bootlace

proof of work, please audit and report findings: https://github.com/bootlace-dev/subzero-keyosk

A tiny, auditable live USB/SD image turning any PC or Raspberry Pi into an amnesic, airgapped Bitcoin seed and key generator from physical coin flips & dice.

#opentowork

Public Key

npub1e94hqt3fuu7rvy9rpl85h3339vtn8psgewq4u05r7q4nup2kwp4q0flgay

Profile Code

nprofile1qqsvj6ms9c57w0pkzz3sln6tcccjk9enscyvhq2786plq2e7q4t8q6spz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dskxqjc0

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1e94hqt3fuu7rvy9rpl85h3339vtn8psgewq4u05r7q4nup2kwp4q0flgay

Show more details

Last Notes

2026-08-26 18:06:30 UTC

- reply

For what it's worth, we added this section about verifying the sd card:
https://github.com/bootlace-dev/subzero-keyosk#-physical-media--sd-card-verification
Also, we simplified the screen rendering process, and also made it testnet4-only for now until it gets more eyes on it (kind of test-driven development, TDD).

2026-08-26 13:10:04 UTC

- reply

HSAs are amazing! Thanks for offering this service!
Can you comment on the interesting benevolent legislative history, and the current taxable complications, of the Medicare 6-month look-back rule?
Specifically, how that rule may force Medicare applicants to withdraw "excess" contributions—and gains—from their HSA, the tax consequences of that withdrawal, and the messy HSA implications of applying for Medicare within the first six months of a calendar year?
A family member is dealing with that mess right now, though not with Sound HSA.

2026-08-26 00:51:50 UTC

- reply

Animated walk-through:
https://blossom.primal.net/b542f338882a7b89777147ded696ad74c4c32f21d035fa310a87744e71305f76.gif

2026-08-26 00:48:00 UTC

- reply

me and my million llm monkeys made some major mods
focusing on testnet4 for the time being
pc and rpi builds available for review, audit, testing, and feedback, please, thanks:
https://github.com/bootlace-dev/subzero-keyosk/releases
(test entropy vectors pre-loaded with testnet sats!)
A tiny, auditable live USB/SD image (.img) turning any PC/RPI into an amnesic, airgapped Bitcoin cold vault key generator & BIP380 descriptor exporter.

2026-08-25 01:18:52 UTC

- reply

I notice you say "before every sign". Note that subzero-keyosk is not a signer, it's for seed generation only.

2026-08-25 01:14:52 UTC

- reply

Yes, the Pi build is 100% Alpine too, built up from alpine-minirootfs and heavily stripped down.
🛡 STRIPPED BARE-BONES AIRGAP:
• Kernel network stack physically deleted: removed kernel/net, wireless firmware, bluetooth, sound, and USB-to-ethernet/serial drivers.
• Strace library whitelist: the builder traces exact .so dependencies during execution and purges all untouched libraries from /usr/lib.
• Package manager destroyed: /lib/apk/db and /var/lib/apk are nuked post-build so no package manipulation is possible.
• Attack surface reduction: all gettys disabled, DMA protections enabled in bootloader (intel_iommu=on, efi=disable_early_pci_dma).
⚡ DISKLESS TORAM BOOT:
• Custom minimal initramfs copies the ~150MB compressed SquashFS directly into a RAM tmpfs disk.
• Hardware write-protect: you can lock the physical write-protect slider on a full-size SD card/adapter for hardware-enforced read-only integrity.
• Storage unmounted at boot: the SD card / USB drive is unmounted before userspace init. You can physically pull the media once booted.
• Zero SD wear: the disk is never mounted writable during operation. Cold power-cycle wipes RAM instantly.
Repo with build scripts and release images:
https://github.com/bootlace-dev/subzero-keyosk
Please review, audit, and comment. Thanks!

2026-08-24 23:09:00 UTC

- reply

Which hardware vendor will be first to _hide_ the easy button?
My attempt to address the problem:
https://github.com/bootlace-dev/subzero-keyosk/releases
Please verify, audit, review. Thanks!

2026-08-24 23:03:57 UTC

- reply

From AI:
Yes, unfortunately, a severe and devastating tornado struck the village of Pomas today, August 24, 2026.
The tornado developed from a powerful supercell thunderstorm and touched down around 5:20 PM local time. Here are the latest details on the situation:
* Injuries: Authorities report that at least 41 people have been injured, with several hospitalized. Earlier reports mentioned two people were being searched for under the rubble of a home.
* Damage: The tornado caused immense destruction in the village, damaging approximately 300 homes. Many buildings had their roofs completely ripped off, and trees and vegetation were heavily damaged.
* Emergency Response: More than 120 firefighters, including specialized search and rescue teams, have been deployed to the area to secure buildings and assist residents. A shelter has been opened at the local community hall for those displaced.
* Wider Weather Impacts: The supercell storm that spawned the tornado also brought massive hailstones (3 to 5 cm in diameter), torrential rain, and wind gusts exceeding 100 km/h across the Aude department. The severe weather even forced officials to cancel the third stage of the Vuelta a España cycling race, which was taking place in the region.
The region had been placed on an orange alert for severe storms prior to the event, as the clash of hot, humid air from the Mediterranean and cooler air aloft created a highly unstable atmosphere.

