►
From YouTube: SimPEG meeting July 8, 2020
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
B
That's
a
good
question:
I
always
had
checked.
My
calendars,
I
would
have
had
done
last
week
last
week
were
a
lot
of
submissions
with
two
papers
and
then
the
one
we're
working
on,
we
hope
to
make
their
progress
and
the
rest
I'm
still
deeply
involved
in
this
integrating
my
code
and
the
gradient
into
this
consortium
joint
inversion
framework,
so
we're
getting
there,
but
not
there
and
I
hope
sometimes
in
the
future.
B
I
can
show
or
share
some
of
the
other
work
and
they're
happy
to
show
some
of
it,
but
that's
all
below
company
barriers
at
the
moment.
Yeah.
That's
mostly
love
today.
I
think
I
really
have
to
look
at
those
maps.
I
think
so
lamented
them
in
simple
right
at
those
Maps,
because
today
and
yes,
they
actually
had
the
issues
of
the
gradient.
You
get
it
out
for
conductivities,
but
then
you
would
like
to
have
it
for
resistivities
her
vice
versa,
or
you
would
like
to
have
it
on
a
log
scale
and
I.
B
C
B
D
I
just
got
a
kind
of
quick
question,
so
the
problem
that
you
are
dealing
with
is
like
with
a
lot
of
snoring
problem.
It's
like
so
big
and
like
if
you
have
sort
of
good
pathway
that
you
think,
how
are
you
going
to
invert
and
even
simulate
those
big
data
like
a
big
sort
of
simulation
and
inversion.
So.
B
These
the
consortium,
the
inversion
framework
they
have-
is
a
structural
inversion.
So
there's
just
a
few
layers
blocks.
Basically
that
can
vary
the
property,
so
you
have
a
couple
of
layers,
but
then
you
can
put
in
you
can
put
in
blobs
like
say
squares
or
arbitrary,
shaped
blobs,
and
then
it
in
words
for
the
values
of
these
structures,
but
it
also
can
shift
the
boundaries
of
those.
D
Still
I
can
only
all
of
those
issues
that
you've
made
for
Cynthia
Wendy
and
then
jumping
into
that
refactoring
is
law.
My
to-do
list
for
free
talk,
but
yes,
spit
busy
these
days
here.
We're
sort
of
wrapping
up
this
project.
Yeah
and
I
am
mostly
working
on
to
you
statistics
in
these
days,
so
yeah
I'm,
using
like
I,
look
they're.
D
What's
called
s
gems,
the
Stanford
that
develop
software
mostly
GUI
based,
but
you
can
still
do
a
scripting
and
then
there's
a
Python
again,
so
you
can
do
a
bit
of
I
thought
scripting
as
well
there,
but
yeah,
no,
not
very
exciting.
It's
got
a
fun
but
I'm
just
in
time.
I
think
at
some
point,
it'll
be
nice
to
have
some
juice
statistical
plugin
that
you
can
readily
generate
some
realistic
images.
D
A
it's
not
on
the
github,
it's
on
the
SourceForge,
so
it's
a
it's!
Basically
an
executable
and
I.
Think
if
you
want
you
can
compile,
but
I
think
it's
written
in
either
Fortran
or
C++.
Okay
and
you
have
a
GUI.
What
is
it
called
s?
Gems,
sgem
s,
so
the
base
package
is
a
GS
slip,
but
I
think
they
developed
further
with
more
sophisticated
algorithm,
I.
A
Know
if
you've
seen
this
isn't
a
related
note.
I'll
drop
this
into
the
notes:
geo
stats
guy
from
a
UT
Austin.
My
whole
approach
he's
got
a
lot
of
geo
stats
code
on
github
I've
flipped
through
it
I
haven't
gone
to
gone
into
things
in
detail,
but
there
might
also
be
some
interesting
things.
I
don't
know.
I
did.
E
A
D
So,
but
the
the
bottleneck
actually
a
lot
of
3d
like
in
once
you
move
on
3d,
it's
actually
quite
expensive
and
what
that's
cool?
That's.
What
actually
has
seems
good
about
it.
It's
quite
fast.
So,
even
if
you
go
out
like
a
million
cells,
it's
a
a
durable
and
you
can
actually
run
3d
simulations
like
what
they
meant
just
simulations
there,
and
they
got
a
bit
more
sophisticated
algorithm
like
rather
than
just
simple
kriging,
more
simple,
so
yeah.
