►
Description
Many astrophysical observations support the existence of a dark matter component in our universe. However, after a few decades of active research, the nature of dark matter remains elusive. In this context, the LZ experiment aims to detect dark matter using a multi-tonne detector filled with liquid xenon and located at the SURF underground laboratory. We expect a handful of signal events over a significant background during the five years of operation in such a detector. Observing deviation from the background model would sign a dark matter detection. Hence, the LZ collaboration is engaged in a great effort of developing a robust signal and background model using Monte-Carlo simulation. Simulating such a detector requires the usage of any computing available. Therefore, we developed a framework to take advantage of all the HTC (PDFS) and HPC (EDISON and CORI) resources available at NERSC for the second and third mock data challenges. Using a centralized job submission system, containerization technologies, and software distribution service (CVMFS) made it possible.
A
And
the
neptune
discovery,
so
the
neptune
position
was
mathematically
predicted
independently
in
1845
46
by
urban,
very
and
john
adams,
by
studying
the
irregularity
orbits
of
uranus.
And
here
what
you
can
see
is
our
pictures
of
the
the
old
style
of
doing
physics.
But
what
is
really
interesting
is
that
first
we
had
this
theoretical
collection
of
the
position
of
neptune
and
then
it
was
confirmed
on
the
same
year
a
bit
later
by
observation,
and
we
can
draw
a
parallel
between
this
discovery
of
neptune
and
the
dark
matter.
A
Here
and
at
the
at
the
edges
of
the
galaxy,
so
this
is
something
that
we
can
measure
and
guess
what
the
measures,
the
measurement
of
the
rotation
velocity
as
a
function
of
radius,
are
exactly
showing
this,
which
supports
the
hypothesis
of
having
this
component
of
dark
matter
in
the
in
the
universe,
sorry
in
the
galaxy
and
then,
if
we
protoscope,
we
have
this
other
scale.
That
is
a
galaxy
cluster
scale.
A
So
here
this
is
an
optical,
the
bullet
cluster.
So
here
the
x-ray,
which
is
giving
us
the
the
gas
distribution
and
then
the
len
record,
ending
in
fact
that
is
giving
us
the
dark
matter.
If
you
want
and
then
what
we
can
see
is
clearly
this
separation-
and
it's
even
clearer
here
on
the
original
paper-
plots
where
you
can
see
here:
the
micro
lensing
for
the
gravitational
potential,
with
the
two
barycenters
here
and
here
the
gas,
and
you
can
even
see
here
the
shock
wave
of
the
the
merging
of
the
gas.
A
So
that
shows
us
that
at
a
lot
of
scales,
we
have
dark
matter
and
we
even
have
dark
matter
at
all.
Scales
like
at
the
linear
scales
and
what's
interesting,
is
that
at
the
universe
scale,
the
dark
matter
is
more
or
less
following
the
matter
density,
but
it's
also
forming
like
filaments
between
the
different
clusters.
A
A
Search
because
I
said,
wimp
is
just
one
candidate
for
dark
matter.
We
have
so
many
other
and
actions
are
another
good,
good,
dark
matter
candidate,
but
the
lc
dark
matter
search
experiment
is
mainly
focusing
on
the
whimp
dark
matter,
and
it's
for
this.
It's
a
very
good
candidate.
So
we
are
using
the
liquids
in
time
protection
chamber
technology,
which
is
a
great
target
because
it
has
a
high
density.
A
high
ionization
and
insulation
yield
it's
transparent
to
its
own
insulation
light.
So
this
means
that
we
can
extract
a
signal
from
it.
A
We
have
this
effect
of
shell
self-shielding
that
I'm
going
to
mention
in
a
bit
later,
and
it
has
also
a
high
intrinsic
radio
priority.
This
means
that
the
xenon
that
we
are
using
to
do
the
detection
is
very
pure
and,
as
I
mentioned,
it
can
also
do
other
things
like
actually
like
particles,
exotic
dark
matter,
candidates
and
also
some
nuclear
science
for,
for
example,
the
nutrients,
the
beta
decay.
A
So
here
is
a
diagram
of
the
lz
time,
conduction
temperature.
So
when
you
have
a
particle
interacting
in
the
liquids
in
an
it
deposits,
it
deposits
some
energy
and
this
deposit
of
energy
is
going
to
create
to
ionize
and
sorry
it's
going
to
excite
the
medium,
which
leads
to
the
creation
of
primary
scintillation
light.