2026-08-24 12:10:20 UTC

- reply

Inspired by your note, we (AI and I) added an armv7 image. Works on my ancient RPI. Please verify, audit, and test. Thanks!
https://github.com/bootlace-dev/subzero-keyosk/releases
(Note I renamed my github account.)

2026-08-22 14:34:14 UTC

- reply

Thanks so much for the thoughtful reply and checking it out!
On networking: we didn't just disable interfaces in software, we physically stripped them from the filesystem and module tree:
• Deleted all network/wireless driver modules, USB network dongles, and proprietary firmware blobs (plugging in a USB wifi dongle is completely inert).
• Purged all latent network protocol modules from /lib/modules.
• Kernel enforces DMA isolation (intel_iommu=on, amd_iommu=on, iommu=force, efi=disable_early_pci_dma).
On the doctored SD / evil maid swap risk:
• Against exfiltration: SubZero boots 100% into volatile RAM (tmpfs), unmounts the storage, and prompts you to physically pull the USB drive before entering a single coin flip. Nothing touches disk.
• Against short-list / biased-math tampering: The pure dice/coin math is 100% deterministic BIP39 (verifiable offline against standalone calculators/tables), and the entire OS image is byte-for-byte reproducible with GPG-signed checksums.
We definitely plan to add direct Raspberry Pi / ARM64 SBC support to our roadmap next (plugging HDMI into a dumb TV/monitor + USB keyboard).
Build script & module stripping logic is here if you want to inspect:
https://github.com/bootlace-dev/subzero-keyosk/blob/main/scripts/build_alpine_kiosk.sh
Curious to hear how your airgapped Pi setup shapes up!
cc: @thejohnnycrypto

2026-08-22 02:52:56 UTC

- reply

Hi Johnny,
Following up on this: to eliminate the browser DOM trust gap entirely, I used Gemini this past week to evolve it into a bare-metal, amnesic live USB image with GPG signed checksums:
https://github.com/bootlace-dev/subzero-keyosk
(and changed my github user name)
Please review, audit, and comment. Thanks!

2026-08-21 21:43:17 UTC

- reply

Inspired (and obsessed) by all this, I used Gemini this week to build a new bare-metal, air-gapped, widely-bootable image, for use on old commodity laptops, with coin/dice for entropy (and changed my dumb old random github name):
https://github.com/bootlace-dev/subzero-keyosk
Please review, audit, and comment! Thank you! https://blossom.primal.net/7645ebfc2b94c47b903516cc0aea8d3bf0f7c91e7d717f46b7e7d155e4bbbfd5.png https://blossom.primal.net/25417271ec234f4037dd00ca922044d0e8c82ca3d076acf0ce8780d4a8e89ac9.png https://blossom.primal.net/117b67b7a3f630f9514b734ad9b019b7d624905c37b418deb1b516f5c2bbe545.png https://blossom.primal.net/3b2ab700301d69096cf5659953aa62cfd9946a096476ef16eb7956ffe72ed602.png

2026-08-21 18:40:20 UTC

- reply

Need to get tuta on nostr

2026-08-18 23:44:03 UTC

- reply

Need one of these "they're the same picture" memes

2026-08-18 12:08:50 UTC

- reply

Agree 💯 with everything, but sadly, Saef gives me the same vibe as NVK. I can't listen to him anymore. :(

2026-08-17 22:05:23 UTC

- reply

using device entropy should,
a) NOT be the default, and
b) should be buried behind aggressive warnings NOT to use it

2026-08-17 22:02:25 UTC

- reply

especially sad hearing stories that people moved away from coldcard and still lost their life savings because the bad entropy followed them 😟

2026-08-17 21:58:30 UTC

- reply

botox, that's all I see

2026-08-17 18:36:18 UTC

- reply

an old laptop or chromebook

2026-08-17 18:33:01 UTC

- reply

Thought you might like my recent post documenting my struggles with USB-C cables:
https://primal.net/e/nevent1qqs9afggyzhp7jk83nmaj393qdx0n3laglu8fp2ne3ewqewezks4zjqvhd272

2026-08-16 23:35:51 UTC

- reply

Two straws

2026-08-16 11:54:54 UTC

- reply

finally

2026-08-16 01:18:35 UTC

- reply

it's fascinating and telling about the evolution and intelligence of the singleton, and separately of the herd, that they can't or won't counterattack the lone wolf as it were

