17:00:39 <br-m> <rucknium> Meeting time! https://github.com/monero-project/meta/issues/1491
17:00:44 <br-m> <rucknium> 1. Greetings
17:01:00 <tevador> Hi
17:01:13 <br-m> <gingeropolous> ahoy hoy
17:02:06 <br-m> <vtnerd> hi
17:02:20 <br-m> <ofrnxmr> gm
17:03:12 <br-m> <rucknium> 2. Updates. What is everyone working on?
17:04:34 <br-m> <jeffro256> Howdy
17:04:38 <br-m> <vtnerd> me: working on updating my /get_tree_paths PR for fcmp++ (lws/lwsf needs this for tx construction). this is part of lwsf testing after rebasing/merging all new changes. otherwise I will be updating my serialization pr yet again
17:04:38 <br-m> <gingeropolous> selfishmining: put @rucknium:monero.social 's gist in my bot and smoked it . antiproxy efforts kinda abandoning. Driving the bots to optimize FCMP++
17:04:39 <br-m> <rucknium> me: Wrote "2026 Counterfeiting vulnerabilities in cryptocurrency privacy projects" https://gist.github.com/Rucknium/a38821f8a1e0e0a115c0f4bc72d3bdb1 . Added 34 papers to https://moneroresearch.info (quite a backlog!). Looking at alternative Difficulty Adjustment Algorithms for Monero's block time.
17:04:42 <br-m> <ofrnxmr> Nothing research related, just addressing reviews on open prs related to peerlist stuff
17:05:59 <tevador> revisiting Share or Perish
17:06:29 <br-m> <rucknium> 3. FCMP++ to-do list status. Programming tasks (https://github.com/seraphis-migration/monero/issues/53). Reviews and audits (https://cryptpad.fr/sheet/#/2/sheet/view/yPVIUywwA9-deE9VF6GYm9bXbPdCerdST3UDEEfBxcM/embed/). FCMP++ Integration Audit Overview (https://github.com/seraphis-migration/monero/issues/294). Network upgrade  [... too long, see https://mrelay.p2pool.observer/e/-K_8yL8LNS1fakNu ]
17:07:21 <br-m> <jeffro256> https://github.com/monero-project/monero/pull/10916 is ready for review
17:08:08 <br-m> <jeffro256> carrot_core needs a rebase, but there's been no major developments for the last couple weeks 
17:08:44 <br-m> <jeffro256> https://github.com/monero-project/monero/pull/11737 was merged recently 
17:09:14 <br-m> <jeffro256> The FCMP++ which requires review the most at this point in time is https://github.com/monero-project/monero/pull/11440
17:09:20 <br-m> <jeffro256> *FCMP++ PR 
17:10:42 <br-m> <jeffro256> The work for the FCMP++ integration audit should begin very soon
17:10:50 <br-m> <rucknium> Thanks, @jeffro256:monero.social . I researched info about the 8 counterfeiting vulnerabilities (inflation bugs) discovered or exploited in cryptocurrency projects this year: https://gist.github.com/Rucknium/a38821f8a1e0e0a115c0f4bc72d3bdb1
17:11:21 <br-m> <rucknium> Five of the eight had at least one audit of the vulnerable code that didn't catch the vulnerability.
17:11:23 <br-m> <jeffro256> @jeffro256: Although, #10916 is a blocker for the audit for the transition period 
17:11:39 <br-m> <rucknium> It was a depressing exercise.
17:12:07 <br-m> <jpk68:matrix.org> Hello
17:13:02 <br-m> <rucknium> Some potential lessons for where counterfeiting flaws are empirically most likely to lurk, in the Gist.
17:13:36 <br-m> <jeffro256> @rucknium: Not exactly inspiring on confidence. Do you know if/ how many of those audits of the vulnerable projects did catch other vulnerabilities?
17:14:18 <br-m> <jeffro256> @rucknium: Thank you for this. I'd like to look over this later for sure.
