Hacker News Reader

Comment by nullc | Hacker News Reader

I (Greg Maxwell) am the one of the two "Bitcoin core developers" mentioned in the MtGox press release. My source is Mark Karpeles directly.

I do not know how much they lost. Most of my discussion with them was before I think they knew exactly how much they lost. I had assumed, by the nature of the issue, that it wasn't likely to be much. I'm a little less sure of it based on their behavior since this weekend and due to finding out that there software was automatically issuing reissuing transactions that it didn't think had been paid out.

View on HN

An enterprising blockchain-spelunker might be able to put a very rough ballpark estimate on amounts lost by any affected repeat-payers, by:

(1) Find all confirmed transactions that have a signature that appears in non-canonical form (and thus likely confirmed under an unexpected TXID);

(2) For those transactions, identify the paid-to addresses and amounts: possible targets of make-up transactions. (Of course, it may be hard to distinguish true targets from 'change'.)

(3) Find later transactions with the exact same paid-to addresses and amount: these may be erroneously-issued repeat payouts.

Of course, if the complaining user offers a different address for the make-up transaction, this wouldn't work. On the other hand, a researcher already working hard to correlate affiliated addresses, now or in the future, still might be able to surmise when duplicate-amounts went to affiliated-addresses in succession during the active exploitation period.

> (1) Find all confirmed transactions that have a signature that appears in non-canonical form (and thus likely confirmed under an unexpected TXID);

Unfortunately, you can't— MtGox had a long standing bug where they would author transactions which were themselves non-canonical form (they encoded the variable length values in DER as fixed length with excessive padding). This is one of the reasons their transactions were getting stuck.

People mutating their transactions removed the padding, making them canonical.

Aha, interesting. Seems someone could still run a similar process in reverse: if MtGox's successful make-up transactions are non-canonical, find those first, then look for canonical precursors, within the suspected-exploitation timeframe.

That seems more prone to false-positives, though, and perhaps after some point the flood of copycat non-canonicals becomes a problem. (On the other hand, if MtGox was fairly unique for a while in issuing a particular kind of non-canonical transaction, it could make mapping their affiliated addresses by some larger effort easier.)

Some researchers might have records of most of the alternate transactions that were circulating without being confirmed. For example, I don't know how comprehensive Blockchain.info's double-spend report – https://blockchain.info/double-spends – is, but I imagine TXID-mutated variations might appear as double-spends there. It's currently reporting 1126 detected double-spends - over 1000 in the last 3 days, but before then, quite rare.)

Reply on news.ycombinator.com