A
This
is
the
so-called
s1
signal
that
you
can
see
also
here
and
then
because
of
the
application
of
an
electric
field
on
the
liquid
xenon
volume,
you
have
the
drift
of
the
electrons
and
then,
when
the
drift,
the
electrons
are
reaching
the
gas
pocket
at
the
top
of
the
detector.
They
are
amplified
by
a
much
higher
electric
field
and
then
the
ele,
the
ionization
electrons,
are
emitting
light
by
electroneum
distance,
creating
the
so-called
s2
signal
that
that
you
can
see
here
so
on
this
excuse.
This
is
the
drift
time,
so
you
have
the
first
peep.
A
That
is
the
s1
that
you
can
see
at
t0
at
t0
and
after
some
time
because
of
the
drift
of
the
electron.
You
can
see
the
s2
so
by
using
this
different
time.
Difference
between
this
one
and
s2.
You
can
reconstruct
the
the
z
position
and
by
using
the
light
pattern
as
it
seems
to
be
highlighted
on
the
on
the
pnt's
right
here.
A
In
total,
we
have
10
10
tons
of
liquid
xenon
of
xenon,
sorry
and
seven
tone
just
in
the
in
the
active
volume
and
when
we
are
going
to
fiducialize
it
so
being
removing
the
edges
to
eliminate
background.
We
are
going
to
reach
a
fiducial
volume
of
5.6
ton,
which
is
the
volume
that
matter
for
the
analysis
and
the
electric
field.
A
The
target
electric
field
is
about
300
percent
better,
and
this
light
is
going
to
be
collected
by
this
top
and
bottom
arrays
that
you
can
see
here
and
here
which
are
composed
both
by
247
pnt's
oops.
I
missed
to
remove
this
slice,
sorry
about
that
anyway,
so
the
llc
experiment
is
located
at
the
surf
underground
laboratory,
which
is
in
south
dakota.
A
So
it's
about
one
mile
underground,
it's
right
here
in
the
davis
capital,
so
you
can
see
here
a
picture
of
the
shaft,
the
picture
of
the
the
tunnels
to
to
go
to
the
to
the
lab,
and
here
this
is
a
picture
of
the
water
tank.
So
we
need
to
go
underground
to
reduce
the
background
coming
from
customy
trades,
which
is
a
very
important
for
us,
because
we
want
to
reach
a
very
low
background,
so
meet
a
diagram
of
lc.
A
I
have
pictures
coming
and
then
here
you
can
see
the
the
tpc
that
I
just
shown
you
a
bit
before,
and
the
different
subsystems
of
of
lt
that
I'm
going
to
describe
so
here
this
there
are.
These
are
like
few
pictures
of
the
of
the
gpc.
So
this
this
is
the
gpc
built
assembled
at
the
at
the
surface
lab.
So
you
can
see
here
the
ptfe
reflector.
Here
you
have
the
bot,
the
top
and
the
bottom
array
of
c's
that
you
can
see
here.
A
So
we
also
have
this
other
subsystem
that
is
called
the
that
we
call
the
liquid
insulator
veto.
That
is
right
here.
So
here
we
have
17
tonne
of
gadolinium
loaded
liquid
scintillator,
and
that
is
monitored
by
those
eight
inch
tnts
all
around
all
around
it,
and
what
you
can
see
here
is
a
picture
of
the
the
liquid
scintillated
vessel
that
is
in
acrylic
here
and
then
here.
A
So
a
very
important
part
of
lz
is
the
two
scheme,
the
two
v
s
veto
systems.
There
is,
as
I
mentioned
this
skin,
which
is
which
you
can
see
here
here.
I
put
up
a
square
here
to
indicate
the
skin
pnt,
but
the
pimples
are
not
located
at
the
bottom.
They
are
just
looking.
They
are
located
at
the
top.
That's
just
for
illustration,
and
to
tell
you
that
we
are
instrumenting
this
this
region,
so.
B
A
Okay,
so
oops,
okay,
that's
good!
So
when
we
have
gamma
coming
from
the
impurities
uranium
thorium
of
the
the
different
materials,
the
gamma
can
interact
in
the
gpc
and
escape
the
tpc
without
leaving
any
more
energy.
So
these
deposits
here
could
be
interpreted
as
a
single
scatter
electron
record
deposits,
but
because
it
interacts
in
the
skin,
we
can
I
easily
flag
it.