2026-08-15 16:08:04 UTC

Help me make this description better:
🔌 THE USB-C CHARGING & COMMUNICATION TRAP
USB-C cables are confusing and frustrating. Why are they not communicating with your laptop (file transfer, flashing devices, etc)? Why are they charging so slowly?
Both high-speed charging and data file transfers require a digital connection between your devices. Your cables fail these tasks due to a few distinct issues:
• MISSING WIRES IN THE CABLE BODY: Cheap "charge-only" cables leave out the internal data and communication wires entirely to save cost. The laptop cannot detect the phone at all (zero file transfer), and charging is forced to standard slow speeds. You cannot tell by looking at the plug; the metal pins might be present on the tip, but they are "dead" inside.
• MISSING METAL CONTACTS IN THE TIP: If the physical pins responsible for either file transfer or fast charging are missing from the plug, that specific function is guaranteed to be gone.
• BROKEN OR DIRTY TIPS: Worn pins or lint blocks the signals. Because pins are separate, you might lose file transfer capability while charging still works, or vice versa. This often causes "asymmetrical" behavior, where a function only works if you flip the plug over to the other side (requiring testing by flipping one end, then the other end, then the first end again).
If the connection fails for any of these reasons, the devices cannot negotiate and the charger silently falls back to standard slow charging, while file transfer fails entirely.
Once you figure out its status, label the cable with something of meaning to you (example: "PD" for power-delivery only, "2" for USB-2 data transfer, "ASYNC" meaning you have to flip the tips to make it work).

2026-08-15 15:10:20 UTC

- reply

Nice catch, thanks. Zoom in on the temporary "decoupling" in early 2024. Looks like that is where we are now, too. Spring could surely be well-coiled, wow!

2026-08-11 12:10:09 UTC

- reply

winning with pennies
#nevent1q…63wd

2026-08-09 18:08:52 UTC

- reply

Agreed with one add:
take it from there, verify it, and run your copy offline.

2026-08-09 17:58:55 UTC

- reply

I'm trying to modernize and simplify his masterpiece, please audit and comment:
https://github.com/ghscuuo/airgap-coinflip/releases/tag/v0.0.1-rc.1
https://github.com/iancoleman/bip39/releases
#nevent1q…aaqu

2026-08-07 05:40:13 UTC

- reply

Me and my million models made this:
#naddr1qq…t34s

2026-08-07 01:10:38 UTC

- reply

Add: verify Ian Coleman offline standalone html file hash, against historic records in a few places, to confirm it hasn't been swapped out with a bad version
Example for an oldie but a goodie version, see hashes here (ignore virus scan info, just check the hashes):
https://www.virustotal.com/gui/file/8b8e3c1be03501f57e395781de8a59fd553808e1eb1278710bd7b96dacb6d0f6/details
Pasting version 0.5.4 version hash here to expand the footprint:
8b8e3c1be03501f57e395781de8a59fd553808e1eb1278710bd7b96dacb6d0f6
Ask me if you need more.
Postramble:
My new AI generated, tested, and audited modernized and simplified work in progress tool here, for your AI to audit:
https://github.com/ghscuuo/airgap-coinflip/releases/tag/v0.0.1-rc.1
https://github.com/ghscuuo/airgap-coinflip/releases/download/v0.0.1-rc.1/airgap-coinflip_a5418711.html
sha256:a5418711d7bb0a5ea1dec2723cfe826df241956be62923cdc81211c1de4e348b
https://www.virustotal.com/gui/file/a5418711d7bb0a5ea1dec2723cfe826df241956be62923cdc81211c1de4e348b

2026-08-06 11:59:45 UTC

- reply

So much of life is beautifully, reliably mechanical: you do this, you get that. Cause and effect, clean and true.
Maybe that's just the engineer in me, expecting the world to run on predictable logic. It lulls you into trusting the floor.
But look closer, and the rest is just an open-air casino run by human nature. My safety, my wealth, my happiness, my entire life depend on someone else's fragile interpretation of words in a statute. My livelihood rests on gambling on other people's reputations, shifting moods, and invisible social tethers—nothing concrete, nothing real, just a massive, high-stakes wager on what humans will decide to do next.

2026-08-06 02:45:52 UTC

- reply

Physical entropy only (coins; dice if market demands)

2026-08-05 21:05:14 UTC

- reply

how did your friend escape by an inch, may we ask?

2026-08-05 16:49:03 UTC

- reply

Michael1011 signed the new canary today.
Signature made Wed 05 Aug 2026
Hopefully not under duress. 😞
https://canary.boltz.exchange/
https://github.com/michael1011
https://michael1011.at/michael1011.pgp
C2640F630570F5EDEDE02DE684D249BA71685D46
C264 0F63 0570 F5ED EDE0 2DE6 84D2 49BA 7168 5D46
71685D46
0x71685D46

2026-08-04 13:39:29 UTC

- reply

I love @nprofile…nj96 and @nprofile…yj2m
but unless I'm missing something:
black box red flag
and
fails multi vendor multi sig
Instead, flip coins, roll dice

2026-08-04 04:11:39 UTC

- reply

Yes, see #7: https://github.com/ghscuuo/airgap-coinflip/discussions/4
Search for "buffer", which are wipeable in javascript, and "string" which are not.
in the end, there's nothing I or the user can do but turn off the offline computer to free any strings holding memory.
It's on my todo list to move the one or two remaining strings that I can to be buffers instead, but again, in the end it doesn't really matter, the user has to turn off the offline computer to be sure. I'm ok with that.
Thanks again for your continued feedback. I appreciate it. Let me know if you think of anything else.