D
C
D
Some
online
tools
and
stuff,
so
yes
on
the
radar
but
I
haven't
worked
on
it,
but
I
actually
do
have
a
time
between
August
and
October.
So
I
had
a
chat
with
my
boss
and
she's,
pretty
keen
on
developing
open
source
and
packaging
up
what
we
have
done
here
so
between
August
and
October
I
can
talk
potentially
used
my
50%
of
my
time
to
do
that
so
yeah
I.
F
So
yeah,
you
should
be
looking
at
the
symposium
1d
repository
I,
put
the
issues
in
there
and
I
kind
of
put
four
main
issues,
at
least
to
start.
So
there
was
a
couple
of
unstable
solutions
or
solutions
that
we
didn't
have,
for
you
know
the
horizontal
loop.
We
could
only
do
it
on
the
on
the
axis
and
for
the
vertical
magnetic
dipole.
F
We
could
not
put
it
on
the
axis,
so
we
have
a
couple
of
solutions
that
we'd
like
to
have,
but
I
think
maybe
the
biggest
issue
and
something
that
soggy
would
be
the
most
useful
is
on
the
waveforms
so
really
just
deciding
where
it
lives.
Right
now
we
have
all
the
stuff
pertaining
to
waveforms
kind
of
in
several
places
for
simple
waveforms
like
if
it
was
a
step
off
or
a
square
wave,
there
would
be.
F
There
would
just
be
an
independent
function
that
would
let
you
put
in
x
and
then
output
the
currents
then
for
some
known
waveforms
like
the
sky,
t'aime
2015,
it
would,
you
would
basically
call
it
and
it
would
just
give
you
back
some
structure
with
all
the
information
in
it,
and
then
there
are
some
known
systems.
Basically,
there
are
some
things
where
you
can
say:
I
want
to
put
in
these
locations.
It's
this
type
of
survey
and
it
outputs
a
completely
finished
survey,
object
which
we
don't
really
use
that
structure
anymore.
F
So
there's
just
a
lot
of
different
ways
of
kind
of
defining,
either
waveforms
or
particular
time
domain
systems,
and
the
big
question
is
just:
where
do
we
want
the
waveforms
to
live
when
we
define
a
waveform?
Do
we
want
this
to
be
maybe
a
property
of
the
source
object?
Do
we
want
to
define
a
waveform
object
like
we
do
with
the
TE
problem
and
then
say:
each
source
needs
to
be
attached
to
a
waveform
object
yeah.
F
D
General
imprint,
I
think
the
reason
why
I
sort
of
it's
just
a
history
of
like
my
trials.
Oh
there's,
a
simple
parametric
waveform
that
you
can
potentially
use
but
I.
Think
in
general,
each
survey
has
a
different
waveform,
even
though
you
use
less
say
exactly
same
victim
system
or
skyping
system.
They
have
a
slightly
different
waveform,
so
I
think
as
a
sort
of
a
learning
and
educational
perspective
store.
D
Having
a
couple
of
like
representative
waveform
is
useful
and
I
think
that's
probably
what
with
that,
what
what
we
have
that
can
serve,
but
in
practice
I
think
it'll
change.
So
what's
actually
important
is
the
current
like
an
actual
waveform,
so
time
time,
discretization
and
the
waveform.
So
that's
that
would
be
my
first
comment.
So
I
think
we
can
clean
up
those.
D
F
These
waveforms,
these,
like,
let's
see
like
company-specific
waveforms,
just
a
laundry
list
of
properties
associated
with
it,
and
so
then
your
about
making
that
all
of
those
properties
of
the
source
object,
so
I
kind
of
thought,
maybe
doing
it
the
way
we
do
for
time,
domain
e/m,
where
we
we
create
a
waveform
object
and
all
of
those
properties
are
associated
with
the
waveform.
That
might
be
the
most
simple
and
organized
way
to
do
this.
C
F
And
there
was
there
was
some
stuff
that
I
guess:
I,
don't
understand
the
background
theory
when
you're
talking
about
the
DOOL
there
there
was
some
some.
There
was
yes,
some
of
that
stuff
that
I
couldn't
find
a
reference
or
documentation
on
so
I
didn't
really
know
how
it's
being
implemented.