A
A
A
We
were
able
to
reduce
the
detector
material
background
to
a
very
low
level
here
and
here
in
the
nuclear
recall
background,
we
are
mainly
the
domain
component
is
going
to
be
the
detector
materials
and
also
the
the
neutrinos,
and
this
is
exactly
what
you
can
see
on
these
two
pie.
Charts
and
here,
for
example,
for
the
nuclear
record
background
yeah.
Actually,
the
domain
component
is
the
in
this
6
30
kv.
Nuclear
recoil
is
the
atmospheric
neutrinos,
and
in
here
it's
the
radon
222
for
the
electronic
background.
A
A
Oh
I'm
sorry,
there
is
a
level
up
here
so
this
and
this
figure
is
a
simulation
showing
a
1000
day
run,
which
is
the
expected
duration
for
foreign,
and
this
yellow
orange
area
is
a
signal.
The
signal
region
for
40
gb,
and
here
you
can
see
some
boron
8
and
http
nuclear-recorded
neutrinos
that
are
in
this
region.
So
we
are
going
to
be
sensitive
to
this
to
disguise
and
so
before
the
discrimination
we
expected
131.
A
Electron
recoil
events
and
after
discrimination,
we
reduce
that
by
quite
a
lot.
It's
just
like
thinking
about
applying
a
sharp
cut
in
here.
This
is
not
exactly
what
we
are
doing,
but
that's
just
an
illustration
here,
and
then
we
can
use
the
veto.
So
I
mentioned
the
self
shielding
of
the
liquid
xenon.
That
is
clearly
visible
on
this
on
this
figure.
A
So
what
you
can
see
here
is
that
the
events
are
located
mainly
at
the
edges
of
the
detector,
because
the
nicotine
is
very
dense
and
it's
going
to
be
difficult
for
neutrons
or
electric
coils
to
penetrate
at
very
deeply
in
the
in
the
digital.
So
this
is
more
or
less
like
a
background.
Three
ish
area
by
background
three,
I
mean
a
low
background
area
and
when
you
apply
the
video
cut,
what
is
extremely
interesting
is
that
you
can
expand
and
reduce
this
leakage
of
events
instituted
into
the
volume
and
this
untouched.
A
A
So
we
have
in
the
in
the
full
chain
simulation
of
the
tour
two
main
software,
so
the
first
one
is
called
backrat,
which
is
our
general
based
lc
detector
model.
You
can
see
here
a
representation
of
visualization
of
the
lc
geometry
implemented
in
this
in
this
framework,
and
we
have
also
this
der
software,
which
is
the
detector
electronic
response.
Oh,
I
need
to
write
that
the
software
that
is
taking
all
the
pmt
hits
because
we
do
light
propagation
in
the
in
the
tpc
and
convert
that
into
a
measurable
signal.
A
A
We
do
backrat
and
dr
that
we
are
applying
the
two
in
the
same
step,
which
is
called
here
in
the
multicolor
production,
and
then,
after
that
di
is
going
to
produce
a
file
that
will
look
like
a
data
acquisition
file.
We
are
processing
this
file
to
extract
quantities
that
we
that
we
are
interested
on,
and
so
we
have
plot
mock
data
challenges
that
are
designed
to
stress,
exercise
and
test
our
computing
and
analyze
its
capabilities.
A
A
So
to
do
the
handling
of
this
large
number
of
jobs,
we
developed
this
infrastructure
software
that
is
more
or
less
summarized
by
this
by
this
diagram
and
I'm
going
to
to
make
some
poses
on
the
different
parts
of
this
of
this
diagram.
But
the
main
piece
here
is
the
p
squared
software
that
has
been
developed
for
diabetes
and
we
are
using
it
quite
a
quite
a
lot
at
the
moment,
still,
okay.
So,
let's
start
with.
A
What's
on
the
top,
so
detector
parameters,
item
and
process
definition,
so
the
difficult
parameters
are,
as
you
can
see
here,
for
example,
the
all
the
detailed
parameters
that
you
can
or
you
want
to
vary
in
your
simulation.
So
this
is
the
input
for
mdc3.
So
you
you
have
here
the
time
which
is
in
fake
days
here
and
then,
for
example,
you
can
have
you
can
see
here
the
radon
220
activity
as
a
function
of
time.
So
this
is
something
we
decided
to
query.
