hi pasquale thanks for the feedback https://groups.google.com/g/d12rg/c/71AixTjBiog
On Tuesday, November 17, 2026, Pasquale Kaboth <[email protected]> wrote: > . > > *1. Backward-linked Release provenance* > > . > > This creates a backward-verifiable release history without requiring > modifications to the existing Debian Release formats. > we just added a proposed set of tags, for solving the external genesis / provenance problem, there's a debian enhancement protocol DEP for this scenario: i don't have a problem with following it through *2. Protection against conflicting but validly signed histories* > > This is a particularly important point. > > Our counter-tests demonstrated that two isolated recipients can accept > different release histories even when both are cryptographically correctly > signed. > > A hash chain alone does not prevent this type of split-view attack. > mmmm it's not an "attack" per se. iirc, rolling releases such as debian/experimental then become debian/unstable automatically after 2 weeks, then debian/testing you get the idea. (much more crucial is ISP transparent cacheing proxies screwing around in ways that no amount of apt/dpkg config fixes will solve - more on that below) when you have Reproducible builds you *want* a stake-in-the-ground Release, otherwise how are you going to report a *reproducible* build bug against a moving *non-reproducible* target?? if the user who reported the bug has the packages they used still installed then the peer-network allows the Maintainer to grab a snapshot of the reproducible-build packages *from the reporter's* system even when the public mirrors have long since moved on. also look up devuan (and the angband.pl no-systemd repositories: with that now offline i have to recompile several key packages after modifying debian/rules and debian/control to 100% remove libsystemd - xserver-xorg-core being the primary but also cups and others. complete pain but pottering's behaviour was so alarming to me, i cannot *personally* trust anything he's been involved in: others are free to make their own assessments) devuan is particularly interesting as they have a transparent proxy that replaces *some* packages and lets others "fall through" to debian stable, it's quite sophisticated and saves them a hell of a lot of mirror hosting costs for such a small team. also it's common practice to add your own repo to /etc/apt/sources.list[.d] or a local apt-mirror to act as a cache, etc etc and then there's canonical who forked the whole damn debian project, with too small a team to cope with that, resulting in incompatible forks of packages with radically different *binary* level dependencies and you could add and mix debian and ubuntu /etc/apt/sources.list entries and get into a dreadful mess... all they had to do was provide a *small* repo with enhancements... sigh i'll shut up now on this one 😁 I would therefore introduce independent witnesses that exchange signed > checkpoints from an append-only transparency log and verify their > consistency. > > Clients would retain their last accepted checkpoint and require a valid > consistency proof before accepting subsequent updates. Conflicting > checkpoints could then be detected through gossip. > you may not be aware that debian packages are required to be lexicographically increasing in revision (makes pinning very powerful), and an even more sophisticated set of "compatibility-with-other-packages" rules. this is *downstream* of the top-level "Release" file. any upgrades, *even* when you accidentally do "apt-get dist-upgrade" on debian/unstable, which i've accidentally done in the past, should always always work. [for a given definition of "always"...] one thing that has royally effed up use of debian/testing for me in the past is ISP transparent proxies! there's almost nothing you can do about that with the possible exception of routing use of apt over a goddamn TOR session. no amount of dpkg or apt configuration solves ISPs doing transparent proxies, particularly ones from Chinese ISPs written in perl (.pl) when one of your /etc/apt/sources.list entries is angband.pl and the ISP's scripts expose their private network infrastructure IP addresses due to their fuckup on gratuitous regexp matches firing http redirects directly onto their *internal* proxy servers. that was fun to watch the broadband ISP callout engineer try to diagnose. twice. A configurable witness quorum could provide an additional acceptance > condition for new checkpoints. > > *3. Decentralised trust management* > > The actual data distribution could remain entirely peer-to-peer, using Git > bundles, BitTorrent, a Kademlia DHT, or other P2P transports. > the current candidates are a local mirror service which is then libp2p enabled. however C.A.T also found a particularly intriguing and very small apt transport plugin that directly uses ipfs. only about... 300 lines of code. Trust decisions would remain independent of the transport mechanism. > yes. that's in the WOT requirements. and the really hard-to-grasp part for people unfamiliar with debian: the transport has *zero* trust associated with it. zero. period. Each user could decide which repository keys, release authorities, and > witnesses they trust. > correct. > Key rotations should be documented as explicitly authorised and signed > transitions. Threshold signatures would also be useful for particularly > critical changes. > there's debian-keyring and debian-archive-keyring they serve different purposes, look them up. > However, a new participant would still require an independently > authenticated initial trust anchor. The DHT alone cannot provide that. > correct. also listed in the requirements. bootstrap-hex0 makes it small enough that you could verify the first boot/compile by hand. will continue tomorrow, lots to cover. l.