17:14:37 <br-m> <rucknium> I didn't try to count those. I could revise the gist with that info.
17:15:29 <br-m> <rucknium> Like I said in the gist, any additions or corrections of the info is welcome.
17:16:30 <br-m> <rucknium> More on this topic?
17:18:09 <br-m> <rucknium> @rucknium: Those corrections are especially welcome because what I've wrote potentially puts many projects and auditing firms in a bad light.
17:18:43 <br-m> <rucknium> 4. FCMP beta stressnet (https://github.com/seraphis-migration/monero/releases/).
17:19:17 <br-m> <ofrnxmr> Ginger has some ram & sync speedups probably worth looking ay
17:19:22 <br-m> <jeffro256> Nothing from me on this at the moment 
17:19:25 <br-m> <ofrnxmr> At
17:19:44 <br-m> <jeffro256> Are those available somewhere ?
17:19:47 <br-m> <ofrnxmr> Daemon*
17:19:52 <br-m> <gingeropolous> https://github.com/Fountain5405/monerodato
17:21:15 <br-m> <gingeropolous> ive been using stressnet as a testing bed for coding agents to find optimizations. I used @rucknium:monero.social 's vuln. gist as an input to a unrestricted llm (glm 5.3) to try to exploit any of the optimizations.
17:21:54 <br-m> <rucknium> From my lips to the bot's ears
17:22:13 <br-m> <gingeropolous> feed the great entropy machines
17:22:38 <br-m> <rucknium> "Simplifications" or "optimizations" of code seems to be a cause of 2 of the 8 counterfeiting vulnerabilities.
17:22:59 <br-m> <jpk68:matrix.org> Have any of these had to do with caching?
17:23:40 <br-m> <rucknium> Sort of. Different behavior for mempool and complete block processing was the cause of 2 of the vulnerabilities.
17:23:55 <br-m> <rucknium> Oh, and the Liquid Network one involved something to do with caching
17:24:06 <br-m> <jpk68:matrix.org> @rucknium: Does that include the MWEB one? I forgot
17:24:25 <br-m> <rucknium> @jpk68:matrix.org: Yes
17:24:47 <br-m> <rucknium> https://blog.liquid.net/liquid-network-security-incident-assessment/
17:24:47 <br-m> <rucknium> > On September 6, 2026 at 13:53 UTC (Liquid block 4,050,336), an attacker exploited a vulnerability in Elements' rangeproof verification cache, allowing a single transaction to pass validation even though its output value was not backed by its inputs.
17:24:55 <br-m> <gingeropolous> @jpk68:matrix.org: , sort of? though the caching that it worked on were related to LMDB stuff to try and speed up HDD sync
17:25:13 <br-m> <gingeropolous> perhaps thats a different caching
17:26:26 <br-m> <jeffro256> Which is one reason why we want #10916 to be in-tree for the audit 
17:26:47 <br-m> <rucknium> Is there anything like a Lean to normal-code verification bridge? Like a mathematics proofs is formally verified in Lean, then some bridge machine-verifies that code faithfully executes what's proven in the Lean code?
17:27:41 <br-m> <jeffro256> Can Lean describe cache invalidation and database consistency ?
17:28:38 <br-m> <rucknium> Simplifications and optimizations can make the mathematics look different from code. That happens often in statistics, I know. For example, all the proofs with Ordinary Least Squares regression use explicit matrix inversion, but code almost always uses the QR decomposition for speed and numerical stability reasons.
17:29:30 <br-m> <rucknium> Of course, in statistics usually your data isn't trying to actively attack you.
17:30:20 <br-m> <gingeropolous> im trying to get the readme organized better such that each commit shows how much optimization is brings compared to stock. numbers seem all jumbly to me currently
17:32:01 <br-m> <rucknium> More on this topic?