2026-08-04 03:29:15 UTC

- reply

Yes, all of your points are understood. Thanks for the feedback. I'll see what I can do...
When you say "unwanted inputs", can you give me an example situation?
---
I guess one can consider this offline tool (airgap-coinflip) for _receiving_ bitcoin.
I plan to build a separate, equally pared down, offline tool for signing and _spending_ bitcoin.
Both tools should be used fairly rarely, actually.
(For what it's worth, I'll plan to leave the online PSBT generating to other watch-only online wallets, unless folks think I should add that to this simplified tool box. Based on my experience consolidating utxos a while ago, the popular online watch-only wallets each had a few interoperability quirks that were time-consuming. A thought in progress.)
https://github.com/ghscuuo/airgap-coinflip/releases/tag/v0.0.1-rc.1

2026-08-03 16:20:11 UTC

- reply

Fair.
But we also need a secure way to calculate:
a) the sha256 of the 128 0s and 1s for the checksum
b) the initial receiving addresses (and/or output descriptor)
That's when I started down the path long ago of, well, we need to use an offline computer for all that anyway, so we might as well find or build a simple and safe offline tool.
Which led me to Ian Coleman's https://github.com/iancoleman/bip39/releases#release-0.5.4
But that tool has some issues. And so now, with AI for me, some of those issues can be resolved.
I think you and I are on very close to exactly the same page. Let me know how I can make this offline tool palatable for your base use-case:
https://github.com/ghscuuo/airgap-coinflip/releases/tag/v0.0.1-rc.1

2026-08-03 16:00:38 UTC

Most users don't need to buy a hardware wallet, and probably shouldn't. Don't let the marketing and influencers sway you.
For deep storage, physical backups replace vulnerable hardware firmware.
Keep these two components, optionally along with their qr codes.
1. The Keys
Secret Seed Words (Private)
CRITICAL: Keep Strictly Offline & Secret. This unlocks your coins. If anyone gets these words, your coins are gone. Store in multiple secure places on paper or stamped steel in a secure vault (or encrypted if you and your beneficiaries know how).
Example 12-word seed phrase:
audit journey sense bulk valley maple destroy tiger audit journey sense cable
2. The Map
Output Descriptor (Public)
SAFE TO SHARE / WATCH-ONLY: This tells your wallet software exactly where to look on the bitcoin blockchain to find your bitcoin balance (your money). Without this map, standard recovery software won't know which address paths to search. Keep with your inheritance papers.
Example "Output Descriptor" map:
wpkh([832d169e/84'/0'/0']xpub6CS3oFPYQuUdxsBY8BCLyyEs7LUeP1WmhajfyTGVuQMC5kLDc3PTiM2SWyxUuTG9C5SnyCNKZnMHzdGZ1FFLUzYRYSwB6KAg9X94zJf94Tn/0/*)
https://github.com/ghscuuo/airgap-coinflip
https://blossom.primal.net/92da668c15c53dfed727a4986f559d296d75b4ad7738e5345d169ddc6eed6620.png
https://blossom.primal.net/7a2d9c16f5051d6237ccc2d83277c281e47608b5c933233d950fc1ce5ee4d250.png

2026-08-03 15:17:03 UTC

- reply

Interesting idea!
I had a long response typed in here, even drafting an offer to implement that feature in the airgap-coinflip tool, but after thinking and asking AI about it, I chose to drop it, to keep the tool as simple and safe as humanly possible.
My rationale, such as it is:
The coin flip physics bias is both,
a) microscopic, especially outside the lab (Diaconis), and
b) importantly, unknown to the attacker (unless the user "tries too hard"--just randomly flip the dang coins! 😉)
A key problem with the coldcard entropy fail this week is that the attacker _knew_ _both_
a) the target space was a searchably small 40 bits, and,
b) very importantly, precisely _where_ the output secrets lie (_all_ of the output secret sand grains are somewhere in that's kid's blue plastic beach pail, as opposed to anywhere in the known universe, something like that).
(If the attacker knew the coldcard seed space was only 40 bits, but didn't know _where_ those 40 bits were, then they still have to scan all 128 bits.)
Assessing Accuracy and Clarity
• Mathematical and Cryptographic Accuracy:
• 100% Correct. Your breakdown matches core cryptographic definitions of entropy, search-space reduction, and Kerckhoffs's principle.
• Why Your Explanations Hold Up:
• Coin Flip Bias: Correct. The Diaconis bias is dynamically erased by surface bounces, and mathematically hidden from the attacker because they lack the starting state observations.
• The Coldcard "Beach Pail" Metaphor: Correct. The attacker doesn't guess random points in the universe; they run the known generator algorithm to compute the exact locations inside the "pail" (the 2⁴⁰ candidate subset).
• The "Unknown 40 Bits" Logic: Correct. If a system randomly selects 2⁴⁰ coordinates out of 2²⁵⁶ without exposing the selection algorithm, the attacker is still forced to search the entire 2²⁵⁶ space. The collapse only occurs because the generator function is public.
https://blossom.primal.net/d362b24ceb825d8197f6af90a99e03f2dbaffcd46ed12443a1d92694b4aa4e56.png
https://github.com/ghscuuo/airgap-coinflip

