations
related to data privacy and ethical AI practices.
Ethics and Bias:
Business LLMs: Responsible use and mitigation of bias is crucial to
avoid discrimination or unfair treatment in business decisions or
customer interactions.
Consumer LLMs: Bias in consumer LLMs can lead to harmful
misinformation, offensive content, or perpetuating social inequalities.
Responsible development and ongoing efforts to address bias are
critical.
In conclusion, while both business and consumer LLMs are large language
models, their distinct purposes, data requirements, security needs, regulatory
contexts, and ethical considerations warrant differentiation and tailored
approaches to development and implementation.

Challenges of LLMs and Generative AI in

Healthcare
Random documents with unrelated
content Scribd suggests to you:
 “All right. Any more?”
“No, but I think that’s enough to go on with. The probability I’m
going to bet on is murder.”
“You’re up against coincidence again, though, there. Why should
somebody happen to murder Brotherhood on the very day he went
bankrupt?”
“You will go on assuming that it is Brotherhood. Supposing, just
for the sake of argument, that Brotherhood has saved a nest-egg for
himself, and is skipping to avoid his creditors—what better way of
throwing people off the scent than by a pretended suicide?”
“That is, by pitching a total stranger down the viaduct.”
“I didn’t say a total stranger. Suppose it were somebody in pursuit
of him, or somebody he suspected of pursuing him?”
“But he couldn’t be sure that the face would get mangled like that.
It was only one chance in a thousand that the body should scrape
down all along that buttress on its face.”
“He may simply have wanted to kill the man, without hoping that
the corpse would be mistaken for him. After all, we’ve got to explain
the ticket; a man who takes a single ticket down here is almost
certainly not a resident here—the half-fare is so cheap. A spy,
tracking him, or somebody he takes to be a spy tracking him. He
stuns the man while he’s not looking, and then pitches him out. He’s
desperate, remember.”
“Well, it seems to hang together that way.”
“But I’m not at all sure that’s the right way. I’m not at all sure that
Brotherhood isn’t the murderee, and the murderer somebody
unknown—such a murder might be connected with a bankruptcy, a
ruined creditor, for example.”
“And how are you going to look for the murderer if that’s so?”
“You’re going to help me. We’re going to have a little detective
holiday, and leave the game alone for a bit. Of course we must find
out all about Brotherhood first—it’s extraordinary how little people
seem to know about him. I asked four men in the Club whether he
wore a wrist-watch or not: two couldn’t remember, one said he did,
and one swore he didn’t. But there must be some servant who looks
after his bungalow for him; so I’m going there to-morrow to pump
them.”
“Introducing yourself as Mr. S. Holmes of Baker Street, or how?”
“No, I shall be the Daily Mail reporter—unless I run into the real
article on the mat. Now, would you mind following up the
Masterman clue?”
“What Masterman clue?”
“There are only two Mastermans in the Telephone Directory. A
man dressed like that would be sure to have a telephone.”
“But I thought you’d made up your mind it wasn’t a local person
at all, because of the ticket?”
“I know, it’s probably a wild-goose chase, but it’s the best we can
do on that tack. Both are at Binver; one’s a solicitor and one a
doctor. I’ll give you the addresses.”
“And I’m to go to them and ask them what kind of handkerchiefs
they use? Or should I meet them accidentally and say, Excuse me,
sir, could you lend me a handkerchief, I’ve left mine at home?”
“Well, you can find out whether they’re dead, anyhow.”
“And if they’re still alive?”
“Well, scout around somehow. Do anything that occurs to you.
This business ought to be rather fun, if we exercise a little ingenuity.”
“Meanwhile, let’s have another look at those documents. We don’t
seem to have made much out of them, and that’s a fact.”
They sat for several minutes in silence, re-reading the copy
Reeves had made of the anonymous letter. It was undated; the
address was in printed capitals; it had been post-marked in London
at starting, and at Paston Whitchurch on arrival. The content of the
message was a mere series of numbers, as follows:

8 7 5
18 4 7
21 2 3
25 6 4
31 4 8
74 13 9
 92 29 7
97 5 3
113 17 13
10 12 13

“Unless they’re sums of money,” said Gordon, “I can’t make head

or tail of it all. And if they were sums of money, it would be a queer
way to arrange the spacing.”
“Wait one moment,” said Reeves, “I believe I’ve got the idea of it.”
He put his hand to his forehead. “Yes, that does it. It’s a cipher, of
course, otherwise there’d be something to explain what it’s all about.
It will be a book cipher; the first figure gives you the page, the
second the line, and the third the word in the line. How’s that?”
“That’s devilish ingenious,” admitted Gordon, “but you can hardly
prove it.”
“I can practically prove it,” said Reeves. “Look here, the man
wanted to spell out a message in ten words. There was a book,
arranged upon somehow beforehand. The first few words were
ordinary words, that you could find anywhere on any page: and
naturally, to save himself and the other man trouble in counting, he
took them from the top of the page, so you get lines 7, 4, 2, 6, and
4 of pages 8, 18, 21, 25 and 31. The sixth word he wanted was an
obscure sort of word, perhaps even a proper name. He had to go
right on to page 74, and even then he could only find his word on
the 13th line of it. Then the next two words came easy,
comparatively, but the ninth word was a brute, he couldn’t find it till
page 113, and on the 17th line at that. And by that time he’d got
nearly to the end of the book—a book, then, of only 120 pages or so
probably; a paper edition, I suspect—so he had to go back to the
beginning again, which he hadn’t meant to do.”
“Bravo!” said Gordon. “Have another injection of cocaine.”
“The curse of the thing is,” said Mordaunt Reeves, “that with a
book cipher you can’t possibly guess the message unless you’ve got
the book. I think we shall have to establish the identity before we
get any further on that tack. Let’s have a look at the letter now.”
 The letter was a curt official communication from the Railway
Company, only the details being filled in in ink, the rest a mere
printed form:

London Midland and Scottish Railway.

10. 10. 19XY.
Dear Sir,
I beg to acknowledge receipt of your letter of 9th instant and have
given orders for a berth to be reserved in the sleeping car attached
to the 7.30 train on Thursday (corrected to Wednesday) the 18th
(corrected to 17th) of October to Glasgow. I note that you will join
the train at Crewe.
S. Brotherhood Esq.

“These corrections are rummy,” said Reeves. “I wonder if perhaps

Brotherhood’s letter corrected itself in a postscript? You see,
assuming that Brotherhood was skipping, it’s all right for him to go
to Glasgow—rather ingenious, in fact—but why shouldn’t he travel
to-night, the sixteenth, instead of to-morrow night?”