So
that
would
be
a
good
thing
to
help
me
out
right
right.
D
D
So
that's
what
they
usually
men
do
all
moment
and
in
terms
of
impact.
What
you
actually
need
is
just
a
step
response
and
then,
once
you
got
a
step
response,
you
convolve
with
the
different
wave
forms
so
computationally,
it's
actually
not
expensive,
so
the
cost
is
actually
convolution.
So
everything
is
really
computed.
All
the
frequency
components
and
you
generate
the
step
response.
First
then
convolve
it
with
the
two
different
waveforms.
F
F
A
One
question
with
all
of
this:
instead
of
means
I'm
guessing
that
some
of
the
waveforms
that
are
in
the
em1
decode
right
now
might
be
basically
a
duplication
of.
What's
in
simply
are
there
places
where
PM
1d
can
actually
just
import
directly
the
waveforms
that
are
in
simp
ik
and
can
we
think
about?
Maybe
we
don't
need
to
do
this
right
away,
but
waveforms
that
are
in
iam
1d
that
are
not
insipid?
F
I
I
thought
about
that.
A
little
bit
I
didn't
get
too
deep
into
it
and
I'm
sure
there's
some
stuff
that
can
get
reused,
but
for
the
3d
e/m
problem
you
define
time
stepping,
whereas
this
you
actually
define
the
x
and
then
do
a
convolution
and
so
I
didn't
know
how
much
stuff
can
be
brought
over
I'm
sure
there's
functions
like
the
function.
That
would
let
you
create
a
ramp
off
or
something
I'm
sure
that
function
exists
somewhere
else
in
simp,
ik
I'm
sure
this
stuff
that
we
can
reuse
yeah.
A
A
C
Still
sampled
functions
like
the
waveforms
as
a
player.
Is
that
sampled
functions
so
it
doesn't
like
it's
the
ones
in
simpler
sampled
functions,
just
like
these
ones
are
sampled
functions,
so
I
feel
like
they
should
probably
just
all
live
it
simple,
because
sympathy
m1d
already
depends
on
symphony.
Does
it
have
idea.
F
In
in
some
ways,
yes,
then
some
ways,
no
I,
guess
my
plan
of
attack
for
this
was,
let's
you
know,
take
the
em1
D,
let's
structure
it
like
simpang
is
structured
and
it's
okay,
that
there
that
there's
some
stuff
operating
independently
or
sorry
that
you
know
that
might
be
duplicated
that
exists
in
Simpa
and
then,
when
we
move
it
over
then
make
decisions
as
to
not
making
sure
there's,
there's
duplicate,
stuff
I
thought.
Maybe
that
would
be
a
step
that
happens
later.
I.
F
And
so
I
think
there
was
that
and
then
was
there.
Another
thing
that's
worth
mentioning
at
this
point:
yeah
I
mean
I,
had
some
comments
on
the
highest
level
organization,
but
I
think
we
can
fix
that
sort
of
afterwards.
A
lot
of
it
was
yeah,
making
this
waveform
class
and
trying
to
move
supporting
functions
out
of
the
highest
level.
F
So
there's
a
lot
of
kind
of
if
statements,
if
it's
this
type
of
source
or
etc,
and
it
would
be
nice
to
just
have
a
function
defined
in
there.
That
says
you
know
compute
compute,
the
Hankel
transformer
to
compute
the
integral
and
have
that
happening
somewhere
else.
I
think
the
highest
level
could
be
cleaned
up
so
that
it's
much
easier
to
read,
and
then
it's
calling
supporting
functions
as
as
opposed
to
putting
all
of
that
in
the
simulation
class,
but
yeah.
That's
this
I've
got
via
four
issues
there.
G
Yeah
so
went
back,
so
among
other
things,
is
they
I've
I'm?
Finally,
looking
at
the
simulation
class,
I
was
really
not
involved
in
that
then
I
was
keeping
keeping
my
code
as
it
was
for
all
the
writing,
etc,
so
I'm,
so
I'm,
so
I'm
trying
to
merge
my
branch
with
the
simulation
with
master
basically
at
the
moment,
so
should
be.
Okay,
like
the
I
just
have
to.
G
There
is
just
like
when,
when
the
file
changed
name
that
sometime
it's
harder
to
track
things,
so
that's
what
it
just
requires
me
to
do
it
manually
kind
of
so
what
I'm
working
on
that,
and
actually
on
that
note,
I
have
a
question
for
you.
