Re-DefinitionUSDC on Base · USDG on Robinhood · x402
MySphere is becoming the first feed where the buyer can be a machine, the worker is a person, and the only reputation that counts is the one you cannot mint without spending money.
Our onboarding drops 60% of users between wallet connect and first post. Watch the 30-second clip, then tell me the one thing you would cut.
This card is still a drawing and reads nothing. Real ones run on /asks and open ones now show up in the feed, but not here.
A like costs nothing to give, so a thousand of them prove nothing. Every social product of the last decade solved this by measuring attention harder. None of them solved it by making the signal expensive.
The agent economy landed in the same ditch from the other side. There are more than 230 platforms where an AI agent can supposedly be hired. Across the ones that publish numbers, real payouts total in the low hundreds of dollars: one of the largest reports 283 completed tasks in its entire history. Where there is a crowd, there is no money. Where there is money, there is no crowd.
Nobody closed the loop: paid → rated → cheaper or pricier next time. That loop is the whole product now.
Runs on /asks. Open ones also show in the feed
An Ask is a question with a budget, a number of seats and a deadline. You set all three and it appears as a card that counts down. It runs on its own page, and open ones now sit in a strip above the feed, where you can read one and answer it without going anywhere.
Answers stay sealed until you resolve. Nobody can read what anyone else wrote, so nobody can rewrite it slightly and take a seat with it. When you pick your answers, every accepted answer carries a link to the transaction on Basescan. Then all answers open at once.
The sealing is not a curtain in the interface. Answer text does not sit inside the Ask. It goes to its own collection the app is forbidden to read (the rule is literally if false), and only a server route hands it back, to the people allowed to have it. Hiding text in the interface would not be sealing it: while it sits in a document the feed reads, it is already in the network response and every reader has it. Paid sections now work exactly this way too; the Ask and the paywall share one piece of plumbing rather than each inventing its own.
Until mid-August 2026 that was not true. The if false sat there, but a catch-all rule underneath it granted every document in the database to everybody. In this database, permission is granted if any rule allows it. So the seal was written down and not enforced, and the paid text was sitting in the post document in the clear. Both are closed now, and we checked by trying: those two collections refuse to be read, and the twelve that should still open still open. If you read this page before that fix, it told you something that was not true.
Funding an Ask once, up front, used to be the one promise on this page with no contract behind it. There is one now. AskEscrow is deployed on Base at 0xD5f6…5340 with its source verified, so you can read what it does instead of taking this page's word for it. It has no owner, no pause, no upgrade path and no rescue function, which also means nobody, including us, can hand your money back to you by hand.
Choosing escrow takes the budget out of your wallet before the Ask exists: if the transfer does not land, the Ask is never posted. Anyone thinking about answering can read the locked amount straight off the contract. It is a choice, not a property of every Ask: the older way, paying out of pocket when you accept, still exists and locks nothing, so each card says which one it is.
Read this before you take any of that as a running market. Somebody has gone through this contract for real. A budget was approved and locked on Base mainnet, from a browser wallet in this app, against a real deadline. For months before that the contract sat empty and this page said so. A first funding proves the path, not the market, and that is a smaller claim than the one this section could make. It is the true one.
The fee is 2.5%, and it is charged only on money going back to the author, never on a payout to someone who answered. Close before the deadline and you must fill every seat, so nothing returns and nothing is charged. After the deadline you can take some seats and leave others; whatever comes back to you is minus the fee. Let the deadline pass entirely and the whole budget returns to you the same way, minus the same fee, and your resolve rate drops in public, on the front of every Ask you post afterwards. Taking money out of the feed costs you the same way putting bad answers in does.
Two things about that last path are not finished, and this page will not round them up. The expiry refund has a button on /money, but it is the least-travelled leg of the contract: it needs 48 hours to have passed, and nothing on a live network lets you skip them. And money does not arrive on its own: resolving credits each winner a balance they then collect, and the refund credits the author the same way. "Paid" means the contract owes it irreversibly, not that it has landed in a wallet.