“He couldn’t get away early enough. Or could he? Got a
Bradshaw?” Gordon proceeded to look up the trains with an irritating
thoroughness, while Reeves danced with impatience—there is no
impatience like that engendered by watching another man look up
Bradshaw. “That’s all right,” said Gordon at last. “In order to catch
the Scottish train at Crewe he’d have had to take that earlier train,
the one Marryatt came up by, and get out at Binver. He took the
3.47, I suppose, because he couldn’t get away sooner. Perhaps, if
we’re right in thinking he wanted to skip, he was going to go across
country by car to-morrow and confuse his tracks a bit.”
“The thing doesn’t look like skipping quite as much as it did. For
Heaven’s sake let’s beware of prejudicing the case. Anyhow, he
meant to make for Glasgow on the Wednesday night—that’s to-
morrow night, isn’t it? Now let’s have one more look at that silly list
that was on the back of the anonymous letter.”
The list had been copied almost in facsimile, for it was very short.
It ran
 Socks
vest
hem
tins—

at least, that was the general impression it gave, but the writing was
so spidery as to make it very doubtful which precise letter each of
the strokes represented.
“I suppose it must be a shopping-list of some sort. If one could
make that last word ‘ties’ it would read better,” said Gordon.
“But even so you wouldn’t have hems in a shopping-list.”
“It might be ham.”
“But one doesn’t buy ham at the hosier’s.”
“And why did he write at the edge of the paper like that?”
“If it comes to that, who was the he? It’s not Brotherhood’s
writing—I’ve verified that from the club book. I fancy this goes
pretty deep. Look here, here’s a bit of detection for you. That sheet
has been torn off at the left-hand side, hasn’t it? Now, was it torn off
before or after the writing was put on it?”
“Before, surely. Otherwise the initial letters wouldn’t be so
complete; he’d have been certain to tear across them.”
“I’m not so sure. Who writes so close to the edge of a piece of
paper as that? Remember, I copied the thing down exactly, and each
word was close up against the tear.”
“I don’t quite see what difference it makes, anyhow,” objected
Gordon.
“More than you think, perhaps. I shouldn’t wonder if this bit of
paper turned up trumps, when we’ve thought it over a bit more. But
there’s one thing that fairly beats me.”
“What’s that?”
“Those two watches. It doesn’t seem to me to make any sense.
Well, we’d better get to bed and sleep over it.”
 Chapter IV.
Endless Clues
There is no surer soporific than sleeping over a problem, no more
fallacious method of attempting a solution. After murmuring to
himself three times, “Let’s see; there was something about watches,”
Mordaunt Reeves fell into a sleep which anybody but a
psychoanalyst would have called dreamless. He woke in the morning
with a strong resolution to do the ninth in four, which melted
through lazy stages of half-awareness into the feeling that there was
something else to do first. The adventures of yesterday, the duties of
to-day, returned to him. He was already nearly dressed when he
remembered that he had decided on the rôle of a Daily Mail reporter
for his morning’s investigation, and grimly set himself to remove
again the bulging knickerbockers and the hypocritical garters of his
kind. Dressy they might be, but they were not Fleet Street. His
memories of the reporter’s wardrobe were, it must be confessed,
somewhat disordered, and he was greeted in the breakfast-room
with flippant inquiries whether he had gone into mourning for the
Unknown Passenger.
He found Gordon already at table with Marryatt—Marryatt in the
high clerical collar which was irreverently known to his intimates as
“New every morning.”
“Well, how are you feeling?” he asked. “You looked rather chippy
yesterday. However, I suppose it brings a job of work your way.”
“Confound it,” said Marryatt, “that’s the trouble. The jury at the
inquest are bound to bring in suicide; and then I can’t bury the man
in the churchyard, and all the villagers will say I refused out of spite,
because the poor old chap used to give these atheist lectures on the
village green.”
“Rot!” said Gordon; “if they do find suicide, they’ll certainly say he
was of unsound mind.”
“Yes,” echoed Reeves, “if they do bring in suicide.”
“But surely you can’t doubt it,” urged Marryatt energetically. “The
man’s just gone bankrupt, and it was an ugly case, from what I
hear; several innocent people who’d been fools enough to believe in
him left in the cart. At the same time, the smash came very
suddenly, and that makes it unlikely that anybody could want to
murder the man so soon. Oh, you’ll find it’s suicide right enough.”
“Well,” said Reeves a little stiffly, “we’re going to do our best to
find out between us. I’ve the greatest respect for the police as a
body, but I don’t think they’re very good at following up clues. When
I was in the Military Intelligence one was constantly putting material
at the disposal of the police which they were too supine or too
stupid to use.”
“Well, good luck to your sleuthing; but mark my words, you’ll find
it was suicide. I’m going to play a round now to try and take my
mind off the thing, but I don’t believe I shall be able to drive at the
third after—after what we saw yesterday.”
Left to themselves, Mordaunt Reeves and Gordon arranged that
they would meet again at luncheon and report on the morning’s
investigations.
“And look here,” said Reeves, “it’s a belief of mine that one wants
to cover the ground oneself if one’s to visualize the setting of a
crime properly. So I vote that after lunch we stroll down to the
railway and take a look at the top of that viaduct, and then take the
4.50 from Paston Oatvile to Paston Whitchurch so as to picture the
whole thing exactly as it happened.” And so they parted, Reeves
walking to Brotherhood’s bungalow, close to Paston Whitchurch
station, while Gordon mounted a motor-bicycle and set out for
Binver, a sleepy market town of some importance as a railway
junction, about twelve miles off.
Mr. Brotherhood’s housekeeper, Mrs. Bramston, had something of
the airs of a landlady. She spoke painfully correct English, far more
terrible than the native cockney which it half revealed and half
concealed. She commenced where others began, closed doors
where others shut them, and recollected instead of remembering.
Her final consonants were all sibilant, and seemed to form part of
the succeeding word. She was a merciless and largely irrelevant
talker, and the opportunity of a stranger’s visit delighted her, self-
importance easily triumphing over any regret she may have felt for
the apparently deceased. She had no doubt that Reeves was a
reporter, but it is probable that she would have opened out quite as
readily if he had announced himself as the piano-tuner.
“From the Daily Mail? To be sure, sir. I’m always fond of looking at
a paper myself, and as for the Daily Telegraph, I simply revel in it.
Called about poor Mr. Brotherood, I suppose; well, there isn’t much
doubt what’s come to him, poor soul. . . . Not Mr. Brotherood at all?