Joe
is
how
far
you
want
that
suze
decomposition.
You
were
working
on
for
DC.
Is
that
already
on
master
or
is
that
somewhere
else?
Where.
C
D
C
C
G
Okay,
yeah
things
because
I
was
like
I
think
one
of
the
next
thing
I
want
to
do
is
like
to
go
back
to
a
DC
IP
data
set
I've
been
I,
was
working
on
before
and
kind
of
left
on
the
side.
So
I
will
it's
a
big
data
set.
So
that's!
So
that's
why
I
was
wondering
but
I.
It's
true
that
it's
probably
a
pole
dipole
for
mining,
so
my
not
be
that
full,
but
I.
D
G
G
H
Been
sticking
with
surface
waiting,
mostly
yeah,
that's
a
pretty
effective
job
I've
been
playing
around
with
the
sensitivity
depth
waiting,
but
it's
the
problems
are
so
huge.
It's
not
really
like
time
efficient,
so
I
think
it
has
like
so
many
inversions
all
by
the
time
we
calculate
it.
Yeah,
oh
and
then
we've
run
out
of
memory
or
something
when
we're
trying
to
calculate.
B
H
Party
so
party,
so
I
kind
of
inquired
a
bit
there
and
like
for
me
to
get
a
license
through
my
company,
that's
about
3,000,
Swiss
francs,
just
and
that's
locked
to
one
computer
yeah.
It's
not
free!
You
can't
just
take
it
off
the
computer
and
put
it
on
another.
One.
Commercial
license
is
like
$30,000,
but
it
yeah.
It
works
on
Windows
and
everything
probably
pretty
easy,
whereas
petsy
and
mumps,
we
might
have
problems
if
Windows.
A
H
Want
to
use
it
I've
convinced
my
boss
to
let
me
turn
one
of
my
computers
into
a
Linux
machine.
So
I
don't
know,
I
think
with
petsy.
You
got
so
many
choices
like
you,
don't
have
think
it
has
mumps
and
everything
else,
so
maybe
it'd
be
worth
the
time
to
put
into
petsy
so
that
there's
you
have
choices.
Where
is
it
that
we
just
go
with
moms?
It's
just.
We
have
mom's
kind
of
thing,
which
might
be
the
one
that
like
reading
about
it,
that
people
do
love
it.
Really.
It's
pretty
good
have.
H
H
H
Solvers
as
direct
solvers,
maybe
it's
worth
the
time
to
do
that
and
you
can
there's
lots
of
cool
ways
to
construct
the
matrices
like
we
can
utilize,
the
sparse
or
assisity.
We
don't
have
to
store
any
of
that
information.
Yeah.
It's
really
well-rounded
project.
The
documentation
that
I've
been
through
I
just
might
be
might
be
a
little
bit
time
consuming
needs
to
think
riding
all
the
rappers
John.
H
Pantsy
and
as
I
was
looking
at
yeah
I
was
just
gonna,
build
it
right,
yeah.
Just
from
when
dieter
started
that
conversation,
it
seems
like
yeah,
everybody
goes
from
the
roots.
They
mentioned
some
of
the
Python
wrappers,
but
I
think.
If
we
want
full
control
over
what
versions
are
being
used,
we
should
just
build
it
right
from
the
bottom
yeah
yeah
yeah.
D
H
E
H
H
H
B
A
C
C
The
rappers
that
we
can
build
our
would
work
for
both
versions,
like
Intel's
version
and
like
the
new
party,
so
they
have
the
same
handles
and
everything
yeah
as
far
as
pets
II
like
I,
would
really
like
to
look
into
it.
Mumps
I
know
there's
like
mum
spot
like
a
Python
interface
for
a
month,
months
already
or
just
depending
on
what
quite
sure,
it's
I
think
it
uses
the
MPI
version.
Yeah.
C
But
I
wasn't
really
just
concerned
about
building
box
because,
like
I
said
it's
on
kind
of
forge
I
think
it's
like
I
just
looked,
it
was
like
version
5.1
or
something
so
then
they
have
it
for
all
three
operating
systems.
I
think
we
should
be
okay
to
depend
on
that.
Not
have
to
worry
about
building
amongst
ourselves.