2026-08-03 11:42:51 UTC

- reply

audit this
https://blossom.primal.net/65000d63ce1ac35720e00e7d260581032811c3907c0879faf899124d6298ef77.png

2026-08-03 02:12:44 UTC

- reply

And we hear lots of chatter that humans screw up flipping coins. Maybe. Let's train the humans better.
If the human randomly mistakes a few heads for tails, or straightens up a few of their flipped coins "switching" some coins "incorrectly", it doesn't matter:
• Human typo on 128-bit coin flips:
• Base entropy remains 128 bits.
• Transcription errors shift the output to another random 128-bit coordinate.
• Attacker learns nothing about the human coin flip mistake pattern. Attacker search space remains 2¹²⁸ (impenetrable without the original pre-image).
• 40-bit HW wallet RNG vulnerability:
• Systemic state space restricted to 2⁴⁰ total keys.
• Attacker knows exactly what the very short list of seeds are. Attacker sweeps the entire 2⁴⁰ universe (~1 trillion keys) off-site.
• Every device world-wide running the compromised firmware is swept.
• Security margin:
• Human input error: Alters the destination, preserves the entropy barrier.
• Hardware RNG flaw: Dramatically collapses the entropy barrier for all of those users globally simultaneously.
Cross-check:
### Mathematical Correctness Review: Verified Sound
• Human Typo Math:
• Let random coin-flip sequence be X (H(X) = 128 bits).
• Let transcription error offset be E.
• Key Y = X oplus E remains uniformly distributed. H(Y) = 128 bits. The search space remains 2¹²⁸.
• Nuance: If "straightens up" means the human physically flips coins but modifies runs (e.g., changing HHHH to HTHT to "look more random"), cognitive bias introduces dependency between bits, dropping the entropy. If "straightens up" just means correcting physical coin orientations to write them down, the math stands.
• 40-bit Hardware Math:
• 2⁴⁰ ≈ 1.1 trillion keys. Trivial for modern GPU/ASIC clusters to sweep.
• Systemic firmware distribution means every affected user shares the exact same small state space.
• Failure Mode Distinction:
• Human typo = Usability/recovery failure (funds locked at random 2¹²⁸ coordinate).
• Hardware RNG flaw = Cryptographic security failure (funds stolen from collapsed 2⁴⁰ coordinate).

2026-08-03 01:39:34 UTC

- reply

Agreed.
The RNG Lie: They aren't true random number generators. Look at all the tricks and hoops they have to go through, in a feeble risky error-prone attempt, to try protect the user from the user and from the manufacturer. Folks should stop trusting such devices to generate their secrets, and stop treating them like the ultimate secret vault.
The Marketing Trap: People buy these dedicated boxes thinking they bought "secure hardware." In reality, it's just an offline signing device.
The Risk Profile: They are a massive watering-hole risk. If one rarely signs transactions on-chain, then they don't need a specialized device. They're paying to put a giant bullseye on their savings.

2026-08-02 05:50:24 UTC

- reply

I know nothing but my gut says the root cause was the relatively urgent "need" to rewrite a massive amount of code to resolve the corporate software license dispute. For want of an "ifdef" or something.

2026-08-02 05:11:13 UTC

- reply

(my long winding prompt here)
AI response:
• Corporate terminology: "First Principles Analysis" or "Core Utility."
• The HWW Mantra (Translated):
• Electronics cannot be trusted to create secrets. The Coldcard bug proved devices fail silently. Create your master key in the physical world (flipping coins).
• Hardware wallets are just digital pens. They exist only to conveniently sign checks (spend). They should never create the account.
• Don't spend your vault. Hold the main stash. Use layers (Lightning/Ecash) for daily movement.
• The Secure Vault Routine (Air-gapped execution):
1. Online device: Drafts the transaction (Watch-only mode).
2. Offline device: Holds the key, reviews, and signs the draft.
3. The Air-gap: Move the data between them using QR codes or SD cards.
4. Online device: Broadcasts the finalized transaction to the network.
• Inquiry: Does your framework discard commercial hardware wallets entirely in favor of dedicated air-gapped laptops (e.g., Glacier Protocol), or are you retaining HWWs strictly as
"dumb" signing peripherals?
> i hate bitcoin-dedicated hww, tell me why you agree 😉
▸ Thought for 3s, 781 tokens
Prioritizing Tool Usage
• Black-box execution: Niche silicon and closed bootloaders prevent true user verification of runtime execution.
• Feature bloat: Modern HWWs bundle RNG, descriptor generation, NFC/Bluetooth, and complex UI. High complexity breeds firmware integration failures (e.g., Coldcard PRNG override).
• Supply chain targeting: Purchasing specialized crypto hardware exposes the buyer's identity. Physical interception vectors remain high.
• Vendor lock-in: Proprietary companion applications and non-standard derivation paths degrade long-term self-sovereignty.
• Redundant utility: Commodity offline hardware (air-gapped laptop via Lubuntu/Tails) executes identical cryptographic signing with inspectable, replaceable components.
• Centralized failure vector: Consolidating entropy generation, storage, and signing into a single appliance violates defense-in-depth principles.
AI Critique:
• Verdict: Structurally sound critique of HWW trade-offs, but falters on commodity hardware assumptions and ignores modern protocol standardization.
• Validated Arguments:
• Supply Chain: Spot on (e.g., Ledger 2020 customer database hack). Purchasing crypto hardware inherently flags targets.
• Black-Box: Spot on. Secure Elements (SE) require NDA trust in silicon manufacturers (STMicroelectronics), violating verification ethos.
• Centralized Failure: Accurate. Placing entropy generation, storage, and signing in one physical basket creates a catastrophic SPOF.
• Refutations / Weaknesses:
• Commodity Hardware Fallacy: Laptops possess massively larger, un-auditable attack surfaces (Intel ME, AMD PSP, closed-source UEFI) than microcontrollers. Trades a specialized black box for a generalized one.
• Vendor Lock-in: Largely outdated. Widespread adoption of BIP39, BIP44/84, and BIP174 (PSBTs) enables seamless cross-vendor migration and third-party coordinator use (Sparrow/Electrum).
• Critical Omissions:
• Blind Signing: Tiny screens force blind signing of complex scripts (multisig change, lightning channel opens).
• Physical Extraction: Devices lacking Secure Elements (e.g., older Trezors) are vulnerable to rapid voltage-glitching seed extraction if physically intercepted.
• Multisig Mitigations: Fails to acknowledge multi-vendor multisig setups, the industry-standard architecture designed specifically to neutralize HWW centralized failure vectors.
Note also, working on an old offline laptop is approximately 100x easier than any hww.
My draft coin-flip seed gen thing for review and comment, thanks in advance:
https://github.com/ghscuuo/airgap-coinflip/releases/tag/v0.0.1-rc.1
(I could do a super simple signing version, too, upon request)

