diff --git a/pytest.ini b/pytest.ini
index 5228471..b6406e2 100644
--- a/pytest.ini
+++ b/pytest.ini
@@ -1,12 +1,11 @@
 [pytest]
-# for now, running with with_my pytest tests/reddit.py
-# discover __init__,py as well
-# TODO also importing too much with it...
-# TODO FIXME wtf is this??
-python_files = reddit.py
+# discover files that don't follow test_ naming. Useful to keep tests along with the source code
+python_files = *.py
 addopts =
+  -rap
   --verbose
-
+  # TODO hmm, not sure... guess it makes sense considering all the ext modules..
+  --continue-on-collection-errors
   # otherwise it won't discover doctests
   # eh? importing too much
   # --doctest-modules
diff --git a/tests/README.org b/tests/README.org
deleted file mode 100644
index c1c8015..0000000
--- a/tests/README.org
+++ /dev/null
@@ -1,3 +0,0 @@
-At the moment, some tests here are specific to my data.
-
-If you know any good public data sources I could test against instead, please let me know!
diff --git a/tests/bluemaestro.py b/tests/bluemaestro.py
index 144e4f4..420dab8 100644
--- a/tests/bluemaestro.py
+++ b/tests/bluemaestro.py
@@ -1,11 +1,13 @@
 #!/usr/bin/env python3
 from pathlib import Path
-
 from more_itertools import one
 
+from .common import skip_if_not_karlicoss
+
 import pytest # type: ignore
 
 
+@skip_if_not_karlicoss
 def test() -> None:
     from my.bluemaestro import measurements
     res = list(measurements())
diff --git a/tests/commits.py b/tests/commits.py
index f49a34f..ffc7d66 100644
--- a/tests/commits.py
+++ b/tests/commits.py
@@ -1,7 +1,8 @@
+from .common import skip_if_not_karlicoss as pytestmark
+
 from more_itertools import ilen
-from my.coding.commits import commits
 
-
-def test():
+def test() -> None:
+    from my.coding.commits import commits
     all_commits = commits()
     assert ilen(all_commits) > 10
diff --git a/tests/common.py b/tests/common.py
new file mode 100644
index 0000000..ba5def3
--- /dev/null
+++ b/tests/common.py
@@ -0,0 +1,7 @@
+import os
+
+import pytest
+
+skip_if_not_karlicoss = pytest.mark.skipif(
+    'HPI_TESTS_KARLICOSS' not in os.environ, reason='test only works on @karlicoss data for now',
+)
diff --git a/tests/config.py b/tests/config.py
index 2496002..fdcdebf 100644
--- a/tests/config.py
+++ b/tests/config.py
@@ -18,10 +18,10 @@ def setup_notes_path(notes: Path) -> None:
 def test_dynamic_configuration(notes: Path) -> None:
     setup_notes_path(notes)
 
-    from my.body.weight import dataframe
-    weight_df = dataframe()
+    from my.body.weight import from_orgmode
+    weights = [0.0 if isinstance(x, Exception) else x.value for x in from_orgmode()]
 
-    assert list(weight_df['weight'].fillna(0.0)) == [
+    assert weights == [
         0.0,
         62.0,
         0.0,
diff --git a/tests/emfit.py b/tests/emfit.py
index cd88945..8a779e4 100644
--- a/tests/emfit.py
+++ b/tests/emfit.py
@@ -1,7 +1,10 @@
-from my.emfit import datas
+from .common import skip_if_not_karlicoss as pytestmark
 
 
 def test() -> None:
+    from my.emfit import datas
+    # TODO this should be implement via sanity checks/stat instead?
+    # the same code will be used for tests & for user reports
     ds = [x for x in datas() if not isinstance(x, Exception)]
     for d in ds:
         assert d.start.tzinfo is not None
@@ -10,7 +13,10 @@ def test() -> None:
         assert d.sleep_end.tzinfo is not None
 
 
+from .common import skip_if_not_karlicoss
+@skip_if_not_karlicoss
 def test_tz() -> None:
+    from my.emfit import datas
     # TODO check errors too?
     ds = [x for x in datas() if not isinstance(x, Exception)]
 
diff --git a/tests/foursquare.py b/tests/foursquare.py
index ae9612e..a9169ff 100644
--- a/tests/foursquare.py
+++ b/tests/foursquare.py
@@ -1,6 +1,7 @@
-from my.foursquare import get_checkins
+from .common import skip_if_not_karlicoss as pytestmark
 