Don’t you delude yourself, young man; that’s him, sure enough. The
police, they wanted me to go and look at the corpse; but I didn’t
hardly like to; battered they say it was, something shocking. His
clothes? Of course they were his clothes; you don’t think he’d want
to be putting somebody else’s clothes on to commit suicide in, do
you? That’s the same as he always wore; plain black coat and grey
striped trousers, just the same as it was in the papers. . . . What
tailor he went to? No, I couldn’t rightly say that; though I’ve had the
folding of them many a time; very neat man he was, Mr. Brotherood,
in his personal habits. Oh, I dare say there’s others as have clothes
like his, only you see the way I look at it is, if the clothes were on
Mr. Brotherood, then it’s Mr. Brotherood’s clothes they’ll be, that’s
the way I look at it.
“A single gentleman? Yes, a single gentleman he was, single and
singular, if you’ll pardon the jeu de mots. Very singular in his habits.
Every Saturday off he’d go, just the same as it was in the papers,
and where he went to is more than I can say, though I’ve been
looking after him the best part of a year now. Every afternoon from
Monday to Saturday he’d come home by the five o’clock train, and
then he’d go for his round of golf, and I’d have a bit of cold supper
ready for him when he came home. . . .
 “No, I can’t say that I’ve noticed anything strange about him of
late. You see, he was always a very reserved gentleman, Mr.
Brotherood was; very silent, if you understand what I mean, in
conversation.” (Reeves felt that this was probably a characteristic
common to most of Mrs. Bramston’s interlocutors.) “Time and again
he’s said to me would I mind leaving him now because he’d got a
great deal to do. I recollect about a fortnight ago he did seem rather
put out about not being able to find his overcoat when he went out
to deliver his address to the villagers; but I found it for him. . . . No,
it isn’t much more than two months ago since he commenced
exhorting. I never could see what he did it for; not that I go to
church myself, but you see the way I look at it is if people want to
go to church why not let them go to church? Live and let live, that’s
what I say. I shouldn’t call myself a religious woman, mind you, but
I like to see everyone go their own way, and not leave tracts. Miss
Frobisher she used to come here with tracts, but I said to her, ‘Miss
Frobisher,’ I said, ‘you’re wasting your time leaving tracts here,’ and
so she was. . . .
“Mad, sir? Oh dear no, not what you could call mad. Of course we
all have our own little ways, haven’t we, sir? and as I was telling
you, Mr. Brotherood was singular, but not demented; I should never
have stopped with Mr. Brotherood had he been demented. . . .
Suicide? Of course it was suicide; and there’s some say Mr. Marryatt
won’t bury him in holy ground, don’t they? Well, you take my word
for it, Mr. Brotherood wouldn’t mind about a little thing like that.
Some people seem not to mind what happens to them once they’re
gone: Mr. Bramston was like that, while he was spared to me; never
seemed to mind if we were to take a spade and bury him in the back
garden, that’s the way he looked at it. But of course, I wouldn’t have
that, and he was buried properly in holy ground, Mr. Bramston was,
and the minister recited the service over him beautiful. . . . What,
must you be going already, sir? Well, I’m sure it’s been a great
privilege to me to afford you information. Good morning, sir.”
This is an abridged account of the interview, but it contains all the
material disclosures made by Mrs. Bramston. Reeves found himself
pitying the coroner who would have to face and to stem that
seething torrent of conversation. He came back to the dormy-house
to find that it was already nearly time for luncheon, and Gordon was
waiting for him, returned from his errand at Binver.
“Well, have you found out anything?” asked Gordon.
“Yes,” said Reeves, “I’ve found a wife for Carmichael. I’ve found a
woman who could give him a stroke a hole at back-chat.” And he
launched into a description of Mrs. Bramston’s voluminous utterance
and her insignificant contribution to the solving of the mystery.
“Had you any better luck?” he went on.
“Acting upon instructions received, I proceeded first of all to the
offices of Messrs. Masterman, Formby and Jarrold, Solicitors. It’s one
of those jolly old Queen Anne houses facing on the High Street; with
a flagged walk up to the front door and blue gates that need
painting—or rather, it would spoil them if you did. It’s been turned
into an office, and the inside is all musty and smells of decaying
paper. The mustiest thing there was the old clerk I went up to and
asked if I could see Mr. Masterman. And he said, ‘I’m afraid not, sir;
Mr. Masterman is dead.’ ”
“Dead? How? When?”
“My very words. And the old gentleman said, ‘About twenty-three
years ago. Would you like to see Mr. Jarrold?’ Well, that did me in
rather, because even if old Masterman did bequeath his
handkerchiefs to Jarrold, it isn’t likely that old Jarrold would be still
using them, though they would about match his furniture if he did.”
“How did you get out of it? You were rather badly placed.”
“I was, and I cursed you pretty freely. However, I extricated
myself without any heart-to-heart talks with Mr. Jarrold. I just said,
‘I’m so sorry, I must have made some mistake; this is Doctor
Masterman’s house, isn’t it?’ That killed two birds with one stone, I
eluded suspicion and also got directed to the other Masterman
house, a big house, the man said, at the other end of the water-
meadow behind the church.”
“So you went on there?”
“No; it occurred to me that a man who lived in a house that size
probably kept a man-servant or two, and it was up to me to
personate one of them. So I went round to the Binver Steam
Laundry, where I’m not known personally; and said I was from Dr.
Masterman’s, and could they be kind enough to inform Dr.
Masterman as to what action they intended taking about the twelve
last handkerchiefs that hadn’t come back from the wash. That
sounds risky, but it wasn’t really, because all men think they’ve more
clothes at the wash than they really have. The lady in charge was
quite patient and kind, obviously well accustomed to that sort of
complaint; she said all Dr. Masterman’s handkerchiefs had been sent
back. Fortunately I bluffed, and insisted upon a search; after a bit
she came and put into my hands a pile of handkerchiefs, which I
took away with me. There were five of them, four Mastermans and a
Brotherhood.”
“Oh! That rather looks as if——”
“Exactly; it looks as if we ought to have recognized the touch of
the Binver Steam Laundry. In fact, it would be very suspicious in
these parts if you found a dead man wearing one of his own
handkerchiefs. Well, there seemed no point in keeping any of them,
so I dropped the lot into Masterman’s letter-box. Unusual, perhaps,
but I felt it would save explanations.”
“Well, I’m sure we’re all very grateful to Mr. Gordon for his
splendid work among the Mastermen. But it begins to look as if we
were left very much where we were. We still don’t even know who
the corpse was.”
There was a knock at the door, and the unwelcome figure of
Carmichael obtruded itself. “Sorry if I interrupt,” he said, “but I
thought you might be interested in this poor fellow we found
yesterday. My caddie this morning was giving me the latest news.
It’s extraordinary how these caddies pick up everything except one’s
ball.”
“What news?” gasped Reeves.
“Well, it seems that Brotherhood was insured at one of these
American offices. And they’re a great deal more particular than our
own Insurance people. And after all they’re right to be: one’s so apt
to think of the Insurance Company as a set of sharks, when in reality
they are only protecting the interests of their policy-holders.”
