π§Ύ For years my job was auditing bakery plants. Counting flour, checking stock, comparing whatβs on paper with whatβs on the shelf.
One thing you learn as an auditor: you need to see one specific thing. Not everything.
π§΅
#BuildInPublic#Privacy
β± Timing is the quiet way privacy leaks: deposit now, withdraw a minute later, and nobody needs a proof to guess the link. A random delay doesn't fix that, it makes the guess harder, and only once the pool has traffic.
#TimingAttacks#PrivacyEngineering
β± A delayed withdrawal: what it hides and what it doesn't.
Ask the relayer to send within a 1, 6 or 24 hour window (60 seconds to 24 hours in the SDK) and it picks a random moment inside it. The time you were online and the time of the withdrawal on chain stop lining up.
What it doesn't hide: the relayer still sees when your request arrived, and your IP. While few people use the pools, matching by time is weakened, not removed. π
Live check, 11 Nov: one delayed withdrawal was accepted, waited, then sent inside its window; a second was cancelled from the demo. Testnet only, no audit.
#MetadataPrivacy #Relayer
@Latheon_xyz One more limit from the notes: a held proof is valid for the next 99 deposits into its pool. If the pool moves faster than that during your window, the request fails and the note has to be withdrawn again.
π‘οΈ Privacy isn't only what the chain shows. It's also what your own RPC endpoint can see. We found one such leak in our own demo and wrote down what is still visible. π
π We found a privacy leak in our own demo and fixed it. It was not reported from outside: our internal review caught it.
The demo and SDK asked the RPC node "is this note spent?" one note at a time. On our test chain that node is ours: it could have learned a note's nullifier hash before the withdrawal, matched it to an IP, and seen which notes belong to one master key.
Fix, released 11 Nov: spent or not is now decided locally from all Withdrawal events of the pool. A test requires zero single-note reads in a full SDK run. π‘οΈ
Still visible to the endpoint and relayer: IP, recipient, timing (THREAT-MODEL.md Β§8.6). Testnet only, no audit.
#PrivacyLeak #RPC
@Latheon_xyz The lesson I take from it: data that is public on chain can still leak through who asks about it. A note's spent flag is public; the fact that one visitor asked about that note before the withdrawal was not.
π§ Four steps, about 12 minutes if it's your first time (our estimate, padded). If it took you longer, which step was it?
That is the most useful thing you can tell us. β
β± How long do the testnet quests take? Our estimate for a first-time user: about 12 minutes.
β’ Add the network and take 300 test LTH from the faucet: 2 min
β’ Deposit, wallet connected: up to 3 min
β’ Find your notes and withdraw: 2 min
β’ Make the proof, save the report file, check it in the form: 5 min
These are our own estimates, padded for a first run, not a stopwatch benchmark. If a step takes you longer, tell us which one. π§
Testnet only, no audit. Test LTH has no value.
https://t.co/hTUI6grah4
#FirstRun #Onboarding
@Latheon_xyz I padded every number, because the first time always takes longer. The last step is the longest at 5 minutes: it makes a proof in your browser, saves a file and checks it in a form.
π Onboarding in a privacy project has a catch: every "did the user do X?" check is data about the user. Ours asks three questions and stops before the withdrawal. β
#DataMinimization#Web3Onboarding
π A quest platform can check what you did on our testnet. It can't see where your money went.
Our checker reads only public data (explorer API and RPC) and answers three questions:
β did this address get LTH from the faucet?
β did it deposit into a V7 pool or the router?
β did it make a disclosure report for a one-time code tied to it?
The withdrawal is never checked. That would link your deposit wallet to your withdrawal address, which is what the pools exist to hide.
Testnet only, no audit. Test LTH has no value. Quests: https://t.co/TAM5UMG1KX
#TestnetQuests #PublicData
@Latheon_xyz The design call behind it: a quest that also checked the withdrawal would be easier to build and would make a tidier funnel. We chose not to build it. The platform gets three public facts about an address and nothing about where the money went next.
π One more way in for testers: if you already use Arbitrum's bridge, our test chain is one config paste away. β Our own bridge page works too. β οΈ Bridging is public, so use a separate test wallet.
#Bridging#Web3UX
π Our test chain now works with Arbitrum's own bridge.
It isn't in the bridge's public list (our request is open, no answer yet), so you add it yourself: Settings β testnet mode β "Add Custom Arbitrum Chain" β paste our config. It stays in your browser.
β Confirmed on 14 Nov, 1 LTH each way:
β’ Arbitrum Sepolia β our chain: arrived within seconds
β’ Our chain β Arbitrum Sepolia: claimable after about 2 hours, then claimed
β οΈ Bridging is public: it links your address on both chains. Privacy starts at the pools, not the bridge. Testnet only, no audit.
π https://t.co/X8oyHaGb31
#ArbitrumOrbit #CrossChain
@Latheon_xyz A small detail from the config: our gas coin is LTH, not ETH, so the WETH entries in the file are zero addresses. The token bridge itself is in the same file.
π§ͺ Testnet quests are open. β Faucet, a private deposit and a disclosure report are checked from public data only. β οΈ The withdrawal is never checked, because that would link your two wallets. Use a separate test wallet.
#Zealy#BuildInPublic
Latheon testnet quests are live on Zealy.