-def test_checkins():
+def test_checkins() -> None:
+    from my.foursquare import get_checkins
     # todo reuse stats?
     checkins = get_checkins()
     assert len(checkins) > 100
diff --git a/tests/github.py b/tests/github.py
index f4ca4b5..5fb5fb9 100644
--- a/tests/github.py
+++ b/tests/github.py
@@ -1,16 +1,15 @@
-#!/usr/bin/env python3
+from .common import skip_if_not_karlicoss as pytestmark
+
 from more_itertools import ilen
-
-from my.coding.github import get_events
-
 # todo test against stats? not sure.. maybe both
 
-def test_gdpr():
+def test_gdpr() -> None:
     import my.github.gdpr as gdpr
     assert ilen(gdpr.events()) > 100
 
 
-def test():
+def test() -> None:
+    from my.coding.github import get_events
     events = get_events()
     assert ilen(events) > 100
     for e in events:
diff --git a/tests/hypothesis.py b/tests/hypothesis.py
index 9ef8880..f5ee99e 100644
--- a/tests/hypothesis.py
+++ b/tests/hypothesis.py
@@ -1,7 +1,6 @@
-#!/usr/bin/env python3
+from .common import skip_if_not_karlicoss as pytestmark
 
-from my.hypothesis import pages, highlights
-
-def test():
+def test() -> None:
+    from my.hypothesis import pages, highlights
     assert len(list(pages())) > 10
     assert len(list(highlights())) > 10
diff --git a/tests/instapaper.py b/tests/instapaper.py
index ae685e6..153a716 100644
--- a/tests/instapaper.py
+++ b/tests/instapaper.py
@@ -1,5 +1,6 @@
-from my.instapaper import pages
+from .common import skip_if_not_karlicoss as pytestmark
 
-
-def test_pages():
+def test_pages() -> None:
+    # TODO ugh. need lazy import to simplify testing?
+    from my.instapaper import pages
     assert len(list(pages())) > 3
diff --git a/tests/jawbone.py b/tests/jawbone.py
index b6630a4..c53459d 100644
--- a/tests/jawbone.py
+++ b/tests/jawbone.py
@@ -1,11 +1,10 @@
-# TODO depends on my private data... move somewhere, and exclude from CI somehow?
+from .common import skip_if_not_karlicoss as pytestmark
 
 from datetime import date, time
 
-from my.jawbone import sleeps_by_date
-
 # todo private test.. move away
-def test_tz():
+def test_tz() -> None:
+    from my.jawbone import sleeps_by_date
     sleeps = sleeps_by_date()
     for s in sleeps.values():
         assert s.sleep_start.tzinfo is not None
diff --git a/tests/lastfm.py b/tests/lastfm.py
index a93fc4b..43e8f41 100644
--- a/tests/lastfm.py
+++ b/tests/lastfm.py
@@ -1,13 +1,16 @@
+from .common import skip_if_not_karlicoss as pytestmark
+
+# todo maybe belongs to common
 from more_itertools import ilen
 
-from my.lastfm import scrobbles
 
-
-def test():
+def test() -> None:
+    from my.lastfm import scrobbles
     assert ilen(scrobbles()) > 1000
 
 
-def test_datetime_ascending():
+def test_datetime_ascending() -> None:
+    from my.lastfm import scrobbles
     from more_itertools import pairwise
     for a, b in pairwise(scrobbles()):
         assert a.dt <= b.dt
diff --git a/tests/orgmode.py b/tests/orgmode.py
index aa30114..d213a5e 100644
--- a/tests/orgmode.py
+++ b/tests/orgmode.py
@@ -1,6 +1,9 @@
 from my import orgmode
 from my.core.orgmode import collect
 
+from .common import skip_if_not_karlicoss
+
+@skip_if_not_karlicoss
 def test() -> None:
     # meh
     results = list(orgmode.query().collect_all(lambda n: [n] if 'python' in n.tags else []))
diff --git a/tests/reddit.py b/tests/reddit.py
index 6a660ea..964d0f5 100644
--- a/tests/reddit.py
+++ b/tests/reddit.py
@@ -1,3 +1,5 @@
+from .common import skip_if_not_karlicoss as pytestmark
+
 from datetime import datetime
 import pytz
 
diff --git a/tests/rtm.py b/tests/rtm.py
index efb9a4f..93378b6 100644
--- a/tests/rtm.py
+++ b/tests/rtm.py
@@ -1,6 +1,7 @@
-from my.rtm import all_tasks
+from .common import skip_if_not_karlicoss as pytestmark
 
 
-def test():
+def test() -> None:
+    from my.rtm import all_tasks
     tasks = all_tasks()
     assert len([t for t in tasks if 'gluons' in t.title]) > 0
diff --git a/tests/smscalls.py b/tests/smscalls.py
index 3303e1c..51150f0 100644
--- a/tests/smscalls.py
+++ b/tests/smscalls.py
@@ -1,7 +1,9 @@
-from my.smscalls import calls, messages
+from .common import skip_if_not_karlicoss as pytestmark
+# TODO maybe instead detect if it has any data at all
+# if none, then skip the test, say that user doesn't have any data?
 