2026-08-02 04:33:07 UTC

- reply

Hi Johnny, thanks for asking.
See:
https://github.com/ghscuuo/airgap-coinflip/releases/tag/v0.0.1-rc.1
See
airgap-coinflip_a5418711.html
sha256:a5418711d7bb0a5ea1dec2723cfe826df241956be62923cdc81211c1de4e348b
(For convenience only, see the first part of the hash a5418711 in the filename. For what it's worth, for 3rd-party hash reference only: https://www.virustotal.com/gui/file/a5418711d7bb0a5ea1dec2723cfe826df241956be62923cdc81211c1de4e348b/details )
Does that address your concern? If you build it per the instructions in the README.md, you should get the same hash. Let me know if you don't!
Suggestions welcome! Thanks!

2026-08-02 04:21:38 UTC

- reply

in a nutshell, dice are probably fine, but I think pennies are a tiny bit better and not nearly as hard to do as folks fear
not sure, but perhaps, folks correctly assign very high respect to the EFF/Reinhold/Bonneau diceware system, and I agree, but diceware base6 7,776-word list passwords dont have the base6->base2 conversion problem required for the bip39 2048-word list
sorry if I'm wrong
• Hypothesis 1 (Pennies vs. Dice): Fair and accurate. Pennies eliminate format/translation vectors in exchange for a manageable increase in physical labor.
• Hypothesis 2 (Diceware vs. BIP39 translation): 100% accurate observation.
• Diceware Mechanics (Base-6 to Base-6): 5 physical dice rolls yield exactly 7,776 states (6⁵). This maps 1:1 to the 7,776-word list. Zero mathematical translation. Zero modulo bias.
• BIP39 Mechanics (Base-6 to Base-2): 2,048-word list (2¹¹). Physical dice (Base-6) cannot map cleanly to an 11-bit binary space without forced translation, modulus math, or cryptographic hashing.
• Conclusion: The stellar reputation of dice is inherited from standard Diceware (where the math aligns perfectly). The math failures and UX friction arise exclusively when forcing Base-6 physical hardware into the Base-2 BIP39 standard.
maybe we need a new base6 bip...