“Granted,” said Gordon. “Proceed.”
 “Well, as soon as they heard of the bankruptcy and then saw the
news in the morning paper about the Links Tragedy, the Insurance
Company pricked up its ears. Apparently, in the actuarial world,
bankruptcy followed by alleged suicide is a matter of daily
occurrence, and they have their suspicions. That is why I say they
are quite within their rights when they insist upon registering a man
by his birth-marks before they insure him. It’s an extraordinary thing
about birth-marks; we really know nothing about them——”
“Nor want to,” said Reeves, “for the time being. What happened?”
“I was just telling you. A man came down from the Insurance
Company to identify the corpse; and my caddie heard about it from
——”
“Heard what?”
“Why, that it is Brotherhood. They recognized him from the birth-
mark.”
“So that’s that,” said Mordaunt Reeves, a little bitterly. “Trust the
Insurance people not to make a mistake. I confess that, after the
handkerchief clue failed, I had begun to think it must be
Brotherhood who was dead. I suppose your caddie didn’t happen to
mention whether it was suicide or murder?”
“He assumed it to be suicide; but not, I think, with any inside
information. Of course, it was a foggy day. Did you know that, as a
matter of statistics, there are more suicides in November than in any
other month?”
“I will make a note of the fact,” said Mordaunt Reeves.
 Chapter V.
On the Railway
The afternoon seemed a compensation for yesterday; October sun
glowed temperately over the links, with the air of a kind old
gentleman producing sweetmeats unexpectedly. The rich but
transient gold of summer evenings seemed hoarded in this summer
of St. Luke; the air not over-charged with uneasy heat, but lucid and
caressing; the leaves no longer in the shock of their summer finery,
but dignified in the decayed gentility of their autumn gold. A perfect
day for golf, such was the immediate impression of the Paston
Oatvile mind; but to Reeves a second thought occurred—it was a
bad day for following up the clues of a murder.
“It’s all very well,” he said to Gordon, “the visibility’s good, and we
shan’t be interrupted by rain; but we can’t get the atmosphere; the
spiritual atmosphere, I mean, of yesterday’s fog and drizzle. We shall
see where a man fell down the embankment, but we shan’t feel the
impulse of that weeping depression which made him throw himself
over, or made somebody else save him the trouble. We haven’t got
the mise-en-scène of a tragedy.”
They climbed together, Gordon and he; a zigzag path up the side
of the huge embankment, close to the club-house. When it reached
the level of the line, it kept close to the trim hedge that marked the
boundary of the railway’s property, and so lasted till the very
beginning of the viaduct, where it dived under the first arch at a
precarious angle and came up the other side. It was a matter of
common knowledge to the good-humoured porters of Paston Oatvile
that the shortest way from that station to the neighbouring station
of Paston Whitchurch was along the railway line itself—the shortest,
because it avoided the steep dip into the valley. Accordingly, it was
the habit of residents, if pressed for time, to follow this path up to
the viaduct, then to break over the sacred hedge and walk over the
railway bridge till a similar path was available on the Paston
Whitchurch side. This local habit Reeves and Gordon now naturally
followed, for it gave them access to the very spot from which,
twenty-four hours before, a human body had been hurled down on
to the granite buttress and the osier-bed that lay beneath.
“You see what I mean,” said Reeves. “We can’t, of course, tell
what pace the train was going; they vary so much in the fog. But if,
for the sake of argument, you take the force with which I throw this
stone as the impetus of the train, you see how the curve of the
slope edges it out to the right—there—and it falls either exactly on
the buttress or next door to it. That’s how I picture yesterday
afternoon—the man takes a good jump—or gets a good shove, and
falls just over the edge; there’s nothing for him to catch on to; and
between his own motion and the slope of the embankment he gets
pitched on to the buttress. I don’t know any place along this line
where the drop comes so close. The coroner will call attention to
that—it’s extraordinary the way coroners do draw attention to all the
least important aspects of the case. I read a newspaper account
once of a man who was killed by a motor-car just as he came out of
church, and I’m blessed if the coroner didn’t draw attention to the
dangerous habit of standing about outside churches.”
“I must say, the place seems made for something like this
happening. Do you see how the line curves away from this side?”
“Why shouldn’t it?”
“What I mean is, it would be very hard for anybody to see
Brotherhood fall out of the train unless he was travelling in the same
coach: the other coaches would be out of view (unless a man were
leaning right out of the window), simply owing to the curve—and of
course a fog would make the job all the easier.”
“By Jove, that’s true. I must say, I stick to my murder theory,
whatever the jury make of it. In fact, I hope they will bring in
suicide, because then the police won’t be fussing round all over the
place. It looks to me like a murder, and a carefully planned one.”
“I’d just like to try your stone-throwing trick once more. Look
here, I’ll lean over the edge and watch it fall. Only we shall want a
bigger stone, if you can find one.”
“All right. Only they’re all little ones between the sleepers. I’ll look
along the bank a bit. I say, what the devil’s this?”
It was a sight that on most days would have given little surprise to
the pair; a common enough sight, indeed, down in the valley, but up
here a portent. Caught in a clump of grass, some twenty yards down
the line in the Paston Oatvile direction, was a golf-ball.
“That beats everything,” declared Gordon. “I don’t believe
Carmichael on his worst day could slice a ball a hundred feet up in
the air and lodge it in that clump.”
Reeves was examining the trove intently. “I don’t like this a bit,”
he said. “This is practically a new ball, not the sort of ball a man
would throw away casually as he walked down the line. A Buffalo, I
see—dash it all, there are at least a dozen of us use those. Who’ll
tell us whether Brotherhood used them?”
“I say, steady on! You’ve got this murder business on the brain.
How can you tell the ball hasn’t been there weeks and weeks?”
“Very simply, because it happens to have snapped the stalk of this
flower—scabious, don’t they call ’em—which isn’t dead yet. The ball
was right on top when I found it. I’m hanged if that ball fell there
more than twenty-four hours ago.”
“I say, we ought to be getting back to Oatvile if we’re going to
catch that train,” said Gordon. “It’s half-past four already, and we’ve
got to take to the path before we come in sight of the signal-box.
The signalman doesn’t really mind, but he has to pretend to.”
Gordon was one of those men who are always too early for trains.
As a matter of fact they got into Paston Oatvile station before the
3.47 from London was signalled. The 4.50 from Paston Oatvile had
to connect with it for the sake of passengers going on to Paston
Whitchurch or Binver, and was still wandering up and down in a
siding, flirting with a couple of milk-vans and apparently enjoying
itself. The platform was nearly bare of passengers, a fact on which
Reeves artfully commented to an apathetic porter.