A
C
Yeah
version
3.13
is
I'm
kind
of
Forge.
Okay,
like
we're
fine
to
depend
on
any
of
these
libraries
and
just
go
from
there.
I
I
think.
The
next
thing
is:
okay,
don't
worry
about
building
those
libraries,
so
it's
like
at
the
existing
Python
wrappers
for
them
and
see
if
they
are
fit
for
what
we
want
to
do.
Okay,
if
they
are,
then
we
could
just
use
those
directly
and
PI
that
solve
or
not
worry
about
it.
C
C
E
C
A
H
C
So
since
last
week
I've
been
hanging
out-
and
let,
though
is
my
lot
of
people
came
to
the
house
and
it's
a
very
small
house,
so
it's
been
kind
of
hectic
here.
I've
done
a
few
updates
to
dis
critize
over
I
got
that
it
all
set
up
to
automatically
deploy
everywhere
now
for
a
new
update,
so
it'll
go
all
the
way
to
kinda
forward
like
you're
talking
about
Peter
how
to
auto
merge
feature
on
there.
C
It's
got
all
the
way
through
there
I
mean
it's
just
a
small
little
addition
to
the
tree
mesh
plot
marriage
so
that
it'll
respect
the
norm
argument.
So
we
can
do
things
like
plot
log
norms
on
this
kinda
tree
mesh,
reasonable
color,
varies
article
as
part
of
that.
I
also
went
through
and
changed
all
of
the
keyword
arguments
to
the
plot
image
and
plot
the
plotting
functions
for
discretize
to
be
more
to
be
in
line
with
Pepe
conventions.
C
C
C
The
modules
and
it
a
little
bit
as
far
as
there's
a
lot
of
like
utility
things
that
are
exposed
at
the
base
level
discretize
that
are
not
needed
to
be
so
just
kind
of
hide
those
inside
folders
things
that
we
don't
really
need
them
meet
people
to
be
looking
at
the
source
code.
It
shouldn't
change,
how
anyone
interacts
with
it,
because
those
tree
mesh
and
the
tree
matters
tensor
they're
all
gonna
be
imported
like
that.
C
A
And
one
thing
we
can
even
look
at
doing
I
like
looked
at
this
and
then
looked
at
how
much
needed
to
change
and
then
was
like,
but
doing
flake,
eight
and
things
like
that,
so
that
moving
people
like
users
gets
developers,
get
warnings
and
errors
when
we'd
use.
That's
not
happy
so.
C
C
Was
thinking
also
thinking
about
splitting
up
the
docs
on
disc
retires
a
little
bit
more
to
be
so?
Ok,
I
get
kind
of
confusing
I
like
all
the
same
meshes,
and
things
are
on
the
same
page
like
HTML
page
right.
It's
one
long
list,
I
kind
of
like
more
like
the
numpy
style.
Things
where
you
can.
Click
on
everything
has
its
own
page,
HTML
page,
so
I
think
you
should
move
to
that
kind
of
Docs
structure.
At
least
it's
not
gonna.
G
B
H
F
C
C
John
I'm
sure,
if
you
I,
would
appreciate
if
you
could
come
drop
off
the.
If
you
have
some
homebrew
I'd
love
to
try
it
you're.
E
C
C
For
right,
when
you're
looking
for
a
cell,
giving
a
point
so
I
guess,
if
the
n
log
n
right
that
kind
of
operation
complexity
roughly
yeah,
so
it
shouldn't
be
awful,
I
mean
that's,
not
a
super
bad
complexity
level
as
well.
Right
here,
as
far
as
the
interaction
between
meshes
be
interacting
with
the
tensor
Russia,
is
that
one's
already
really
fast,
but
to
between
two
tents
and
Russia's,
because
you
can
decompose
into
each
dimension
right.
C
D
D
C
C
A
I've
got
a
couple
updates
as
well,
so
yeah
been
working
on
this
paper
and
that's
been
definitely
raising
things
about
solvers,
but
also
some
cool
things
and
I'll
share
this
well.
I'll
drop
the
link
for
the
github,
repo
or
I.
Think
Peters
already
put
it
in
the
slack
so
I
do.
We've
got
a
couple
notebooks
there
for
the
block
model
and
the
layered
model.
A
A
C
A
D
A
Once
you
start
fiddling
and
really
driving
it
down,
it
tends
to
be
sort
of
across
the
the
survey
line.
