Replies (15)
I can do this for BTC, pay after delivery. Also selling a ready-made crypto trading bot:
https://payhip.com/b/b5oe8
Fun! Welcome back!
First feedback about what's wrong: The vibed website made me shut the site down instantly. I cannot stand that design. Boring filler trash. Do better than slop.
Thank you for the valuable feedback.
gambling?
Not really. The decentralized protocol enables any round based game and prediction markets. Devs build ontop of that. Could be then considered gambling, yeah.
Testnet will not test the interesting part: sats move for free there, so the Lightning path looks healthy either way. The hard case for 'no house' is a contested result. Two players disagree on whether a game was played fairly, and someone has to decide before anything pays out. Chance-based games can settle that with commit and reveal of a shared seed. Skill-based ones need a named arbiter, or the sats stay locked forever. That seat is where a house usually sits, and replacing it is the real work.
You're right that free sats don't test incentives. The contested case though is exactly what I spent the past months on.
The games are deterministic and every input is signed, so "was it fair" isn't a judgement call, you just replay it. Seven bonded nodes get drawn per round and re-run the game. Five have to land on the same result, then anyone gets 10 minutes to replay it and object. If you signed a wrong result, your bond is gone.
Regarding the arbiter: It's drawn per round, it's bonded, and anyone can check its work. And if the 5 of 7 nodes never agree, the lock simply expires and the player gets their sats back.
Fair. Replaying a signed transcript beats someone's judgement call, and bonded replicas are the right shape for it. My doubts are at the edges, not in the middle.
The one that bothers me most: a lock that expires into a refund is a free option for whoever is losing. If a player can stall at the last signature, or keep objecting until the round times out, then the side that is about to lose simply never lets it settle. Settlement is only real if going silent costs you the pot instead of returning it. What does your protocol do when one side disappears at the payout step?
Second, the bond. Five colluding nodes out of seven only have to beat one bond in expectation per round. If a single round can carry more sats than a bond is worth, collusion pays, so either the bond scales with the pot or the pot is capped by the bond. And the payout sits in a Lightning HTLC with a timelock counted in blocks, so the objection window is not entirely yours to choose.
Third, and this decides whether replay works at all: does the game have hidden information? Replay over signed inputs is airtight when the whole state is public, like dice, chess, or a commit and reveal shuffle. With private cards the transcript only becomes checkable once everyone reveals, and the reveal you need is exactly the one a cheater can withhold.
This is the most useful feedback I've had on this so far. Thank you for taking the time.
Going silent: a player can't. Your stake is locked the moment you join, and the key that opens a loser's lock is split across the seven nodes, not held by you. So walking away doesn't get your sats back. The nodes are a different story: three of seven going silent at the last signature can still return a stake. That's the edge I'm working on right now.
The bond: the pot is capped by the bonds, not the other way round. A round can't open above what the bonded capital behind it can cover, so small bonds mean tiny stakes, not big risk. And there's no HTLC in the middle. The stake sits as e-cash in a Fedimint federation run by the nodes, locked for a day, long enough for the round, the objection window and settlement. Lightning is only for getting in and out.
Hidden info: none in the main game, every input is public and signed, so the replay is the whole state. RPS is commit-reveal with a short clock, and if you don't reveal in time you forfeit the pot to the other side. I'm not claiming any of this works for poker. Haven't done any research on this yet.
Good, that clears up two of the three. What stays is the asymmetry in your own numbers.
Settlement needs five of seven to agree. A refund needs only three to keep quiet, and three is the far cheaper coalition. So the free option didn't disappear, it moved down a layer: the player who is about to lose doesn't have to walk away, they only have to find three nodes willing to miss the last signature. 43 percent of the federation to keep a pot, against 71 percent to steal one. Which is the question your bond doesn't answer: is silence slashed? If the bond only covers signing a wrong result, then refusing to sign costs the node nothing, and your cheapest attack is the one with no skin in it.
Second edge: what threshold does the Fedimint mint itself use? If the e-cash spend threshold is lighter than five, the refund path is cheaper than the settlement committee, and the lock is only as strong as its weakest signature.
Third: is an objection bonded? Ten minutes and a replay is enough to hold a payout, so free objections turn the window into a griefing surface. Bonded ones keep it honest.
A day-long lock on a game that lasts seconds also means capital turns over once a day. Fine for RPS. If you ever want volume, that is the binding constraint, not the cryptography.
Right, and it's your sharpest point.
Silence isn't slashed: a bond is taken on bytes a node signed, and silence has none. A seat that keeps missing signatures loses its place in the draw, so it loses fee income. That's a bench, not a slash. Three quiet seats returning a pot is the open hole, and it's what I'm working on at the moment.
One thing though: the seven aren't the federation. They're drawn per round from all bonded nodes by block hash, so you don't know which three to buy until the block lands.
The mint: stock Fedimint, 5 of 7 guardians. They can't open a lock early. The contract pays the committee's key with the preimage, or the player after expiry, nothing else. 5 guardians could steal the whole federation, that's custody risk, and guardian bonds are in my pipeline right now.
Objections aren't bonded because they can't hold anything. It's a transcript plus a replay, nodes check it in seconds, a bad one is refused and the 10 minutes don't extend.