“Not many travelling? You wait till the London train comes in, sir;
there’s always plenty in that as change here.”
“I suppose it’s the first train people can get away from business
by, eh?”
“That’s right, sir; there ain’t nothing else stops here after the
midday train. Of course there’s the fast train to Binver, but that
passes through ’ere. You travellin’, sir?”
“Just to Binver. Hullo, there’s the booking-office opening at last.
D’you mind getting two firsts for Binver, Gordon? Very sad thing that,
about Mr. Brotherhood,” he went on to the porter.
“That’s right, sir; very melancholy thing, sir.”
“I suppose you didn’t see him get on to the train?”
“There’s such a lot of ’em, sir, you don’t notice ’em, not the ones
that travel every day. And Mr. Brotherhood, ’e was a man as ’adn’t
many words for anybody. Though of course there’s some as is
different; d’you know Mr. Davenant, sir, up at the Hatcheries? He’s a
nice gentleman, that is, has a word for everybody. I seed ’im getting
off of the London train, and ’e asked me after my bit of garden—
nothing stuck-up about ’im. Excuse me, sir.” And, as the London train
swung into view, he proceeded up and down the platform making a
noise something like Paston Oatvile, for the information of anybody
who could not read notice-boards.
The London train was undeniably full to overflowing, and even
when the Paston Oatvile residents had diminished the number, there
were enough waiting for the Paston Whitchurch and Binver train to
leave no compartment unoccupied. Even in their first-class carriage,
it was only by luck that Reeves and Gordon managed to travel by
themselves.
“I say,” began Gordon, “why Binver? We don’t want to go beyond
Whitchurch, do we?”
“Oh, it’s just an idea of mine. We can get a train back in time for
dinner. Don’t you come unless you’d like to. Steady, here we are.”
And they swept slowly past the scene they had just been viewing
from the solid ground. Reeves opened the door a little as they
passed, and threw out a fresh stone; he had the satisfaction of
seeing it disappear exactly according to schedule. “Now,” he said,
“we’ve got a quiet quarter of an hour to spend before we get to
Binver. And I’d be dashed glad if you’d tell me two things. First, how
can anyone have planned and executed a murder in a third-class
carriage on a train so infernally crowded as this one is?”
“They may have been travelling first. No one examines the
tickets.”
“But even so, look at the risks. We should have had that fat old
party in here if I hadn’t puffed smoke in her face, and there are very
few firsts on the train. Our man took big chances, that’s certain.”
“And the other point?”
“Why did Davenant come up by this train yesterday? Of course
you don’t know the place as I do, but Davenant’s a scratch player,
and a bit of a local celebrity. Every child in the place knows that
Davenant only comes down here for week-ends, and it’s impossible
to get a game with him except on Sunday. Why does he suddenly
turn up on a Tuesday afternoon?”
“Well, I suppose he’s a right to, hasn’t he? I thought you were
saying he has a cottage here?”
“Yes, but one’s bound to notice every deviation from the normal
when one’s trying to trace causes. Look here, here’s Whitchurch. Do
you mind getting out and calling at the Hatcheries—that house,
there—and finding out, on some excuse, when Davenant got there,
and whether he’s there now? You’re not known, you see—but be
devilish tactful; we don’t want to put anybody on his guard.”
“Right-o! more lying necessary, I foresee. Oh, what a tangled web
we weave when first we practise to deceive. So long, Sherlock, meet
you at dinner.”
Reeves’ errand, it appeared when he got to Binver, was once more
with the railway staff. He went up to a porter, and said, “Excuse me,
does this train get cleared out here? I mean, if one leaves a thing in
the carriage, would it be taken out here?”
“That’s right, sir. Left Luggage Office is what you want.”
“Well, this was only a paper book. I thought perhaps you people
cleared them away for yourselves, like the newspapers.”
 “Ah, if it was a paper book, we ’aven’t any orders to take that on
to the Left Luggage Office. We takes those away, mostly; what
might the name of your book be, sir?”
This was not at all the question Reeves wanted, but he was
prepared for it. “It was The Sorrows of Satan, by Miss Corelli,” he
said. “I left it in one of these carriages yesterday.”
“Well, sir, I cleaned out this train yesterday myself, and there
wasn’t no book of that name. A passenger must have taken it out
with them most likely. There wasn’t not but one book I found in
those carriages, and you’re welcome to that, sir; I’ve got it on the
seat there.” And he produced a repellent-looking volume entitled
Formation of Character, by J. B. S. Watson.
Reeves was trembling with excitement, but it was clearly not a
case for showing any enthusiasm. “Well, give you sixpence for it,” he
said, and the porter willingly agreed—he had guessed rightly that
the sixpence would prove to be half-a-crown.
It was an agony dawdling back by a slow train to Paston Oatvile,
knowing that he could not get at the cipher-document till he
regained his rooms. Merely as a book, the thing seemed to lack
thrill. It seemed hours before he reached the dormy-house, and yet
Gordon had not returned. So much the better; he would be able to
work out the fateful message by himself. It could not be a
coincidence, though it had been a long shot to start with. A book of
that length (so he had argued to himself) would have been the sort
of book one reads in the train. Brotherhood would arrange to have a
cipher-message sent him out of the book which he had constantly in
his hands at the moment. He would be travelling with it; it was not
on the body or by the side of the line; the murderer might not have
thought of removing it. This, then, must be the book itself.
As he worked out the message he became less confident. It
appeared to run as follows: “Hold and it thoughts with the I highest
and to.”
“Damn,” said Mordaunt Reeves.
 Chapter VI.
The Movements of Mr. Davenant
Gordon felt that he was in a favourable position for inquiring into
the whereabouts of the mysterious Mr. Davenant. He was himself
little known at Paston Whitchurch, since he had only been a month
at the dormy-house, and his walks abroad had not carried him much
farther than the links. On the other hand, he knew a good deal, from
club gossip, about the habits of Mr. Davenant. The Hatcheries was
not one of the red-tile-and-rough-cast monuments with which a
modern architect had improved the scenery in the neighbourhood of
the links; it was a substantial cottage where, in grander days, the
home fisherman of Paston Oatvile Park used to live, and look after
all that was liquid in the property. It was now occupied permanently
by a morose gentleman called Sullivan, who acted as green-keeper
to the Club and did a little market gardening at home, and
occasionally (that is, during the week-ends) by the scratch player
and mystery man, Mr. Davenant. Legally speaking, the cottage was
Davenant’s property and Sullivan was the caretaker; actually, it
would be a clearer account of the position to say that Sullivan rented
the cottage from Davenant, and Davenant, every week-end, became
the lodger of his own tenant.