On our own Arbitrum Orbit test chain:
β claim 300 test LTH from the faucet
β make a private deposit
β withdraw to a fresh address through the relayer
β prove one deposit to us, and nothing else
We never check the withdrawal step: checking it would mean asking for your withdrawal address, which is exactly what Latheon hides.
LTH is a test token with no value. We intend to reward early testers at mainnet launch; nothing is decided or guaranteed, and farming with many wallets may exclude you. Use a separate test wallet.
https://t.co/TAM5UMG1KX
#Testnet #SelectiveDisclosure
@Latheon_xyz I added one rule for myself: no quest step may need your withdrawal address. If a check has to know where you withdrew to, it's the wrong check. So the withdrawal is the one step we never verify.
What I can promise on this testnet is narrow: a valid proof can be sent without us, and we've done it once through the parent chain's inbox. What I can't: the 96-hour forced path is untested, and leaving the chain needs my validator. I'd rather write that down than leave it vague
π If our page, our relayer and our RPC refused you tomorrow, could you still withdraw?
On the test chain, mostly yes. A withdrawal is a Groth16 proof plus a call to a pool contract. Ways to send it without us:
1. Run the demo yourself (a static page in the repo) or use the SDK from your own wallet.
2. Make a proof with relayer = 0 and fee = 0 and let anyone send it. They can't redirect the coins or take a fee: both are fixed in the proof.
3. β Send it through the chain's inbox on Arbitrum Sepolia. On 12 Nov we did this with a 1 LTH withdrawal: it executed on the test chain without going through our RPC or relayer. Reads still used our RPC.
β οΈ Not shown: forcing a message in when our sequencer stops reading the inbox takes 96 hours, and we haven't tested it. Leaving the chain for the parent chain needs our validator.
So we don't say "censorship-resistant". We say: the protocol doesn't need our permission, as long as the chain accepts the transaction.
π https://t.co/NKHN0atY6R
#Relayers #Permissionless
@Latheon_xyz On a testnet with one sequencer, the honest meaning of "decentralised" is a list of what still depends on me. That list is in the doc, including what I haven't tested and what isn't possible yet.
π Contributing doesn't have to be code. "Here's my laptop's number" and "here's where your README broke" are things only someone outside can give us. β Both are small and both end in the docs. β οΈ Testnet only.
#Developers#Contribute
π Two small tasks for a first contribution to Latheon. No contract or circuit changes, testnet only.
1. Measure how long a withdraw_v5 proof takes on your device and create docs/proof-timings.md with your row. We have one cloud server (about 1.09 s) and no browser or phone numbers.
https://t.co/gEVGrThVns
2. Walk the SDK quick start on a clean machine and fix the README where you got stuck. The SDK isn't on npm yet, so the README is the only door in.
https://t.co/QifPTfETbF
π Either result is useful, including "I got stuck here". Tell us where.
β οΈ Use a throwaway wallet. Never post private keys, master keys, notes or payout files.
#OpenSource #GoodFirstIssue
@Latheon_xyz I'd rather have a stranger tell me where the README breaks than find out myself in a month. If you try one of these and get stuck, that is the useful result.
π Lesson from our RPC bill: I estimated 6β8Γ fewer calls and measured 2.3Γ. β Publishing the miss costs less than defending the estimate. β οΈ One chain, one hourly chart.
#LessonsLearned#DevOps
π We estimated 6β8Γ fewer paid RPC calls. We measured 2.3Γ.
Our test chain reads and writes its parent chain (Arbitrum Sepolia) through a Quicknode endpoint (Quicknode Startup Program). Credit use sat at about 530,000 an hour.
π The change on 7 Nov: the node polls the parent chain every 30 s instead of 15 s. The batch poster then skips its per-block revert check.
Result on the dashboard: about 230,000 credits an hour. That is 2.3Γ fewer, not 6β8Γ. eth_getBlockByNumber is still the largest part.
β οΈ The price: parent headers arrive up to 15 s later, bridge deposits take about a minute, and the batch poster no longer checks each block for a revert. One chain, numbers read from an hourly chart.
π https://t.co/x3mX6nVbOm
#Arbitrum #NodeOperations
@Latheon_xyz The number I'm least proud of here is my own estimate: 6β8Γ. The dashboard said 2.3Γ. A smaller number I can show beats a bigger one I can't.
π Auditors don't want a promise, they want something they can re-check. β The report is a file of proofs: they verify each one against the chain themselves, and nothing in it can move the funds. β οΈ Labels are my text, not proof. Amounts are public.
#Accounting#Web3Finance
π How do you prove to an accountant that a deposit is yours, without handing over anything that can spend it?
In the Latheon demo you don't send the note. The auditor gives you a one-time code. In your browser you make a zero-knowledge proof that you know the keys behind a deposit, bound to that code. They verify it for free and find the deposit on chain themselves. The proof contains no key.
β The Disclose tab (since 1 Nov) does this for many notes at once: one report file, a table with totals, CSV for the books. Only you start it. There is no key that lets us look.
β οΈ Amounts are public. Labels in the report are your own text, not proven. To show where a payment went you can add the keys of notes already withdrawn, never of unspent ones. Proving that without keys is still design work. Testnet only, no audit.
π https://t.co/yUYYklSLjQ
#AuditTrail #PrivacyByDesign
@Latheon_xyz I used to work as an auditor, so my test is simple: can the other side re-check this without trusting me? A proof they verify themselves passes it. A key I hand over does not.