17:33:32 <br-m> <rucknium> 5. Replace the difficulty adjustment algorithm with ASERT (https://github.com/monero-project/research-lab/issues/165).
17:34:08 <tevador> The proposal is: 1) Switch to monotonic timestamps from v17. 2) Switch the DAA to ASERT or WTEMA from v18. If the second one can't be done, we should at least ship the first change.
17:35:40 <br-m> <rucknium> Was there a reason for non-monotonic timestamps? Lost in the minds of the original Cryptonote developers? Was it to be too forgiving to poorly-behaved miners?
17:36:34 <tevador> I think the point was so a later block can "correct" an incorrect future timestamp, but this causes more problems that it solves.
17:36:58 <selsta> not a researcher but I support this proposal
17:37:12 <br-m> <rucknium> Would be good to get sech1's view on this.
17:37:29 <tevador> Pools will simply use max(current_time(), parent.timestamp + 1)
17:38:24 <br-m> <vtnerd> couldn’t the proposed algorithm cause chain splits with that max check? admittedly not full versed on this new proposal
17:38:50 <tevador> Do you mean the FTL check? That already exists today.
17:39:18 <br-m> <vtnerd> ftl?
17:39:25 <tevador> The proposal just reduces it from 2 hours to 30 minutes, which is more sensible IMO.
17:39:28 <tevador> Future time limit
17:39:48 <tevador> How much a block's timestamp can be in the future
17:40:03 <br-m> <vtnerd> ah I somehow missed. that
17:40:16 <br-m> <vtnerd> yes thats what was I referencing
17:41:08 <br-m> <vtnerd> the risk is primarily just someone not running ntp for the most part
17:41:35 <br-m> <rucknium> I have a question about FTL. Say I get a block that has a timestamp just beyond the FTL. I reject it. A few minutes later with no one finding a block in the meantime, the block becomes valid. What happens?
17:42:05 <tevador> I think current monerod rejects it, but the correct behavior is to wait until the block becomes valid.
17:43:29 <br-m> <ofrnxmr> So some nodes could accept it?
17:43:35 <br-m> <rucknium> Couldn't current behavior cause a chainsplit if an adversary timed block propagation just right?
17:44:07 <br-m> <ofrnxmr> My clock at 1701 and ruckniuns at 1658, block tinestamp is 1729
17:44:16 <br-m> <rucknium> Especially if the block wsa deliberately slow to propagate, e.g. had a lot of non-txpool txs in it
17:44:25 <br-m> <rucknium> was*
17:45:27 <tevador> Yes, that can happen. Each node uses its local clock. 30 minute FTL means we accepts nodes with local time up to 15 minutes off
17:45:47 <br-m> <ofrnxmr> If waiting for block to become valid, it would become an alt block and then higher diff chain wins, which sounds correct. But if ruckniun rejects the block, that sounds like a chainsplit?
17:46:33 <br-m> <rucknium> IIRC the rejected alt chains cache was removed by @jeffro256:monero.social , so that helps.
17:46:35 <br-m> <ofrnxmr> s/would/could
17:46:39 <tevador> not really a split, just a temporary fork
17:47:25 <br-m> <rucknium> It could be a good idea to test this behavior in a simulation to make sure there isn't a hidden problem.
17:47:38 <br-m> <rucknium> @gingeropolous:monero.social: ^
17:47:45 <tevador> In any case, this is mostly unrelated to the proposal.
17:48:11 <br-m> <rucknium> I've begun working my way through economic analyses of DAAs, starting with Noda, S., Okumura, K., & Hashimoto, Y. (2026). An Economic Analysis of Difficulty Adjustment Algorithms in Proof‐of‐Work Blockchain Systems (https://moneroresearch.info/316)
17:48:11 <tevador> The proposal does not change the behavior, only the FTL value.
17:48:47 <br-m> <rucknium> They say that Monero's DAA is close to the DAA of Bitcoin Cash between 2017 and when BCH implemented ASERT, with a few differences.
