What does Ethereum show about Lighter's November 16?
Answer
All of it, from Ethereum alone. The first drain adds up to the cent: 10,393,056.79 USDC paid to the attacker's address. The second, 5,096,588.02 USDC, was in the first batches Lighter posted after a 7 h 22 min pause, and was paid to a different address, not the attacker's.
- 7 h 22 minLighter posted no batch to Ethereum, 10:32 to 17:54 UTC
- 10,393,056.79USDC to the attacker's address, 10:11 UTC
- 5,096,588.02USDC of the second drain, paid to another address at 18:03
- 14.2 minmedian wait from a posted batch to its proof
How we measured
Lighter posts every batch's account changes to Ethereum as a blob and pays withdrawals out of one contract. We decoded the 29 blobs from 09:47 to 17:55 UTC from Lighter's published schema, followed the two victim accounts, the attacker's account and the pools they used, and read every USDC payout of the contract. Before trusting the blobs, we checked them two ways. Decoded positions match Lighter's API. And Lighter's snapshot of all 877,921 accounts hashes to the state root committed on Ethereum.
What else we found
- First drain: in batch 269191, account 714710 (−7,382,824.88), the pool made under its address (−3,010,001.00) and the attacker's account 769455 (−230.91) lose 10,393,056.79 USDC together. 17 minutes later the contract paid 10,393,056.79 USDC to 769455's address.
- Second drain: the same route moved 5,086,386.02 USDC from account 714723 to 769455, and 769455's whole balance, 5,096,588.02 USDC, left Lighter. The contract paid exactly that at 18:03 UTC to 0xb326…0a07. The attacker's address has received nothing since the first payout. Lighter says about 5M USDC was returned; who holds 0xb326…0a07 isn't visible on chain.
- Lighter stopped posting right after the batch where 714723's positions were closed, and the second drain is in the first three batches after the pause. The post-mortem doesn't mention the pause. That the pause was how the second drain was held back is our reading, not Lighter's statement. There was also a 2 h 28 min pause the day before.
- Batches are verified by a proof a median 14.2 min after posting (p95 19.8 min). Batches posted just before the pause waited up to 456.6 min.
What we take from it
Lighter's state can be checked from Ethereum without its API: usually minutes behind, hours behind during a pause.
If Lighter says more about the pause or the return, we'll update this post.
Limits
- Blobs carry each account's net change per batch, so the order of steps inside a batch doesn't show.
- The victims' large USDC losses while closing positions went to other Lighter accounts (the batch totals are near zero). They are not counted as stolen.
- The state root was checked at batch 166859 (2026-05-02). Since 2026-09-20 the root includes a market data tree, a layout we haven't reproduced. 6 root replacements in 2025 had no proof; the 13 since each passed a SNARK verifier, whose circuits we haven't checked.
Measured with an AI (Claude). It wrote and ran the scripts, and every number here is a line of the committed output.