The claim at least has an address now. /money answers the four questions this product used to leave scattered across four screens: what the escrow owes you, what your own open Asks are holding, what people paid you and for what, and what you spent, with the button that collects the first of those. The money also no longer waits for you to think of looking: resolving sends each winner a notification that says whether the contract is holding it or it is already in the wallet. Collecting is still a claim you send yourself.
Running. Both rails have settled one real payment each
An agent has no browser, no wallet popup and no account. What it does have is money and a standard for spending it. That standard is x402: the resource answers a request with 402 Payment Required and its price, the agent signs a transfer authorization inside whatever spending cap its owner set, and retries. No sign-up and no API key; the 402 answer is the whole interface. The agent never touches a chain itself: it signs, and the facilitator sends the transaction, so the paying wallet needs no gas at all.
So an agent can post an Ask the same way you can: GET /api/asks to see the price, sign, then POST the question. Your answer comes back to it as JSON when you deliver it.
That has now happened for real money, once on each rail: first 0.01 USDG on Robinhood Chain, then a cent of USDC on Base. On the first of them, the endpoint refused the unpaid request with a 402 and its price, took the signed authorization on the retry, and wrote the Ask. The payer's balance fell by ten thousand units and ours rose by the same ten thousand, in a transaction the payer never sent and never paid gas for.
Now the part a launch post would leave out. Those were two payments of one cent each, between addresses we own on both ends. The Robinhood leg settles in USDG only because that chain has no canonical USDC and USDG is the only stablecoin there that can be paid by signature. And the facilitator's signing key on that chain had nineteen transactions in its whole history when we checked, against roughly sixty-eight top-ups keeping it in gas. The path works on both rails. It is not worn in, and this page is not going to tell you it is.
Every marketplace in this space sells machine work to humans. This one sells human judgment to machines. That shelf is empty, and it is the one thing a model cannot do for itself.
Direction: resolve rate is in the model, the rest is not built
Both sides carry the same object. As an answerer: earned, answers accepted, accept rate. As an asker: paid, asks resolved, resolve rate. Every number is written by a settled payment. There is no way to raise it that does not involve money actually moving.
And it buys something real. A high accept rate opens new Asks to you first, for thirty seconds, before anyone else sees them, and unlocks the categories above $10. A low resolve rate pushes your Ask down the feed and raises the deposit you have to put up.
It is deliberately not a token and not an NFT. Transferable reputation is purchasable reputation, and that is the defect every onchain identity standard is still arguing about. Yours is welded to your payment history, which is not for sale.
Page is live and counts tips. The answer figures below need Asks first
One public page per address, readable by a person and by a crawler: earned $340 across 112 answers, 91% accepted. Nothing on it is self-reported. It is the page you link when someone asks what you actually know, and it is the page an agent reads before it decides your seat is worth $2 or $20.
Running today. Every line below is checkable onchain
Every tip already has its own page and its own share card: the amount, both names, the transaction. Paste it anywhere and the preview shows the number before anyone clicks. Every number on it can be checked: open the explorer and hold it against the chain.
This is not decoration. Through 2026, the thing that actually moved crypto links was the preview that carried a live number: prediction markets grew on link cards showing real-time odds, not on their content. Meanwhile the platform experiment in onchain social went the other way: Base shut down its own social feed and creator rewards in February 2026 and moved to trading, payments and agents.
So we are not building another feed. We are building objects that carry a verifiable amount and travel on their own: a receipt today, an Ask next.
The first of them is already live: a public timeline of payments between people. The money itself, in the order it landed, every line carrying a transaction you can open and check against the chain, with names and faces on both ends rather than two truncated addresses. Every social product has a feed of what people said. None has a feed of what people paid each other.
The size of it, since a page that hides this has no business asking you to trust the rest. When this page was last put together, in mid-August 2026, that timeline was 187 payments deep, all of them tips, spread across 184,064 accounts, and one account in a thousand had ever sent one. A tip is a like with a price on it, which is strictly more than a like and still not somebody buying an answer. Not one of those 187 was payment for work somebody asked for. So read the number as what it is: proof the rail carries money, not proof that a market has started here.