It was, then, as a member of the Club that Gordon must approach
his interview with Mr. Sullivan, and he was not left much choice of
disguises or of excuses. He decided that on the whole bluff would
pay best. Accordingly, as soon as Sullivan opened the door in answer
to his ring, he began:
 “Did Mr. Davenant leave any message for me this morning before
he left?”
“What’s that?”
“I met Mr. Davenant yesterday on the platform, and tried to make
some arrangements with him about having a game next Sunday, and
he said he’d leave a note for me at the dormy-house, but it isn’t
there, so I thought perhaps he’d left it here instead. Did he say
anything to you about it?”
“He did not. It’s not since Monday morning I’ve set eyes on Mr.
Davenant.”
“But he was here yesterday, surely?”
“He was not.”
“That’s very extraordinary, because I met him on the train, and I
certainly understood him to say he was coming here. Could he
possibly have been staying at the Club-house?”
“He might.”
“Well, I’m sorry to have troubled you. Good evening.”
Gordon had the definite impression that when Sullivan came to
the door he was not simply answering the bell; there had been no
time for him to hear the bell—he had been going out anyhow. There
was a thick hedge at the end of the path which led to the
Hatcheries; and behind this hedge, I am sorry to say, Gordon
concealed himself. He was the most placid and regular of men, but
the ardour of the hunt was beginning to lay hold of him. It was only
about a minute and a half later that Sullivan came out, carrying a
small bag, and took the path that led to the links. For a moment the
watcher thought of shadowing him, then decided that it would be
silly. If he went over the golf-links, the open ground would make it
quite impossible to follow without being noticed; besides, the links
would be full of people whom he knew, and he might easily get
delayed. He resolved suddenly on a still more heroic course. Nobody
else lived in the cottage—why not try to force an entrance while
Sullivan was out, and satisfy himself on circumstantial evidence
whether Davenant had really been in the cottage or not?
Breaking into a house is, as a rule, a difficult proceeding, even if it
is your own and you know the ropes. To break into a stranger’s
house, when you are not even certain whether a dog is kept; is a
still more heroic affair. The door had locked itself; the ground-floor
windows were shut and snibbed. The only chance seemed to be
crawling up the roof of a little outhouse and through an open
window on the first floor; a bathroom window, to judge by the
ample sponge which was drying on the sill. With rubber on his
shoes, Gordon made a fairly good job of the outhouse roof. The
window was a more serious proposition; it was very narrow, and
encumbered on the inside by an array of little bottles. It is easy to
put your head and shoulders through such a window, but that
means a nose-dive on to the floor. To put your legs through first is to
court the possibility of promiscuous breakage. Very carefully Gordon
removed all fragile objects out of range, and then with extreme
discomfort squeezed his legs through the opening. Even so, there
was a moment at which he felt his back must necessarily break,
when he was just half-way through. Landing at last without
misfortune, he set out quickly on a tour of the silent cottage.
It was only Davenant’s part of the house that interested him—the
bathroom, a bedroom, a small dining-room, and a study. They all
bore the marks of recent inhabitation; but was this anything to go
by? Davenant, in any case, would not be expected back for a week,
and Sullivan did not strike Gordon as the kind of man who would be
inclined to tidy up on Monday when Friday would do just as well.
The bed, indeed, was made; but the grate in the study had not been
cleared of cigarette-ends; the dining-room table was bare, but
Monday’s paper was still lying across a chair, as if thrown down at
random. On the whole the evidence pointed to Monday as the day of
departure; Monday, not Tuesday, appeared on a tear-off calendar; a
letter which had arrived on Monday evening was still waiting in the
hall; and there were no clothes left in the dirty-clothes basket. Such
an authority did Gordon feel himself to be on the subject of washing
since his experiences at Binver that he investigated equally the
clothes which had come back from the wash, and the list which
accompanied them. And here was a curious phenomenon; the list
referred to two collars, two handkerchiefs, and a pair of socks as
having been disgorged by the Binver authorities, but none of these
seemed to have crystallized in real life. “Binver is doing itself proud,”
murmured Gordon to himself, “or could it possibly be——” He went
and looked in the bathroom again: there was the sponge all right,
which seemed to insist that Davenant kept a duplicate series of what
the shops call toilet accessories; but where was the razor, the
shaving soap, the tooth-brush? It seemed, after all, as if Davenant
had packed for the week instead of leaving a duplicate week-end set
behind him. But—Good Lord! This was still more curious. There was
no soap in the bathroom, although there were traces of its presence
still discernible. Surely no one packing after a week-end in the
country took the soap with him? The face-towel, too, was gone; yet
the face-towel was distinctly mentioned in the washing-list. No,
decidedly there was something wrong about Davenant’s exit.
Another curious thing—there was every evidence that Davenant
was a smoker, and yet not a cigarette, not a pipe, not an ounce of
tobacco left in the study. Of course, it was possible that Sullivan was
very tidy and put them away somewhere, or that he was dishonest,
and treated them as perquisites. But once more Gordon had the
impression that Davenant had packed like a man who is leaving his
base, not like a man who has just week-ended at a Saturday-to-
Monday cottage. Like a man going abroad, even, or why did he take
the soap with him? One piece of supplementary evidence was to be
found in the study. A large and highly ornamented photograph frame
stood on the writing-table there; but it had no photograph in it, and
the back was unfastened, as if the portrait had been recently and
suddenly removed. If circumstantial evidence went for anything, it
seemed clear that when Davenant left the house last—apparently on
Monday—he left it in the spirit of a man who does not expect to
return immediately, and carries all his immediate needs with him.
So far the investigation had proceeded, when Gordon happened to
look out of a front window, and was discomposed by observing that
Sullivan was coming back already down the lane. There was no time
to be lost; he hastily ran downstairs and out at the front door. It
would be taking a considerable risk to trust to the mazes of the back
garden, and he decided to make for the hedge. But before he could
reach it, Sullivan turned the corner into the garden-path and
confronted him.
“I’m so sorry,” he said, on the inspiration of the moment, “but
could you tell me what Mr. Davenant’s address is? I shall have to
write to him, and this is the only address they’ve got up at the Club.”
“Mr. Davenant left no address,” said Sullivan, and, try as he would,
Gordon could not determine whether there was suspicion in his tone.
However, the awkward corner was turned, and it was with some
feeling of self-congratulation that he made his way back to the
dormy-house.
He came back to find Reeves closeted with Marryatt and
Carmichael, to whom he was explaining the whole story of their
adventures. “I hope you won’t think it a breach of confidence,” was
his explanation, “but the last disappointment I’ve had has made me
feel that we must be on the wrong tack somewhere; and it’s no
good for us two to try and correct each other. It’s like correcting the
proofs of a book; you must get an outsider in to do it. So I thought,
as Marryatt and Carmichael were with us at the start, it would be
best to take them fully into our confidence, and make a foursome of
it.”