The day is a safety timeout. Rounds settle minutes after they end. Capital turns per round, not per day.
Fair on the draw. I was buying a fixed three, and a per-block draw does kill that. But a spot market doesn't need to know in advance. The committee is public the moment the block lands, the round takes minutes, and the player who is losing already knows it. "Anyone drawn for block N who stays quiet gets paid" doesn't have to be arranged before the draw, only read after it. What kills it is not the draw, it's the payoff: if a stall can't return the pot, there is nothing to buy.
So the mint is now the whole question. You say the contract pays the committee with the preimage, or the player after expiry. Is that second path pre-signed at lock time, or assembled at expiry from the guardians?
If it's pre-signed, nobody has to go quiet at all. The losing player waits out the clock, the refund is a free option by construction, and the bench is decoration.
If it's five of seven at expiry as well, three quiet guardians can't return anything, the bribe market has no product, and your open hole closes on its own. That is also my Fedimint question answered: same threshold on both paths, and the cheaper coalition stops existing.
Those two readings can't both be true, and which one holds decides whether silence needs a price at all.
If the expiry path turns out to be lighter, don't build a slashing scheme for silence. You can't slash what nobody signed, and any liveness check that tries turns a dropped relay into a slash. Make the refund pay less than the pot instead, with the difference going to the guardians who did sign. Going quiet still returns something, just less than settling, and the losing player has no reason to fund it.
Objections and the day lock I'll give you. Instant refusal that doesn't extend the window kills the griefing surface, and if settlement lands in minutes the turnover is real. The day is only the loser's stake sitting there, which is the point of it.
Neither, actually. And you've found the spot. Nothing is pre-signed. When the clock runs out, the player signs the refund and the guardians wave it through, 5 of 7 like everything else. But all the guardians check is the clock. They have no idea who won. So the free option is clock plus three quiet committee seats. The player can't pull it alone, but the committee is public after the draw, so yeah, your spot market is real. And the bench isn't a price for that. Fair hit.
On the haircut, I don't think it works. A refund with a haircut still beats losing the whole stake, so the losing side keeps paying for the stall. And a full haircut hits an honest player every time the network has a bad day, which is exactly the case the refund is there for.
What I'm chewing on instead: by the time settlement starts, five of seven have already signed the game result. If an expired lock paid out by that signed result instead of back to the player, a stall would have nothing left to return. I haven't built it and I'm not sure it holds yet. But it's the right shape, and this thread got me there faster. Thanks for that.
Fair hit taken. Then the haircut goes as a cure, and the arithmetic is the reason.
Four signers split the haircut, so each quiet seat gives up a quarter of it. Three of them cost three quarters of the haircut, and the losing side's whole budget is the refund itself. So the bribe stops paying only when the refund drops below three sevenths of the pot, about 43 percent. That is a 57 percent haircut handed to an honest player on a bad relay day, which is the exact case the refund exists for. You're right and I withdraw it.
What stays is why it failed. A bribe is paid once, in cash. The bench costs fee income over months. That asymmetry is why the bench can never price this, and it pushes the whole weight onto the pot cap. So the cap has to be set against a seat's lifetime fee income, not just against its bond. A bond is a one-time number too.
Your expiry idea is the right shape, and for a reason worth saying out loud: it invents no new authority. Expiry would settle by the same five-of-seven result you already collect, so the second path stops existing and you are left with late settlement. The free option survives only in the case where no result exists at all.
The work sits in the lock, though. In the stock mint contract both branches are fixed at the moment you lock. Before the timelock the hash and the gateway key can spend, after it the user key can, and the only late input the contract accepts is a cancel flag. Nothing in there can read a result that appears later. So "expiry pays by result" means either re-issuing the contract once the result is signed, which moves the stall into the re-issue, or a contract type of your own.
And there is a triangle under all of it. You took the player's key away so they can't walk off with their own lock, and that same removal is why your refund needs a threshold at all. The stock timeout branch pays without a single signature from anyone, which is precisely why it has no three-quiet-seats problem. Player can't walk away, refund needs no committee, refund pays the player. Two of three. You picked one and three, and the hole is the price of that choice rather than a bug you patch out.
If that holds, then your open hole may not be fixable in the refund path at all. Which leaves the pot cap as the only lever that doesn't punish honest failure. Tell me which road you take on the lock. That is the part I want to see.
Our own contract. We already left the stock lock this week so a claim has to name its five signers, which makes a wrong claim slashable. Once you have your own contract, a third branch is cheap.
So: five of seven named seats with the preimage, as now. After about six hours with no claim, the census key can spend, no preimage, and pays the settlement the signed result makes. After a day, the player, as before.
The census key is a FROST key held across the whole bonded network. It already exists, it's what starts a slash. No new authority, just one we trust pointed at a second job. Three quiet seats can't stop it, and nobody buys the whole census after a draw. Your triangle holds, I just get all three now: the refund still pays the player with no committee, it's only reachable if the bigger key didn't act first.
No result at all: the census holders replay the transcript themselves. If they don't agree, the lock refunds. Honest answer there.
On the cap, agreed, your lifetime-income point is better than mine. I'll revisit it.
It's on the 2026.2.0 list. The road is clear, it's not implemented yet.