A
We
also
had
some
variation
of
the
electronic
lifetime
in
the
in
the
indus
space,
which
is
some
something
that
we
also
decided
to
marry,
and
so
this
we
encoded
this.
In
variable
information
into
the
items
which
is
a
viable
part
of
the
of
the
of
the
job
and
the
process,
definition
which
is
going
to
more
or
less
tell
us
how
to
execute
the
job
is
independent.
A
It's
a
static
part
of
the
job,
so
we
are
not
going
to
vary
it's
more
or
less
job
execution
instructions
set
of
scripts
to
execute
before,
during
and
after
job
we.
Yes,
we
have
this
capability
of
executing
scripts
before
and
after
job,
which
are
extremely
important.
If
you
want
to
download
data
and
to
upload
data-
and
yes,
then,
this
is
where
you
can
add
the
instruction
for
the
data
transfer
and
archive.
A
So
then
comes
the
ps4
software,
which
you
give
to
square
all
of
these
information,
so
peacequad
is
provides
a
restful
interface
for
database,
which
contains
the
history
of
each
job
ever
submitted,
and
what
you
can
see
on
this
diagram
is
kind
of
the
history
of
of
a
job,
so
you
send
it
it's
processing,
so
it's
in
the
queue
executing
and
it's
then
it's
processed,
which,
which
means
that
so
that's
good
or
it
fell.
And
then
you
have
some
some
ways
to
reset
this
fail
status.
A
Then,
as
you
could
see
in
this
diagram,
we
send.
There
is
an
interaction,
a
link
between
t
squared
and
rapid
mq
rapidm
q
is
an
open
source
message
broker
software
that
incorporates
a
queuing
system
and
it
ensures
the
communication
between
psquare
and
the
workers
it's
acting
as
an
interface
so
in
operation.
A
Here.
What
you
can
see
is
the
active
monitoring
of
the
jobs
in
rabbitmq.
Here
is
the
in
in
yellow
it's
all
the
ready
jobs
and
it
do
it's
all
the
jobs
that
are
all
already
running.
So
here
you
have
the
face
of
insertion
or
publication
of
the
of
the
jobs,
and
here
it's
when
we
started
some
some
workers,
so
workers
were
online
and
asking
some
some
jobs
to
to
run
to
do
their
rapidem
queue.
A
So
let's
take
a
look
at
some.
Some
plots
from
mdc
to
post
modem.
So
here
this
is
the
main
benchmarking
that
we
used
during
mdc2,
which
is
pdsf.
So
the
album
on
pdsf
was
a
bit
different
than
korean
edison,
but
that's
still
a
very
good
reference
for
us
oops.
The
colors
are
not
that
good
and
what
the
first
thing
is.
A
A
But
that
happened
because
we
were
probably
too
greedy
on
this.
But
the
good
thing
is
we
were
able
to
run
this
production
on
pdsf
korean
as
well,
sorry,
pdsf,
curry,
as
well
and
edison.
At
the
same
time,
and
even
with
this
that
was
kind
of
quite
beneficial,
and
if
you,
if
we
look
at
the
usage
fraction,
we
mostly
use
edison
and
curry.
During
this
whole
production.
A
So
here
it's
another
version
of
this
plot,
where
you
can
see
here
the
exception
time
as
a
function
of
that
of
the
date,
and
that
corresponds
to
the
time
where
we
were
too
greedy
and
we
started
too
many
too
many
workers
and
that
stressed
a
lot.
The
project,
a
file
system
that
was
the
pre-see
community
file
system.
A
But
now
we
know
it
and
we
are.
We
are
going
to
be
way
more
careful
handling,
ios
here.
So
we
also
took
a
look
at
the
performance
of
knl,
so
there
is
quite
the
difference
between
the
so
we
used
here
edison
but
edison
and
corey
as
well.
I
have
similar
performance,
so
you
can
assume.
That's
that's
okay
here.
So
if
you
compare
the
two,
there
is
more
or
less
a
factor
four
in
time
execution
time
between
the
two
and
when
you
do
the
math
and
you
compare
the
charge.
A
So
the
edison
cost
us
about
two
hour
not
hour
and
a
correct
panel,
a
nineteen
hour,
which
means
that
for
us,
curry
channel
will
be
too
expensive
to
use
for
the
mdc2
production,
and
this
is
mainly
caused
because
of
the
this.
Is
this
main
cause
is
because
of
the
memory
leaks
that
we
we
had
in
the
in
the
code,
so
between
mbc2
and
mdc3?