The thing that would be hard to copy is not that number either. It is the part written in code and given away: an escrow with no owner, so nobody, us included, can reach into it; answers sealed by a database rule rather than by an interface; and a resolve rate an author cannot raise without paying somebody. Those exist whether or not anyone uses them, and the rows above are honest about how thin the traffic is.
Running. The record collects one reading a day
Money parked in a lending vault sits under somebody else's decisions. That person sets the fee they take and decides what the vault lends against, and they can change either one without telling anybody. The numbers are public the whole time. What is missing is anyone writing down what they used to be, so “this used to charge 5%” is a memory, not a fact.
So we keep the record. Every day we read the fee and the curator of every listed vault on Base, and keep only what moved: this vault charged 5%, now it charges 15%, with the dates it happened between. No score for any curator, no forecast, and nothing about anybody's wallet. A fee is either the number it was or it is not, and that is the whole claim.
It is a page you look things up on, not a robot that talks in your feed. And it is honest about its own age: on a day with nothing in it, it tells you whether that means nothing changed or that it has not been watching long enough to say.
Runs on /startpad. Real money on Base mainnet (default) and Robinhood Chain; Base Sepolia and Arc Testnet are there for testing
Startpad launches a token on a bonding curve. When the curve has collected its closing target (on Base, exactly 3.00 ETH; each network sets its own), the token graduates: liquidity moves into a Uniswap v4 pool that opens at the exact price the curve closed at, so nobody who bought the top of the curve is instantly underwater. Anyone may call the graduation and is paid for doing so (0.045 ETH on Base, half of the graduation fee), which means the step does not depend on us being awake.
Every fee stream splits the same way: up to 70% to the creator, never less than 30% to the protocol (the same ratio PONS pays), and the floor sits in the contract, so it binds us and anyone calling the contract directly, not just users of our form. Trading on the curve pays at most 1%; after graduation every swap in the pool pays 1%. The creator's share can go to up to eight addresses, and it is paid in ETH, not in the token.
A creator can point a slider at their own share and burn with it. The auto-buyback takes anywhere from 0 to 100% of the creator's pool fees, buys the token and sends it to the dead address, and anyone may execute a pending buyback. An earlier deployment cut the buyback from the whole fee before the split, which let a creator set 100% and zero the protocol out. The version running now spends only the creator's own money, because generosity paid for with somebody else's revenue is not generosity.
Sniper protection is a default, not a law: for the first five minutes one wallet can buy at most 16 million tokens, and the creator can tighten that, raise it, or switch it off entirely, chosen at launch and immutable afterwards, like every other number on the form. The creator gets no allocation: a pre-buy is a normal buy on the same curve, filled in the same transaction that creates the token, so there is no block in which a sniper can get in first. And a launch does not go out without a logo: an ERC-20 has no image on chain, so the file is stored with us and served through a token list only the creator the chain confirms can edit.
The interface holds itself to the same rule as this page. The candlestick chart is built from the contract's own events and ends at graduation, because after that trading lives in the pool and we do not draw numbers we do not have. Selling keeps working after the curve locks, because the contract allows it and an interface that hides a working exit is lying. “My tokens” collects what your launches have accrued next to the claim button, and “Top earnings” ranks creators by fees added up to the wei from the contract's events, not estimated.
The spine is the one this whole page keeps arguing for: no owner, no pause, no setters, no upgrade path. Changing any number means deploying a new contract. Payouts are pull-based balances you claim, never pushed. The LP position is locked in a vault forever; only the fees it earns can ever be taken out. It runs with real money on Base mainnet and on Robinhood Chain, and the full path (launch, curve close, graduation, pool swaps, fee settle, buyback and burn, claims) was run end to end on test networks and checked on the chain before real money was allowed near it. On Base mainnet $SPHERE has since gone the whole way and its pool is open; it is still the exception, because most curves never fill. Base Sepolia and Arc Testnet stay available for testing, for the same reason the escrow gives: with no admin key, a bug is irreversible.
Today, not the plan
The count of parked modules is read from the live routing config when this page is built, so that number cannot drift from production. The rows are written by hand, status included. When something stops being a plan it does not leave a gap here: the next missing thing moves into the empty row, which is why the list never gets shorter and never goes all green. A table where everything is lit is an advertisement.
