diff --git a/common/modules/help-cs.org b/common/modules/help-cs.org
index b944ba7..a211f4c 100644
--- a/common/modules/help-cs.org
+++ b/common/modules/help-cs.org
@@ -1,135 +1,151 @@
 #+COLUMNS: %40ITEM %10BEAMER_env(Env) %9BEAMER_envargs(Env Args) %10BEAMER_act(Act) %4BEAMER_col(Col) %10BEAMER_extra(Extra) %8BEAMER_opt(Opt)
 #+INCLUDE: "prelude.org" :minlevel 1
 #
 # Research challenges
 #
 * Selected research challenges: building the archive
   :PROPERTIES:
   :CUSTOM_ID: main
   :END:
 ** Metadata alignment
+  :PROPERTIES:
+  :CUSTOM_ID: metadata
+  :END:
 *** Many concepts related to source code
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     - project, archive, source, language, licence, bts, mailing list, ...
     - developer, committer, author, architect, ...
 *** Many existing ontologies, catalogs
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     - DOAP, FOAF, Appstream, schema.org, ADMS.SW, ...
 # mostly manual
     - Freecode (40.000+), Plume (400+), Debian (25.000+), FramaSoft (1500+),    OpenHub (670.000+), ...
 # OpenHub is mostly automatic
 # Wikipedia ?
 *** Challenge : scale up metadata to millions of projects
      :PROPERTIES:
      :BEAMER_act: +-
      :END:
     - /reconcile/ existing ontologies
     - /link/ and /check/ existing catalogs with Software Heritage
     - handle /inconsistent data/ and /provenance information/
     - synthesise missing information (machine learning)
 
 ** Software phylogenetics
+  :PROPERTIES:
+  :CUSTOM_ID: phylogenetics
+  :END:
 *** The Software Diaspora
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     - Code often /migrates/ across projects: forks, copy-paste
     - Code gets /cloned/: reuse, language limitations, code smells
     - Projects /migrate/ across forges: fashion, functionality
     - Projects get /cloned/: mirrors, packages
 *** Challenge: tracing software evolution across billions of files
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     - rebuild the history of software artefacts
     - identify code origins
     - spot code clones
     - build project impact graphs
 ** Distributed infrastructure
+  :PROPERTIES:
+  :CUSTOM_ID: infrastructure
+  :END:
 *** The software graph
     - files
     - directories
     - commits
     - projects
     all de-duplicated in Software Heritage
 *** Challenge: design efficient architectures and algorithms
     - replication and availability
     - navigation
     - what happens to CAP? (updates are nondestructive!)
     - query
+** Efficient updates
+*** Many sources
+    - GitHub (tens of millions of projects)
+    - Gitlab (tens of thousands of instances)
+*** Frequent changes
+    - millions of commits/push per day
+*** Keeping up
+    - what is the most efficient way to keep up with this stream of data?
 * Selected research challenges: using the archive
   :PROPERTIES:
   :CUSTOM_ID: using
   :END:
 ** Code search: an old problem
 *** A natural need
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     - Find the definition of a function/class/procedure/type/structure
     - Search examples of code usage in an archive of source code
     - you name it...
 *** A natural approach
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     - Regular expressions
 *** We have all used /grep/ since the 1970's!
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     \hfill where is the challenge?
 ** Finding a needle in a haystack: size matters!
  How do we search in /millions/ of source code files?
 *** Google code search (open 2006, closed 2011)
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
 #+latex: see {\small \url{https://swtch.com/~rsc/regexp/regexp4.html}}
 #+latex: reborn in Debian {\small \url{http://codesearch.debian.net/}}
 *** how 
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     - build an inverted index of /trigrams/ from all source files
     - /map/ regexps to trigrams
     - /filter/ files that may match
     - run /grep/ on each file (using the "cloud")
 *** performance
     :PROPERTIES:
     :BEAMER_act: +-
     :END:
     \hfill scaled reasonably well up to /1 billion lines of codes/
 ** Challenge: scaling up code search
 *** What about /all the source code/ in the world?
      :PROPERTIES:
      :BEAMER_act: +-
      :END:
     Software Heritage is /two orders of magnitude/ bigger already
       - over /two billion/ unique source files
       - /hundreds/ of billions of LOCs
     We need new insight for handling this.
 *** Beyond regular expressions?
      :PROPERTIES:
      :BEAMER_act: +-
      :END:
     Advanced code search requires
      - language specific /patterns/
      - working on /abstract syntax trees/
     Regular expressions are a nice /swiss-army knife/ approximation, can we build a specific tool that scales?
     
 ** Software as Big Data = "Big Code"
 *** Remember the numbers
-    - 30 million repositories ingested (10M next in line)
-    - 700 million commits
-    - 3 billion unique source files / 200 TB of raw source code
-    ... and growing by the day!
+    - tes of millions repositories ingested
+    - over 1 billion commits
+    - billions of unique source files ... and growing by the day!
 *** Challenge: what can machines learn here?
     - programming patterns
     - developer skills
     - vulnerabilities
     - bugs and fixes