# coldcard-weak-rng-research

Independent, defensive investigation into the COLDCARD firmware entropy failure
disclosed 2026-07-30/31 and the ~594.5 BTC coordinated sweep observed the same day.
Everything here is verifiable: real commits, real blockchain data, real compiled
binaries.

**[📄 Read the report (English)](docs/en/report.md)** · **[📄 Leé el informe (Español)](docs/es/report.md)**

## What this is

On July 30, 2026, roughly 594.5 BTC was drained from 500 Bitcoin addresses in about
fifteen minutes. The victims used COLDCARD hardware wallets, and the root cause was a
firmware bug that made supposedly random wallet secrets predictable. Coinkite (the
manufacturer) and Block published official advisories. This repository does not just
summarize those advisories, it independently re-derives the parts that can be
re-derived: the fix commits, the on-chain flow of the stolen funds, and the bug itself,
compiled and proven with real tooling.

## Three things you can check yourself

1. **The code.** [`evidence/commits/findings.md`](evidence/commits/findings.md) verifies
   the fix commits directly against the Coldcard firmware repository, including one the
   official advisory doesn't mention.
2. **The money.** [`evidence/onchain/`](evidence/onchain) traces the theft on-chain from
   scratch, starting only from the public consolidation address, not from any published
   transaction list. Result: 500 transactions, 1,324 UTXOs, ~594.5 BTC, matching public
   reporting. Every transaction ID is in
   [`drain-transactions.csv`](evidence/onchain/drain-transactions.csv).
3. **The build.** [`evidence/builds/`](evidence/builds) compiles the vulnerable and
   patched firmware from source and uses `arm-none-eabi-nm` on the real binaries to show
   which function the "random number" call actually resolved to, before and after the
   fix. See [`comparison.md`](evidence/builds/comparison.md).

## What's not here

No PoC that recovers a real seed, scans for vulnerable addresses, or moves funds that
aren't the researcher's own, and none ever will be. See [`DISCLAIMER.md`](DISCLAIMER.md).
A synthetic RNG simulator and a Bitcoin regtest demonstration are planned but not built
yet; open items are tracked honestly in [`RESEARCH_GAPS.md`](RESEARCH_GAPS.md), including
one significant one: the Mk3 fix version Coinkite announced (4.2.0) has no matching git
tag as of this writing.

## Sources

Every claim traces to a primary source (code, commit, or on-chain data) or a rated
secondary source. Full list with reliability ratings in
[`references/sources.yml`](references/sources.yml).

## How to cite this

```
Álvaro P., "The COLDCARD entropy failure and the ~594.5 BTC sweep: incident report",
coldcard-weak-rng-research, 2026-07-31.
https://github.com/alva-p/coldcard-weak-rng-research
```

## License

Code is MIT-licensed ([`LICENSE-CODE`](LICENSE-CODE)). Reports and documentation are
CC BY 4.0 ([`LICENSE-DOCS`](LICENSE-DOCS)).

## Security and scope

Found a vulnerability, or a secret accidentally committed somewhere in this repo? See
[`SECURITY.md`](SECURITY.md). Do not open a public issue with real seeds, keys, or
victim data.

<img width="1672" height="941" alt="image" src="https://github.com/user-attachments/assets/dedd6943-385e-4a29-b0c2-50af3bdab7f7" />