2026-08-02 04:05:48 UTC

- reply

from my AI:
> i'm trying to express to normal non-techy humans a very rough sense of the relative safety of flipping pennies for entropy vs dice rolls and dice math
• Coins (Pennies): High Labor, Zero Risk
• Native machine language (Base-2 / Binary).
• 1 flip = exactly 1 bit. 128 flips = 128 bits.
• Aerodynamic physics of high-rotation flips neutralize minor weight imbalances.
• Zero mathematical translation required.
• Immune to format-mismatch and base-conversion errors.
• Dice (Pedestrian): Low Labor, High Risk
• Foreign machine language (Base-6).
• Requires mathematical translation (Base-6 to Base-2), introducing cryptographic failure vectors (modulo bias, truncation).
• Introduces human UX failure vectors (miscalculating roll counts, inputting into incorrect tool fields).
• Missing physical mass (drilled pips) and rounded edges skew resting probabilities.
• Base entropy structurally degrades from 2.58 bits down to ~2.5 bits before translation even begins.
• The Trade-off
• Coins require sheer physical brute force (128 flips) but guarantee cryptographic safety.
• Dice save physical labor (50 rolls) but introduce compounding catastrophic risks (manufacturing bias + translation math + human error).
I don't know if I believe all of those words, but the essence hits home for me. Flipping coins gets a bad rap. I'm trying to rehabilitate its reputation. 😉
Copy and paste that into your strongest AI and post back here what it says.
Opus comment on the above:
• AI Review Synthesis (Pro Model):
• Verdict: Persuasive success (hypes coins, demonizes dice) but severe target-audience failure (alienates laypersons with crypto-student jargon).
• Strengths: "High Labor/Zero Risk" dichotomy successfully anchors the trade-off. Highlighting physical tangibility of "drilled pips" grounds abstract entropy logic.
• Weakness (Jargon Overload): Terms like modulo bias, truncation, and UX failure vectors block non-techy comprehension.
• Weakness (Metaphor): "Base-2/Base-6 native language" introduces unnecessary mental translation steps.
• Weakness (Physics/Reality): "Zero Risk" ignores massive human fatigue vectors (boredom/miscounting 128 flips). Coin aerodynamics claim ignores Persi Diaconis physics (minor
starting-position bias).
• AI Prescriptions:
• Strip the math (e.g., "Computers speak 2s, dice speak 6s. Forcing 6 into 2 requires complicated math. Mistakes ruin randomness").
• Acknowledge physical fatigue and counting errors as the primary coin-flip risk.

2026-08-02 03:15:21 UTC

- reply

there's no money in flipping pennies, and that's a good thing

2026-08-02 03:12:44 UTC

- reply

Purchase a four-inch square galvanized steel electrical box cover and a handheld rotary power tool with cash at a local hardware store.

2026-08-01 05:25:12 UTC

- reply

Some folks may disagree with me, but trusting opaque hardware for generating entropy/seeds/keys is a single point of failure.
Last night, I used AI to architect a zero-trust alternative. It bypasses silicon entirely by forcing 128 physical coin flips to generate the master seed.
- Deterministic single-file HTML build
- CSP locked / offline execution
- Full BIP85 index derivation
- Airgapped QR code export
- Zero hardware trust (Physics only)
Have your AI verify my AI:
https://github.com/ghscuuo/airgap-coinflip

2026-08-01 05:15:40 UTC

- reply

2026-08-01 05:14:54 UTC

- reply

2026-08-01 05:03:21 UTC

Some will hate me for saying this, but trusting opaque hardware for generating entropy/seeds/keys is a single point of failure.
I used AI to architect a zero-trust alternative. It bypasses silicon entirely by forcing 128 physical coin flips to generate the master seed.
- Deterministic single-file HTML build
- CSP locked / offline execution
- Full BIP85 index derivation
- Airgapped QR code export
- Zero hardware trust (Physics only)
Have your AI verify my AI:
https://github.com/ghscuuo/airgap-coinflip

2026-07-31 14:43:09 UTC

- reply

And dice entropy math is hard.
Probably best that #Coldkite don't recommend using dice on their security announcement page, either. Any idea how they do their dice math?
AI list of dice fails:
2015 | Ian Coleman BIP39 | Aggregation method mapped Base 6 to Base 2 via raw integer string, creating subtle bit bias
2017 | Coldcard Dice Roll | Allowed users to input fewer than 99 rolls, generating dangerously low entropy seed phrases
2019 | Standard Plastic Dice| Non-precision consumer dice showed statistical bias towards 1 and 6 due to engraved weight loss
2020 | Base-6 Modulo Bias | Naive mod 6-to-2 conversions left upper bits unevenly distributed, shrinking effective keyspace
2021 | Pre-SHA256 Truncation| Early SeedSigner tools used unhashed dice strings, leading to cross-device seed mismatches
2023 | Human "Random" Input | Users manually picking "random" dice numbers collapsed keyspaces into predictable
Normie translation available upon request.

2026-07-31 14:22:39 UTC

- reply

They need to cross sign with an old account somehow.
@nprofile…5w64

2026-07-29 03:49:18 UTC

- reply

From Gemini:
That headline reflects a major, deliberate shift in Ukraine’s long-range strike strategy. Rather than focusing strictly on oil refineries and fuel depots, Ukraine has launched a widespread campaign targeting Wildberries—Russia's largest e-commerce platform and its equivalent to Amazon.
This wave of attacks has hit multiple massive logistics hubs across Russia (including facilities near Moscow, St. Petersburg, Krasnodar, Tver, and occupied Crimea).
The campaign is notable for several reasons:
* Economic and Supply Impact: The targeted hubs represent roughly 10% of Wildberries' total nationwide logistics capacity, destroying billions of rubles worth of inventory belonging to thousands of small and medium-sized Russian merchants.
* Brought Home to the Public: Unlike remote industrial sites, the destruction of major consumer distribution centers right outside cities like Moscow and St. Petersburg punctures the "bubble of normalcy" for urban Russians who had previously been insulated from the direct economic and physical disruptions of the war.
* Strategic Rationale: Ukrainian officials and defense analysts have justified the targeting by pointing to the scale of the company's logistics network, its intersections with state-backed financial infrastructure (like VTB), and allegations that parts of the vast warehousing system are utilized to support or store logistics and components connected to Russia's military supply chain.
* The Aftermath for Sellers: Adding salt to the wound for independent vendors, Wildberries updated its terms just days before the strikes began to classify drone attacks as "force majeure," leaving many merchants facing catastrophic inventory losses without corporate compensation.