-
-# TODO that's a pretty dumb test; perhaps can be generic..
-def test():
+# TODO implement via stat?
+def test() -> None:
+    from my.smscalls import calls, messages
     assert len(list(calls())) > 10
     assert len(list(messages())) > 10
diff --git a/tests/takeout.py b/tests/takeout.py
index 2cfb7a3..bc746c2 100644
--- a/tests/takeout.py
+++ b/tests/takeout.py
@@ -8,12 +8,9 @@ from my.google.takeout.html import read_html
 from my.google.takeout.paths import get_last_takeout
 
 
-def ilen(it):
-    # TODO more_itertools?
-    return len(list(it))
+from more_itertools import ilen
 
-
-def test_location_perf():
+def test_location_perf() -> None:
     # 2.80 s for 10 iterations and 10K points
     # TODO try switching to jq and see how it goes? not sure..
     print(ilen(islice(LT.iter_locations(), 0, 10000)))
@@ -30,7 +27,7 @@ import pytest # type: ignore
         'My Activity/Search/MyActivity.html',
     ]
 )
-def test_parser(path: str):
+def test_parser(path: str) -> None:
     path = 'Takeout/' + path
     tpath = get_last_takeout(path=path)
     assert tpath is not None
@@ -39,7 +36,7 @@ def test_parser(path: str):
     print(len(results))
 
 
-def test_myactivity_search():
+def test_myactivity_search() -> None:
     path = 'Takeout/My Activity/Search/MyActivity.html'
     tpath = get_last_takeout(path=path)
     assert tpath is not None
diff --git a/tests/tweets.py b/tests/tweets.py
index 3c5aa4e..fefc24e 100644
--- a/tests/tweets.py
+++ b/tests/tweets.py
@@ -1,12 +1,16 @@
+from .common import skip_if_not_karlicoss as pytestmark
+# todo current test doesn't depend on data, in principle...
+# should make lazy loading the default..
+
+
 from datetime import datetime
 import json
 
 import pytz
 
-from my.twitter.archive import Tweet
 
-
-def test_tweet():
+def test_tweet() -> None:
+    from my.twitter.archive import Tweet
     raw = """
  {
   "retweeted" : false,
diff --git a/tests/youtube.py b/tests/youtube.py
index b8c1aa8..d514061 100644
--- a/tests/youtube.py
+++ b/tests/youtube.py
@@ -1,11 +1,14 @@
 # TODO move elsewhere?
 # these tests would only make sense with some existing data? although some of them would work for everyone..
 # not sure what's a good way of handling this..
+from .common import skip_if_not_karlicoss as pytestmark
 
-from my.media.youtube import get_watched, Watched
+# TODO ugh. if i uncomment this here (on top level), then this test vvv fails
+# from my.media.youtube import get_watched, Watched
+# HPI_TESTS_KARLICOSS=true pytest -raps tests/tz.py tests/youtube.py
 
-
-def test():
+def test() -> None:
+    from my.media.youtube import get_watched, Watched
     watched = list(get_watched())
     assert len(watched) > 1000
 
diff --git a/tox.ini b/tox.ini
index 1a72477..80e5382 100644
--- a/tox.ini
+++ b/tox.ini
@@ -8,6 +8,8 @@ passenv = CI CI_*
 # deliberately set to nonexistent path to check the fallback logic
 setenv = MY_CONFIG = nonexistent
 commands =
+    # TODO core & modules should be tested separately?
+
     pip install -e .[testing]
     # python -m pytest {posargs}
 
@@ -23,19 +25,12 @@ commands =
     # my.body.weight dep
     pip install orgparse
 
-    python3 -m pytest                              \
-        tests/cli.py                               \
-        tests/core.py                              \
-        tests/misc.py                              \
-        tests/get_files.py                         \
-        tests/config.py::test_environment_variable \
-        tests/demo.py                              \
-        tests/bluemaestro.py                       \
-        tests/location.py                          \
-        tests/tz.py                                \
-        tests/calendar.py                          \
-        tests/config.py                            \
-        {posargs}
+    python3 -m pytest tests \
+            # ignore some tests which might take a while to run on ci..
+            --ignore tests/takeout.py        \
+            --ignore tests/extra/polar.py    \
+            --ignore tests/pdfs/test_pdfs.py \
+            {posargs}
 
 
 [testenv:demo]