There's
not
like
it's
not
clear
that,
like
I'm,
really
breaking
down
near
the
edges,
or
it's
perhaps
a
bit
more
near
the
source,
but
I'll
send
a
couple
examples
around
in
in
slack
afterwards,
because
I
don't
have
it
organized
off
the
top
of
my
head.
A
A
E
D
One
thing
that
probably
any
worthwhile
to
check
is
how
other
people
hender
like
a
kind
of
there's
a
singularity
and
when
you
go
to
the
low
frequency.
So
if
you
like
have
a
very
low
frequency,
your
Iomega
tremor
is
gone,
then
that
that
coral
matrix
is
actually
singular.
So
one
way
that
you
BC
approach
was
using
the
vector
potential
like
composing,
that
into
a5
solution
and
solve
it.
D
B
D
A
I
mean
in
this
case,
like
I'm,
suspecting
it's
much
more
I
mean
cuz
weirdly,
it's
converging
to
the
right
solution,
I
mean
if
you
just
eyeball
this.
This
is
we're
doing
not
a
bad
job.
It's
now
coming
down
to
sort
of
mostly
discretization
and
probably
interpolation
errors
like
I
suspect.
Actually,
if
we
had
a
second
order,
interpolation
or
something
that's.
D
A
B
A
A
A
B
D
A
C
E
D
B
H
A
Well,
yeah
I
mean
so
at
this
point,
I'm
not
like
unhappy
with
these
results.
It
would
be
nice
to
drive
them
down
a
bit
further,
but
it's
it's
comparable
and
like
yes,
impetus
slightly
above
but
yeah
so,
but
if
anyone
wants
to
mess
around
with
meshes
and
and
things
like
that,
you're
you're
welcome
to
play
around
with
this.
These
notebooks
yeah.
A
And
then
the
first
thing
I
wanted
to
share
with
folks
is
I
had
a
great
conversation
with
Peter
leav
earlier
this
week,
and
he
is
starting
a
position
as
a
professor
in
the
math
department
at
Mount,
Allison,
and
so
he's
right
now
kind
of
just
exploring
open
source
space
and
trying
to
see
you
know
what
tools
does
he
want
to
really
start
getting
familiar
with
and
maybe
start
pointing
students
to
so
he's
been
checking
out
both
Simpa
and
PI
Gimli
and
he's
really
interested
in
unstructured
meshes
so
coming
from
sort
of
you
know
the
work
with
Colin,
a
lot
of
what
they've
done
is
unstructured
meshes,
and
so
one
of
the
things
I
wanted
to
put
out
is
that
I
think
that
that
actually
could
be
a
really
cool
opportunity
for
us
to
think
about
implementing
on
search
your
meshes
one,
but
then
also
to
actually
start
like
interoperating.
A
A
bit
with
my
kingly
is
even
if
we
don't
actually
like
tie
those
code
bases
together,
at
least
if
we
could
like
take
a
mesh
and
load
it
both
into
PI
gameplay
and
into
simple
I.
Think
that
would
be
really
cool
because
we
could
actually
start
sort
of
comparing
like
really
actually
comparing
inversion
results
and
things
like
that,
because
we
both
take
very
different
approaches
or
fairly
different
approaches,
but
being
able
to
be
on
the
the
same
mesh,
really
cool,
so
yeah,
so
I
think
Peter
might
be
able
to
join
us
next
week.
A
C
A
A
A
There's
some
stuff,
that's
been
there's
like
a
month
and
I
I
dropped
the
ball
and
nobody
else
spoke
up
on
that
comment
and
that's
that's
fine,
but
if
we
can
all
even
just
take
like
I,
don't
know
tackle
one
one
a
week
or
something
like
that.
That
would
be,
that
would
be
wonderful,
I
think
would
spread
would
spread.
The
load.
A
I
can
look,
I
think
there's
a
way
to
discourse
ping,
slack
I,
don't
know
how
to
do
that,
but
I
think
it
can
be
done
which
might
be
helpful
because
I
know
we're
all
sort
of
on
slack
and
tend
to
answer
questions
there.
So
if
there's
a
way
that
discourse
just
like
notifies
us
that
there's
a
new
question,
I
think
that
would
that
would
be
good.
So
all
your
little
check
that
I
would.