A Bitcoin block can carry full proof of work and still be invalid.

I've catalogued 143 established cases from 2010 to 2026, including two previously unreported F2Pool timestamp failures this year.

https://deadmanoz.xyz/posts/2026/invalid-blocks

89 blocks in 2012 contained the same P2SH-invalid transaction, mined by pools that hadn’t upgraded.

Archived explorer transaction lists and preserved headers establish its inclusion in all 89. Only four complete block bodies could be rebuilt.

Both recent F2Pool headers survived in merge-mined chains, with no known Bitcoin-side capture.

These chains also yielded ~1,400 stales new to the stale-blocks collection. Monitoring them is essential to a holistic Bitcoin monitoring effort.

127 of the 143 blocks have no complete body, yet their failures remain independently checkable.

10 further reports still need evidence. Old node files, pool logs or archives could help.

Contributions and corrections: https://github.com/bitcoin-data/invalid-blocks

Time to give my 10 year old servers another 10 years of life 😬

Opus 5 is a terrible model, it constantly makes mistakes and communicates in vacuous and confusing language

In late 2024/early 2025 when I was “getting back on the tools” after a few years of management, I realised that I didn’t really want to go back to dev mode again.

Fortunately I’d been experimenting with LLMs for work, early GitHub and Cursor with the tab and code- block completion, and it was clear that these tools would increasingly allow me to design and build things without having to dev mode again

I, like many others, have been aggressively pushing to leverage the tools for maximum utility. They are still lacking in areas, and you do need to invest heavily in experimentation, but it’s abundantly clear that human time and effort is better spent up the value chain and not writing any code whatsoever.

Maybe we haven’t yet passed the threshold for all codebases or projects, but it’s only a matter of when, not if.

This is both very interesting and very scary

https://www.youtube.com/watch?v=87DyyMV0kCY

Introducing Mempool Atlas

Mode 1: breakdowns and visualisations for ALL transactions in a single nodes mempool

Mode 2: similar to the above, but comparing between 2 nodes (default atm)

I have hooked up 3 nodes to this for the BIP-110 fork - core, knots & libre

(been iterating on the idea for a bit, but rushed to get this out today, probably a few bugs)

https://atlas.deadmanoz.xyz/

The boys are back in town !!

I am pretty diligent in record keeping, but it’s clear I need to ramp up what is recorded when doing impactful stuff.

For example, recording the versions of software, firmware and hardware.

Pre 4.0.X versions of the COLDCARD mk3 were ok, but nothing following (at least until the recent fix) 😬

Claude law “given enough tokens, all bugs are shallow”

@Max on PBJ

https://en.wikipedia.org/wiki/Linus%27s_law

Phew.. spent the last 6 hours or so frantically trying to get a handle on the coldcard issue and then assisting potentially impacted family and friends

Thankfully none impacted, but seen reports online of people that have been, my heart goes out to them 😢

“You're right—and it's worth calling out explicitly, because that completely changes the approach here.”