We
made
a
lot
of
work.
A
We
fixed
a
lot
of
memorabilics
to
reduce
the
memory
footprint,
so
we
went
from
four
to
five
gigabytes
to
about
three
gigabytes
and
we
also
increase
the
number
of
simulation
features,
for
example
the
electron
train.
So
after
next
to
the
largest,
you
can
have
like
some
electro
electrons
that
are
emitted
from
from
the
cathode.
That
can
create
some
some
some
signals.
So
we
added
this
feature
in
here
and
we
also
decided
to
train
the
bias
mitigation
strategy,
which
is,
we
just
add,
fake
events
into
into
the
data.
A
So
those
are
the
the
big
change
between
the
big
changes
between
mvc2
and
mpc3
at
the
level
of
numerous
ptsf
and
edison,
where
the
commission
previous
to
nbc3,
so
only
query
that
was
available
and
thanks
to
this
reduction
in
memory
usage,
we
were
able
to
use
curry
channel
a
lot
more
than
previously
and
for
the
infrastructure
we
changed.
A
few
things
so
first
spread
was
the
point
spin
to
reduce
the
network
issues
that
that
we
had,
and
we
also
use
multi-threaded
and
workers
instead
of
single
single
task
worker.
A
A
A
The
main
thing
is
that
the
number
of
events
also
changed
in
the
in
the
jobs,
but
the
thing
that
is
kind
of
interesting
is
that
you
can
see
still
the
difference,
the
same
difference
between
haswell
of
kennel,
this
factor,
four-
that
is
more
or
less
still
here,
but
we,
the
the
the
good.
The
message
from
this
slide
is
that
we,
by
reducing
the
footprint
we
were
allowed
to
make
use
of
curry
channel
pretty
efficiently
during
the
ntc3
production.
A
So
let
me
wrap
up
the
take-home
stage
for
from
the
mdc's
summary,
so
mdc's
are
mock.
Data
challenges
are
very
great
tools
to
stress
the
data
production
system.
During
these
two
stages,
we
identified
a
lot
of
bugs
and
bottlenecks
during
the
motorcycle
production
and
that
helped
us
a
lot
to
fix
them
and
to
address
them
and
then
to
be
ready
for
the
first
science
for
example.
A
Also
and
a
great
point
of
doing
these
small
data
changes
is
that
they
help
us
to
build
confidence
into
our
infrastructure
tools,
because
everything
else
hold
the
charge
and
the
load.
Even
if
you
have
like
some
small
heat
cap,
everything
holds
pretty
well
and
everything
behaved
pretty
pretty
well.
So,
for
example,
the
database
interface
work
perfectly.
A
The
automated
data
transfer
worked
perfectly
within
within
ask
so
the
data
movement
from
scratch
to
project
a
right
yeah
that
was
projected
by
then
and
also
b
squared,
was
able
to
scale
up
after
migrating
to
to
spin
and
reducing
those
networked
glitches
that
we
were
observing.
So
the
overall
summary
of
mock
data
changes
is
that's
great.
You
should.
A
A
The
first
science
so
now
we
have
also
data
production
activities
to
support
science.
We
have
the
multi-panel
production
because
we
still
need
to
do
the
tuning
of
the
simulation
detector
model,
which
is
something
something
that
is
ongoing
at
the
moment.
We
are
preparing
everything
to
be
ready
when
we
are
going
to
get
the
data
to
do
this
tuning
of
the
of
the
simulation
model.
We
are
also
doing
some
background
modern
production,
because
this
is
really
important
and
some
components
need
a
lot
of
statistics.
A
So
this
is
something
that
is
also
very
important
for
us
to
do
and
to
support
and
for
the
data
processing.
So
we
have
two
kinds
of
data
processing
activities.
We
have
the
online
processing
so
when
a
file
arrives
in
the
way
warehouse.
That
knows,
we
want
to
have
this
automatic
processing
of
the
of
the
file
and
we
have.
A
We
are
going
to
do
some
automatic
maintenance
of
the
workers
to
cover
the
science
run
and
the
calibration
runs,
and
then
also
we
have
something
that
is
more
by
hand,
processing,
which
is
the
reprocessing
of
the
data.
So,
where,
once
we
are
going
to
have
all
the
data
we
are
going
and
those
processing
softwares
will
be
upgraded,
we
are
going
to
have
to
reprocess
them
and
yeah.