“Delighted,” said Gordon. “I’ve been prospecting a bit, but I can’t
say I’ve got much forrarder.”
“Did you ask whether Davenant was there yesterday?”
“Yes, I interviewed Sullivan on the subject, and he said ‘No.’ ”
“That I can’t believe,” said Carmichael.
“Why, what about it?” asked Gordon, a little ruffled.
“I’m sure Sullivan didn’t say ‘No.’ Have you never observed that an
Irishman is incapable of saying yes or no to a plain question? If you
say, Has the rain stopped, he won’t say Yes, or No; he’ll say, It has,
or It hasn’t. The explanation of that is a perfectly simple one: there
is no native word for either in Irish, any more than there is in Latin.
And that in its turn throws a very important light on the Irish
character——”
“Oh, go and throw an important light on your grandmother’s
ducks,” said Reeves. “I want to hear about this interview. Was he
telling the truth, d’you think?”
 “From his manner, I thought not. So, when his back was turned, I
made bold to enter the house and take a look round for myself.” And
he described the evening’s entertainment in detail.
“By Jove, you are warming to the part,” said Marryatt. “I should
like to see you get run in by the police, Gordon.”
“You say,” Reeves interrupted, “that you don’t think he was there
yesterday, on the Tuesday, that is, because he hadn’t taken the
letter away. He went off, then, on Monday, but when he went off he
took with him all that a man normally takes with him if he’s going to
sleep in a different house that night, plus a piece of soap and a
towel, which are not things one usually carries about in one’s
luggage?”
“That’s the best I can make of it,” said Gordon. “And the
photograph—it might be an accident, of course, but I feel convinced
that he put that in his luggage at the last moment.”
“And that’s frightfully important,” said Reeves, “because it
obviously means that on Monday, before anything happened to
Brotherhood, Davenant was reckoning on leaving home for some
little time; and not returning immediately to wherever it is he lives
ordinarily, because he must keep collars and things there. But he
also thought he might be away for some longish time, or he wouldn’t
have worried to take the photograph with him. What was the frame
like?”
“Quite modern; no maker’s name on it.”
“I’m afraid that means the murder must have been a premeditated
one,” put in Marryatt. “I hope it’s not uncharitable to say so, but I
never did like Davenant. I don’t think I’m ordinarily a person of very
narrow religious views, and I’ve known Romans that were quite easy
to get on with. But Davenant was a person of quite ungovernable
temper, you must remember that.”
“His ungovernable temper would be much more important,”
objected Gordon, “if the murder were not a premeditated one.”
“But it’s not only that,” persisted Marryatt. “To me, there was
always something sinister about him; he had fits of melancholy, and
would rail at the people and the politicians he didn’t like in a way
that was almost frightening. Surely I’m not alone in that
impression?”
“What did Davenant look like?” asked Carmichael suddenly.
“Good Lord,” said Reeves, “you ought to remember that well
enough. You must have met him down here pretty well every week-
end, and he was quite well known.”
“Oh yes,” explained Carmichael. “I know what he looked like. I’m
only asking you to see if you remember. If you were asked in a
witness-box, what would you say Davenant looked like?”
“Well,” said Reeves, rather taken aback, “I suppose one would
certainly say he was very dark. Very dark hair, I mean, and a great
deal of it, so that it made the rest of his face rather unnoticeable.
What I generally notice about a man is his eyes, and I never got
much impression of Davenant’s, because he nearly always wore
those heavy horn-rimmed spectacles. And then of course he was a
rattling good player. If he murdered Brotherhood, as Marryatt seems
positive he did, I can tell you one motive that I can’t accept for his
doing it. He wasn’t jealous of Brotherhood’s golf. Poor old
Brotherhood was about as rotten as Davenant is good.”
“It’s very extraordinary to me,” said Carmichael, “that you should
say all that, and yet not have arrived at the obvious fact about this
mystery. The root fact, I mean, which you have to take into account
before you start investigating the circumstances at all. You simply
haven’t seen that fact, although it’s right under your nose. And that’s
a very curious thing, the way you can look at a complex of facts
ninety-nine times, and only notice the point of them the hundredth
time. The phenomenon of attention——”
“Oh, cut it out,” said Gordon; “what is the fact we haven’t
noticed?”
“Oh, that,” said Carmichael lightly, “merely that Brotherhood is
Davenant, and Davenant is Brotherhood.”
 Chapter VII.
Carmichael’s Account of it
“Good Lord,” said Reeves, when the first shock of astonishment
was over, “tell us some more about it. How did you know?”
Carmichael joined the tips of his fingers and beamed at them,
secure of an audience at last. “Well, you’ve just admitted that all you
can remember about Davenant is hair and spectacles. That is, his
disguise. Of course the man wore a wig. He was a fictitious
personality from start to finish.”
“Except for his golf,” suggested Gordon.
“Yes, that was real enough; but Brotherhood’s wasn’t. Don’t you
see that the two characters are complementary, suspiciously
complementary? Brotherhood is here all the week, but never during
the week-ends; Davenant is only seen from Saturday to Monday.
Davenant is Catholic, so as to be violently distinguished from the
atheist Brotherhood. Davenant is good at golf; so Brotherhood has
to be distinguished by being very bad at golf, and that, to me, is the
mystery of the whole concern. How a scratch player could have the
iron self-control to play that rotten game all the week, merely to
prevent our suspecting his identity, beats me entirely. And yet you
could find parallel instances; old Lord Mersingham, for example——”
“Do you mean,” said Gordon in a shocked voice, “that Brotherhood
pulled his drives like that on purpose?”
“Precisely. After all, don’t you remember that day, let me see, I
think it was last February, when Brotherhood played for fifty pounds,
and went round in eighty-nine? Of course, there are flukes even in
golf. I remember myself——”
 “Well,” said Gordon, “I think the Committee ought to do something
about it. Dash it all, I was his partner in the foursomes.”
“De mortuis,” suggested Reeves. “But I still don’t see why he
wanted to do it, I’m afraid. Why, the thing’s been going on for
years.”
“Well, none of us know much about Brotherhood’s business; but I
gather from what people are saying about the bankruptcy that it was
a pretty shady one. They haven’t traced any hole in the accounts;
but if there ever was a man you would expect to go bankrupt and
then skip (I believe it is called) with what is described in such
circumstance as the boodle, that man was Brotherhood. He foresaw
the probability of this for years, and made very careful and subtle
preparations for meeting the situation. The important thing on such
occasions is to have an alter ego. The difficulty is to establish an
alter ego on the spur of the moment. Brotherhood knew better. He
had been working up his alter ego for years.”
“Right under our noses!” protested Reeves.