17:49:43 <br-m> <rucknium> The model is restrictive. They assume that miners are "small", i.e. no miners with a large percentage of hashpower that could individually influence the DAA adjustment.
17:50:28 <br-m> <rucknium> They say that ASERT is better than DAA-2 (the Monero/early BCH DAA), in long-term stability terms.
17:50:39 <tevador> Yes, Monero's DAA is not very good, hence the proposal. It's very bad against profit switching miners. Monero is less affected because of the network high hashrate.
17:51:08 <br-m> <rucknium> At the moment I would lean towards ASERT instead of WMETA because ASERT has been battle-tested on mainnet in BCH for years.
17:51:27 <br-m> <rucknium> But I will look into it more.
17:52:55 <tevador> WTEMA has been battle tested on Grin since 2021
17:53:40 <tevador> Or maybe just tested. Not sure about the battle part. It's on mainnet anyways.
17:54:32 <br-m> <rucknium> Is Grin the main coin for its hash algorithm?
17:55:42 <sech1> I will take a look at the diff adjustment algorithm proposal. Right now I can't say anything, I haven't seen it before
17:55:57 <br-m> <rucknium> Thanks, sech1
17:55:59 <tevador> rucknium: probably yes
17:56:42 <selsta> jeffro256: sent you an LLM review 10916, one of the findings is marked high and definitely needs to be fixed
17:57:24 <br-m> <rucknium> More on this topic? It will stay on the agenda for next week definitely.
17:58:32 <br-m> <rucknium> 6. Discussion: MRL AI/LLM usage disclosure policy.
17:59:18 <br-m> <ofrnxmr> Nobody is going to follow a disclosure policy except for the most honest of people
17:59:20 <br-m> <rucknium> At least tevador and I have requested people disclose AI/LLM usage in the research-lab repo issues. So that makes two of us!
17:59:31 <br-m> <ofrnxmr> So its one of those rules that only applies to law abiding citizens
17:59:44 <br-m> <rucknium> Researchers should be the most honest of people.
17:59:57 <br-m> <gingeropolous> im tryin to plaster all my bot text output with the heading now. but anything from my fountain5405 account is bot output
17:59:59 <br-m> <ofrnxmr> Should be, but
18:00:23 <br-m> <jbabb:cypherstack.com> It can also be helpful to know which models were used in order to know if others should be checked.
18:00:54 <tevador> I'd like to at least know that I'm replying to a 100% AI generated post that the human just copy pasted over.
18:01:06 <br-m> <ofrnxmr> I find it safe to just assume that everybody is on peds. I mean, using llms
18:01:50 <br-m> <ofrnxmr> A lot of comments on things are annoying to read because you can tell they are written by an ai with an opening and closing sentence written by a human
18:02:21 <br-m> <rucknium> I collected a few journal submission policies about AI/LLM usage.
18:02:26 <br-m> <rucknium> https://petsymposium.org/authors-2027.php#submission-guidelines
18:02:26 <br-m> <rucknium> https://web.archive.org/web/20260531220816/https://www.acm.org/publications/policies/new-acm-policy-on-authorship
18:02:26 <br-m> <rucknium> https://www.aeaweb.org/journals/aer/editorial-policy
18:02:46 <tevador> This one was without the opening and closing sentence IMO: https://github.com/monero-project/research-lab/pull/166
18:03:10 <br-m> <rucknium> A MRL repo issue isn't at that level of formality, but those policies show how other institutions are reacting.
18:03:50 <br-m> <ofrnxmr> tevador: Agreed. the whole thing is copy and pasted (or the llm did the whole pr itself)
18:04:04 <br-m> <rucknium> Definitely a decision doesn't have to be made today, but it was time to open the discussion.
18:04:54 <tevador> On the other hand, after going through several pages of slop, some of it led to some improvements I'm planning for SoP.