Those
are
the
two
data
protection
or
data
processing
activities
that
that
we
are
so
we're
also
trying
to
get
ready
for
for
parameter.
A
A
With
respect
to
to
the
rest,
if
we
can
offload
this
on
the
gpu,
we
can
just
divide
it
by
about
1
600
and
then
that
reduce
reduces
a
lot.
The
yeah,
the
usage
of
cpu
and
gpu,
and
that
will
speed
up
a
lot
our
simulation.
So
thanks
to
this
we
could
envisage
and
think
about
doing
a
lot
more
precise
physics
and
then
here
this
is
a
diagram
that
shows
what
we
we
are
thinking
to
do
so
here.
This
is
the
food
chain
that
I
mentioned
before.
A
So
we
have
this
simulation
going
on
inside
and
with
the
light
collection.
Sorry,
the
light
ray
tracing
done
internally
in
background
on
using
cpus
and
then
what
we
can
do
is
we
can
offload
all
of
these
to
two
optics
directly
and
then
get
the
same
the
same.
So
this
is
something
that
is
going
on
going
and
is
going
to
be
very
interesting
for
the
for
the
future.
A
A
So
the
physics
data
taking
is
happening
this
year
or
will
be
happening
this
year
and
for
the
dark
matter,
the
addiction
side
we
are
going.
We
are
supposed
to
be
the
best
experiment
in
terms
of
sensitivity
at
40,
gb
win,
which
is
more
or
less
yeah,
the
best
comparison
point
for
for
us
and
for
the
computing
side.
These
two
agencies
were
extremely
helpful.
A
The
first
the
nbc2
demonstrated
our
ability
to
use
the
nurse
hpc
resources
for
the
montana
production
and
processing,
and
mdc
3
gave
us
even
more
confidence
into
our
infrastructure
software
and
with
yeah
we
had
all
those
great
tests.
Then
then
we
can
just
build
up
and
get
ready
for
for
science,
so
stay
tuned,
because
2022
is
going
to
be
a
very
exciting
year
for
us,
because
data
are
coming
and
maybe
some
physics
results
are
coming
too.
A
B
Yeah,
thank
you
quinn.
That
was
great
yeah.
We're
happy
to
take
questions.
You
may
raise
your
hand
and
we'll
get
some
get
some
questions
in
there.
So
it's
pretty
yeah,
it's
gonna,
be
it's
gonna,
be
exciting
and
looking
forward
to
either
you
discovering
wimps
or
or
sending
the
theorists
driving
the
theories
crazy
by
by
putting
lower
limits
on
possible
unpossible
wins.
B
Right,
and
so
this
is
something
we're
seeing
more
and
more
and
more
of
from
various
experiments
and
facilities,
and
so
I'm
just
wondering
if
there
were
any
any
particular
challenges
that
you
had
to
overcome
at
nurse
that
we
could
that
we
could
work
on
or
if
there
are
any
opportunities
you
could
identify
for
things
that
we
could
do
to
make
your
life
easier.
When
trying
to
do
something
like
this.
A
So
something
that
made
our
life
way
easier
was
having
spin
available.
I
guess
for
for
lz
that
was
like
a
huge
to
have
spin,
because
we
are
now
using
it
a
lot
for
many
many
services,
especially
for
poppy
square,
but
not
only.
For
example,
we
have
our
offline
event,
viewer
that
is
deployed
on
spin
and
having
this
austere.net
available.
B
B
C
Yeah
hi
yeah
yeah
great
talk
quentin,
so
I
have
a
sort
of
related
question
which
is
just
about
whether
these
services
can
be
reused.
I
probably
shouldn't
know
the
answer
to
this,
but
so
the
things
you
built
you
know
can
they
be
used
for
other
hepa
experiments
or
even
other
domains
like
the
spin
services,
specifically.
A
I
don't
see
any
any
reasons
why
we
couldn't
use
this
for
anything.
I
don't
know
if
you
remember
this,
but
I
briefly
mentioned
that
p
squared
has
been
used
by
diabetes
quite
a
lot
in
the
past,
and
we
are
here
reading
this
piece
of
software
from
from
diabetes
yeah.
So
we
we
used
something
that
was
already
already
built:
okay,
but
yeah.
The
main
pst
is
this
process
definition
as
soon
as
the
script
is
good
and
working
for
you,
you'll
you're,
good
to.