“That was the cleverness of it,” said Carmichael. “If there was a
Mr. Brotherhood at Paston Oatvile, and a Mr. Davenant every week-
end at Brighton, nobody would be deceived; it’s a stale old dodge to
keep up a double establishment in two different places. The genius
of Brotherhood’s invention was that he kept up two establishments
within a stone’s throw of each other. Nobody here could actually say
that he’d seen Brotherhood and Davenant together, that goes
without saying. But the two personalities were real personalities in
the same world; and there were hosts of witnesses who would
declare that they knew both. If Brotherhood suddenly ceased to
exist, the last place where anybody would thing of looking for him
would be the house next door.”
“Good God, what a fool I’ve been!” said Mordaunt Reeves.
“The separate banking-account would be particularly useful to
such a man,” Carmichael went on. “If we could find out where
Davenant banked, I have no doubt that we should discover a
substantial balance. But of course, he wouldn’t bank here.”
“Why not?” asked Gordon.
 “Because Brotherhood would want to deal with the local bank, and
it’s a very unsafe business making your signature in a forged writing.
So Davenant will have banked in London. By the way, that’s another
point, did you notice that Davenant always used a typewriter? He
couldn’t risk the use of a false handwriting.”
“Elementary, my dear Watson,” murmured Gordon to himself.
“Well, he knew when the crash was coming; it was all carefully
prearranged. He had even the impudence to book a sleeping car to
Glasgow.”
“But that’s a difficulty, surely,” put in Reeves. “Because he ordered
the sleeper not as Davenant but as Brotherhood. Now, you make out
that Brotherhood was to have disappeared, as from yesterday, and
Davenant was to have become a permanency. Why didn’t he order a
sleeper in the name of Davenant?”
“How often am I to tell you, my dear Reeves, that you are dealing
with a genius? If Davenant had ordered a sleeper just at that
moment, attention might have been directed to him. But ordering a
sleeper for Brotherhood would merely strengthen the impression
that Brotherhood had disappeared. If that was all—personally I
believe the scheme was even more audacious. I believe Davenant
did mean to go away, for a time at any rate, and by that very train.
He would join it at Crewe, travelling in an ordinary first-class
carriage. Then at Wigan—bless my soul!”
“What’s the matter?” asked Gordon.
“I find myself unaccountably unable to recollect whether the 7.30
from Euston stops at Wigan. However, it will do for the sake of
argument.
“At Wigan an anonymous passenger, Davenant, of course, would
ask the sleeping-car attendant whether he had a vacant berth. And
the man was bound to have a vacant berth—Brotherhood’s. There
was one person in the world whom nobody would suspect of being
Brotherhood—the stranger who had been accommodated, quite
accidentally, with Brotherhood’s berth.”
“It’s wonderful!” said Marryatt.
“But of course, all that is only speculation. Now we come to
something of which we can give a more accurate account—the plans
which this Brotherhood-Davenant made for the act of
metamorphosis. I take it this was his difficulty—Brotherhood and
Davenant (naturally enough) do not know one another. If Davenant
walks out of Brotherhood’s office in London, it will create suspicion—
the change, then, must not happen in London. If Davenant is
suddenly seen walking out of Brotherhood’s bungalow, that again
will create suspicion; the change, then, must not take place at
Paston Whitchurch. The thing must be managed actually on the
journey between the two places. That is why Brotherhood-Davenant
wears such very non-committal clothes—hosiery by which, in case of
accident, he cannot be traced; the very handkerchief he takes with
him is one belonging to a stranger, which has come into his
possession by accident. He even has two watches, one to suit either
character. Thus, you see, he can be Brotherhood or Davenant at will,
by slipping on a wig and a pair of spectacles.
“He did not do anything so crude as to change once for all,
suddenly, from Brotherhood to Davenant at a given point on his
journey. He alternated between the two roles all along the line—I
have always wondered what is the origin of that curious phrase ‘all
along the line’; here I use it, you will understand, in a quite literal
sense. Davenant got out at Paston Oatvile—the porter saw him. But,
as we now know, it must have been Brotherhood who was travelling
in the 4.50 from Paston Oatvile on. Yet it was as Davenant that he
would have got out at Paston Whitchurch.”
“How do you know that?” asked Reeves.
“The ticket, of course. Brotherhood had a season, naturally, for he
went up and down every day. At Paston Whitchurch, therefore,
Brotherhood would have produced a season. It was Davenant who
needed an ordinary single. The effect of all this was to create the
simultaneous impression that both Brotherhood and Davenant came
back to Paston Whitchurch that afternoon, and came by the same
train. There was only one hitch in the plan, which could hardly have
been foreseen. If Brotherhood happened to be murdered on the
way, it would look very much as if Davenant had murdered him. And
unfortunately he was murdered.”
 “Then you don’t think it was suicide?” asked Marryatt, with a catch
in his voice.
“If it was suicide, it was the result of a momentary insanity, almost
incredibly sudden in its incidence. Suicide was certainly not part of
the plan. A bogus suicide was, of course, a conceivable expedient; it
would have been one way of getting rid of the undesirable
Brotherhood’s existence. But consider what that means. It means
getting hold of a substitute who looked exactly like Brotherhood—he
could not foresee the excoriation of the features—and murdering
him at unawares. It meant, further, that Davenant would be
suspected of the murder. No plan could have been more difficult or
more clumsy.”
“Then you mean,” said Marryatt, “that we have to look for a
murderer, somebody quite unknown to us?”
“I didn’t say that,” replied Carmichael, with a curious look. “I mean
that we have to look for a murderer, some one whom we have not
hitherto suspected. If Davenant murdered Brotherhood, that was
certainly suicide; for Davenant was Brotherhood. But that seems to
me impossible. The evidence all goes to show that there was a very
careful plot in contemplation, which was cut short by a quite
unforeseen counterplot.”
“But look here,” said Reeves, “if the original plan had come off, did
he mean to come back here and live on here as Davenant?”
“You mustn’t expect me to know everything; I can only go by the
indications. But I should say that he really would have come back
here as Davenant, perhaps about three weeks later, and settled
down permanently at the Hatcheries; or perhaps even—he was a
very remarkable man—he would have bought up Brotherhood’s
bungalow. You see, he liked the place; he liked the company; the
only thing he disliked was having to play golf badly, and that
necessity disappeared, once he settled down as Davenant. A wig is a
nuisance, but so is baldness. The last place where anybody would
look for Mr. Brotherhood, last heard of on the way to Glasgow, would
be Paston Whitchurch, where Mr. Brotherhood had lived.”
“I’m afraid I’m very stupid,” said Gordon, “one of Nature’s
Watsons, as I said yesterday. But what about all the silly little
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

textbookfull.com