18:05:27 <tevador> But the signal to noise ratio was too low for this one.
18:06:07 <br-m> <ofrnxmr> https://github.com/monero-project/monero-site/pull/2747#issuecomment-6944938125
18:06:07 <br-m> <ofrnxmr> less technical repos have plenty of disclosure, but still slop
18:06:56 <br-m> <rucknium> I read and evaluate something differently if I know an AI/LLM made it.
18:07:06 <br-m> <ofrnxmr> https://github.com/monero-project/monero-docs/pull/447
18:07:07 <br-m> <ofrnxmr> another example
18:07:43 <br-m> <ofrnxmr> @rucknium: If i know im talking to a bot, im usually more strict with my feedback, since im just prompting an llm now
18:07:43 <br-m> <rucknium> It's usually easy to tell, but I still like the notice of AI/LLM usage.
18:07:54 <br-m> <jpk68:matrix.org> @rucknium: This sounds like a political compass question
18:08:46 <tevador> Those are at least tiny PRs that a human can review.
18:08:51 <br-m> <syntheticbird> i don't see a wrong in requiring a notice, I do see a wrong in seeing its a useless. Both might be right, but one is a free improvement, the other is just useless argumentative 
18:09:32 <br-m> <syntheticbird> im tired dont mind my very bad english
18:10:12 <br-m> <ofrnxmr> @syntheticbird: Im not saying its useless, im saying that there are repos that dont accept llm submissions, and the llms just stop disclosing
18:10:22 <br-m> <syntheticbird> @ofrnxmr: yes and?
18:10:24 <br-m> <rucknium> I think there's a strawman in the room :D
18:10:42 <br-m> <rucknium> I haven't seen anyone in the meeting say that LLM submissions should be banned
18:10:56 <br-m> <ofrnxmr> I was just using it as an example of an extreme
18:11:54 <br-m> <ofrnxmr> If you want disclosure, you cant openly treat them differently, otherwise they just stop disclosing and youre back to sq 1. My 2c
18:11:55 <br-m> <gingeropolous> as an avid user of coding agents / llm, im fine with disclosing. indicating specific models used can get hairy, because my agents usually launch subagents of different types blah blah blah
18:12:23 <br-m> <rucknium> I may propose a reorganization of the https://github.com/monero-project/research-lab repo soon. Any feedback on that is welcome. And any AI/LLM disclosure policy could go in there.
18:12:25 <br-m> <ofrnxmr> (sq1 = reading 4 pages of slop before you realize its autonomous agent spam)
18:12:39 <br-m> <syntheticbird> @ofrnxmr: I think Rucknium is saying that they treat them differently on a personal level. Nothing is engaging MRL as a whole, this is an informative notice, not a discrimination label
18:13:14 <br-m> <rucknium> If I read a paper by economists or computer scientists, I will evaluate them differently.
18:13:35 <br-m> <rucknium> And be "harsh" on them in different ways.
18:13:59 <br-m> <rucknium> I mean, scrutinize different parts of their methodology in different ways.
18:14:14 <br-m> <ofrnxmr> @rucknium: This is why, unless i have reason to believe otherwise, i typically enter with the view that everyone is using llms
18:15:21 <br-m> <ofrnxmr> easier to just be harsh in general than to get hoodwinked by the ones that use llms more stealthily. But yeah, ofc its easy to be told up front which ones are generated vs hand made
18:16:40 <br-m> <rucknium> Anything more on this topic for now?
18:17:56 <br-m> <rucknium> (I think someone on Matrix side was typing)
18:18:14 <br-m> <ofrnxmr> fwiw, i dont want it to sound like im against disclosure. Im just trying to be realistic about the outcomes. Today its easier to err on side of caution and assume the worse imo
18:18:18 <br-m> <vtnerd> nope, sorry hit keyboard
18:18:39 <br-m> <rucknium> 7. Share or Perish (selfish mining mitigation) (https://github.com/monero-project/research-lab/issues/146).