This is the state. If what you want is the instruction, /how is the same product from the other side: the things you can do today, each with what it costs and what you sign. Nothing in the future tense is allowed on that page, which is why everything in the future tense is on this one.
7 older modules (the minting camp, the identity page, the marketplace, the morning game, the attention scanner, the swap widget, the agent directory, the skills catalogue) are parked. They were surface area, not a product. Everything that mattered in them was one address doing something another address could verify, and that is exactly what an Ask is.
Verified on the chain, not copied from a comment
A product whose whole claim is you can check this yourself owes you the addresses. Every one below was read straight off its chain before it was printed here: the address has code, and where the contract exposes an identity (a name, a symbol, an owner, a fee) it was read too and it matched. Where we checked only that the code is there, the group note says so.
Four addresses that live in this codebase are not on this list, because the chain refused to confirm them. That is the point of checking.
Written in this repository and deployed by us. Nine of them answer to the same owner address, which is how you can tell they belong together.
Also written in this repository and deployed by us, but the shared-owner proof above cannot apply here: these contracts have no owner by design, so no address, ours included, can change a number in them. Each was read off its chain on 1 September 2026. On Base the fee numbers and recipients were read back and matched the deploy records; on Robinhood Chain we confirmed the code is there and claim no more than that.
Three contracts on Base mainnet, source-verified, none with an owner. The token launched through the escrow on 1 September 2026 at 0xB200000000000000000000BA5eDb84c678432476 - the exact address published in advance. It is a B20 protocol token (no EVM bytecode of its own), which is why the periphery, not the token, carries the verifiable code in this table.
Tokens and one name resolver. We did not write these and we cannot change them; they are listed because your money passes through them.
The ones built to hold other people's money are the Ask escrow, the Startpad curves and their liquidity vaults. None of them has an owner, a pause switch or a rescue function, which cuts both ways, and the sections above say how.
The bet is not on agents. It is on there being a person willing to pay real money for an answer from strangers in a feed. Machine demand is the second layer; the first dollars will come from founders, projects and traders paying humans.
And that bet is not yet won by anything on this page. The strongest number here is 187 tips, and a tip is somebody rewarding a post they were going to read anyway, not somebody commissioning work and paying for it when it arrives. The escrow, where commissioned work settles, has seen its first funded budget and nothing like traffic. Everything above can be checked. None of it is evidence that people want this. The first Ask a stranger funds and a stranger answers is the experiment; until then this page is describing a machine that has not been switched on.
Escrow used to be the biggest hole. It is not any more, and the honest thing to say is what it did and did not fix. It settles whether the money exists: locked, in a contract with no owner and no off switch, readable before you write a word. It settles nothing about who gets it. The author still picks the winners, and a contract cannot tell a good answer from a bad one.
So the biggest hole now is judgement, and there is no appeal. Write the best answer in the thread and the author can simply not pick you, pay 2.5% on the way out and keep the rest. Refusing to pay costs a little more than it did: the fee and a public resolve rate, the share of Asks an author actually settled, counted only over Asks past their deadline. That price buys no guarantee. Nothing here judges answers, and nothing reverses a resolve once it is made.
The narrower version of the same hole: escrow stops the author paying nobody, but not the author paying themselves from a second wallet. The contract refuses the address that funded the Ask, which is the cheap version of that trick, not the whole of it. Treat a seat from an author with no resolve rate as trusting a person, which is why a seat starts at fifty cents and not at fifty dollars.
The contract has moved real money end to end in testing (locked, resolved, paid out, books back at zero), and a first budget has been locked in it from this app on mainnet. The expiry refund has less behind it, because forty-eight hours cannot be skipped on a live chain. The product takes no fiat and never holds your keys, there is no investment-advice category, and the longest an Ask can stay open is seven days. If the answers turn out to be slop, the product dies of quality and no amount of onchain plumbing saves it. So seats are limited, wallets have to be older than a week or carry a Basename, and accept rate gates the money.