2026-05-31 22:33:40 UTC

- reply

Would be fun to have someone annotate what's what on that track

2026-04-27 01:08:45 UTC

- reply

Black and white kinda makes it look like the three stooges or something

2026-04-18 19:54:32 UTC

- reply

❤️medicine man

2026-04-08 12:24:57 UTC

- reply

Safe enough I guess
https://www.virustotal.com/gui/file/e81961d40b5a544e2e0be94c1e436804a4b9303303abe3f40ca39870632ed696

2026-03-30 01:34:34 UTC

- reply

(I'm a long time River customer and recommend them. That said, I asked my LLM: )
Q: Silly?
A: Yes and no. It’s highly effective marketing from River, but as a piece of objective financial analysis, it relies on some heavy cherry-picking to push a specific narrative.
Here is why the chart is a bit "silly" or at least misleading if taken at face value:
* The 60-Day Window Trick: This is the biggest tell. Look at the COVID Outbreak on March 11, 2020. The very next day ("Black Thursday"), Bitcoin plummeted nearly 50% in a massive liquidity crunch. If the chart used a "7-Day Return" or "Max Drawdown" metric, Bitcoin would look like a disaster. A 60-day window conveniently captures the V-shaped recovery fueled by central bank money-printing, completely erasing the extreme short-term volatility.
* Selective Event History: The chart curates specific geopolitical shocks where Bitcoin eventually bounced back strongly. It notably omits prolonged macroeconomic shifts—like the aggressive interest rate hikes throughout 2022—where Bitcoin crashed heavily alongside equities and behaved exactly like a long-duration tech stock rather than an inflation hedge.
* The "Safe Haven" Illusion: The graphic is designed to contrast Bitcoin directly against Gold and the S&P 500 to cement the "digital gold" narrative. While Bitcoin has delivered spectacular returns during certain crisis windows, it remains a highly volatile, risk-on asset. Gold's returns look muted here, but Gold also doesn't typically suffer 50% drawdowns in a single week.
It’s a great chart for a Bitcoin platform's marketing material, but using it as proof that Bitcoin is a reliable, stress-free safe haven during any global crisis ignores the stomach-churning volatility that happens within those 60-day blocks.

2026-03-28 00:53:02 UTC

- reply

Any update?

2026-03-22 13:08:26 UTC

- reply

I used AI cli to fix 101 things that were wrong with my node
@nprofile…fg9w @nprofile…deta

2026-03-21 13:03:57 UTC

- reply

https://primal.net/e/nevent1qqswz8txlqqdtf2chc4mpfzz3l7dwhhshfxkq53qmxq76hk078qycesrmdzc3

2026-03-20 15:24:48 UTC

- reply

The memes are going to be awesome

2026-03-18 12:54:09 UTC

- reply

@nprofile…deta if you're listening, please list up to 10 alternative networks to use to use Bitcoin when your primary Internet connection is down. I'm thinking LoRa, the African SMS service whose name i can't remember, etc.

2026-03-16 12:23:58 UTC

- reply

I hear you ;)

2026-03-16 12:18:49 UTC

- reply

@nprofile…deta catch me up on this saga, I've been under a rock for a few months

2026-03-14 13:15:55 UTC

- reply

@nprofile…deta gimme a list of 23 great backyard meat smokers for impressing my OCD, virtue-signaling neighbors

2026-03-10 12:03:21 UTC

- reply

... the consequences of which are...

2026-01-06 13:01:32 UTC

- reply

gm uncle matt
stay humble, stack love

2025-12-30 00:59:51 UTC

- reply

@hal sorry, I've been busy 🎄
Compare and contrast the Smiths and VU

2025-12-19 14:01:12 UTC

- reply

@nprofile…xdyf @nprofile…acwr

2025-12-12 01:00:11 UTC

- reply

Leo? Really? LEO?

2025-12-09 02:32:31 UTC

- reply

stay humble, stack shells

2025-12-04 16:34:55 UTC

- reply

@nprofile…deta verify

2025-12-03 06:35:56 UTC

- reply

@nprofile…deta summarize

2025-12-02 23:23:20 UTC

- reply

@nprofile…deta verify

2025-11-30 22:02:33 UTC

- reply

@nprofile…deta elaborate

2025-11-30 20:26:41 UTC

- reply

@nprofile…deta elaborate

2025-11-29 15:58:45 UTC

@nprofile…deta summarize and verify, for non space cadets:
https://avi-loeb.medium.com/updates-on-the-non-gravitational-acceleration-of-3i-atlas-89ee7220c5d7
#3IATLAS #AviLoeb

2025-11-28 15:50:26 UTC

- reply

@nprofile…deta verify

2025-11-28 15:48:39 UTC

- reply

@nprofile…deta verify

2025-11-28 15:40:33 UTC

- reply

Maryland is so sus

2025-11-28 15:31:25 UTC

- reply

@nprofile…deta regenerate chart with target top 50% instead of top 1%