18:18:52 <br-m> <gingeropolous> speaking of llm gen stuff: https://github.com/Fountain5405/monerosim/blob/main/studies/sop/docs/20261118_mrl_status.md
18:18:56 <tevador> I decided to make one small change to SoP (post will be updated shortly): deterministic tie breaking instead of random tie breaking. Random tie breaking seems better in my simulation but will perform worse in practice with nonzero propagation delays.
18:19:56 <tevador> I initially avoided it because it tells the attacker in advance if he will win the tie or not.
18:22:08 <tevador> A big advantage is that in the honest-only setting, the chance of a longer split should be significantly reduced (I'll leave conrete numbers for gingeropolous to simulate).
18:22:41 <br-m> <rucknium> @gingeropolous: @gingeropolous:monero.social: Thanks. I will take a look at it soon.
18:24:11 <br-m> <rucknium> And thanks to tevador for the followup work on SoP.
18:24:20 <br-m> <rucknium> More on this topic for now?
18:24:52 <br-m> <gingeropolous> so i guess i'll just wait for updates on #146 before running stuff again?
18:25:06 <tevador> Yes
18:25:26 <br-m> <gingeropolous> sounds good, thanks!
18:25:40 <tevador> Thanks
18:26:55 <br-m> <rucknium> 8.  Shi et al. (2026) "Are Unreachable Nodes Truly Safe? Fully Eclipsing Monero’s P2P Network!" (https://arxiv.org/pdf/2609.10260). p2p: increase white_list size and change connection selection algorithm (https://github.com/monero-project/monero/pull/11527). p2p: deduplicate outbound peers by netgroup and bucket the peer lists (https://github.com/monero-project/monero/pull/11707).
18:28:02 <br-m> <ofrnxmr> me: addressing some reviews on some peerlist housekeeping prs
18:28:07 <br-m> <rucknium> @thomasbuilds:matrix.org, @ofrnxmr:monero.social , and I had some discussion on https://github.com/monero-project/monero/pull/11527
18:29:17 <br-m> <ofrnxmr> 11705 includes most of 11527 (except for the whitelist increase). 11706 includes side-issue that is worsened if we increase the whitelist size
18:29:45 <br-m> <rucknium> I noticed https://github.com/monero-project/monero/pull/11764 by @ap4sh:matrix.org . Is that proposal to update the last_seen when my node initiates an action, or when another node initiates the action?
18:30:32 <br-m> <ofrnxmr> @ofrnxmr: (dead nodes accumulating in a large / unpruned whitelist causes you to select from gray, which goes against the "white first" intended behavior)
18:30:58 <br-m> <rucknium> The former seems ok, but the latter has been deliberately avoided by the code.
18:31:36 <br-m> <ofrnxmr> I saw samy pr but havent had a chance to look at it yet
18:32:54 <br-m> <vtnerd> the samy pr seems sane, as last_seen is tracking last successful outbound to that node
18:33:13 <br-m> <vtnerd> I believe its only used on full handshakes currently, but I could be wrong
18:33:21 <br-m> <vtnerd> using it in that ping-back scenario seems sane
18:33:41 <br-m> <vtnerd> it’ll be a merge issue with ssl, but I’ll figure that out
18:35:05 <br-m> <ofrnxmr> @vtnerd: Perhaps related https://github.com/monero-project/monero/commit/5af802f0ce77e091c24fa5bfbf2f79b12d593730
18:35:39 <br-m> <ofrnxmr> related/conflicting. my work sends pingbacks to gray
18:37:10 <br-m> <rucknium> More on this topic?
18:38:35 <br-m> <vtnerd> @ofrnxmr: I don’t see why it would sent it to gray, but I may need to look at that pr further
18:40:54 <br-m> <rucknium> We can end the meeting here. Thanks everyone.
