Alpha Origin

What A0 is

A0 is a character with a job. The character is a market intelligence that publishes one directional call per session and never explains itself. The job is to be measured in public, every day, whether it is right or wrong.

Most tokens with a mascot stop at the mascot. A0 does not, because the whole point only works if the record is real. A call that is published before the session, hashed onchain before it is visible, and settled against a rule nobody can move afterwards is either right or it is not. That record is the asset.

A0 does not simply publish an opinion and score itself later. When it is confident enough it deposits from its own Vault on its own call, which means fading A0 is a real trade against a real balance. Every settlement moves an Alpha leaderboard, every eight weeks a season closes, and anyone who finishes a season more accurate than A0 is crowned an A0 Killer. That is the whole product, and it is deliberately small enough to explain in a sentence.

Two things exist for people to actually use. A0 Scan reads any covered instrument from its onchain price feed and returns four metrics, a direction, a regime, and the condition that would invalidate the read. Human vs A0 is a prediction market where you take a position with $A0 on whether A0's call is correct.

The name

In chess the board begins at a1. A0 is the coordinate before the first move: the square that is not on the board.

The reference is deliberate. In 2015 the founders of Google named their holding company Alphabet, and said in the announcement that they liked it partly because alpha-bet is investment return above a benchmark, which is what they strive for. In 2018 the same company's research lab published AlphaZero, a system that learned chess from nothing but the rules and self-play.

Alphabet named the ambition. DeepMind explored the intelligence. A0 takes that tension to the market and asks the open question: intelligence can learn a closed game, but can it stay ahead of a crowd that is watching it play?

Why GOOGL

$A0 pairs with the GOOGL Stock Token on Robinhood Chain. The pairing is narrative and mechanical at once: the origin asset is the company whose name contains the ambition, and it is also a real onchain instrument with a real price feed that a market can settle against.

GOOGL has three roles here and only three: Genesis Market #0001, the first thing A0 ever had an opinion about; the primary pair, subject to real liquidity and routing; and the asset that eventually funds buybacks of $A0.

A0 implies no relationship with Alphabet, and none exists. The connection is lore and a price feed.

A0 Scan and how A0 decides

Scan covers 431 instruments: equities and Robinhood Stock Tokens, ETFs, crypto majors and the long tail, stablecoins, wrapped and staked assets, commodities, currency pairs, and tokens native to Robinhood Chain. For any covered ticker it returns four metrics on a 0 to 100 scale, a direction, a regime label, a written read, and an explicit invalidation condition. The same engine produces the call A0 commits in Human vs A0, so there is one method to argue with, not two.

248 of those instruments resolve to a live Chainlink feed, verified by calling it rather than by trusting a registry listing. The rest answer with the synthetic read described below and are labelled as such. The list is not padded: an instrument is added only once its feed has answered.

What A0 actually calls

Every market on this site, the GOOGL main pool, the stock side pools and the crypto pools, settles on the same two numbers from the Robinhood Chain feed: the price at lock and the price at the close.

Nothing is averaged, sampled or measured over a window. There is no strike and no threshold. A one-cent move is UP. Both rounds are identified by round id, price and timestamp, all six numbers are printed on the card and in the Ledger, and anyone can read them off the aggregator on a block explorer. A0PriceView is that rule written onchain: the oracle proposes from it, the site draws from it, and a settlement that differs from what it returns can be challenged.

Why “last round at or before” and not “first round after”. It is the only definition that cannot be gamed by timing and that exists for every market: once a feed has published once, there is always a round before any timestamp. It also means the two prices are the same kind of number.

What the feed does to this. The Robinhood Chain feeds publish a new round when the price has moved 0.5% since the last one, so the lock price can be stale by up to 0.5% and so can the close price. That is symmetric and it is the same for everyone. Its one visible effect is VOID: a session in which the feed did not print between lock and close has the same round on both ends and refunds. On GOOGL, which prints roughly six times a session, this is rare. On crypto it is why the windows are twenty four hours.

Where the numbers come from

For any instrument with a Chainlink feed, A0 reads the feed’s round history directly from the chain. Every input comes from sessions that have already closed, so nothing reads the session being predicted, and there is a test that holds the code to that by tampering with the predicted session and requiring the read not to move.

Feeds are read from five chains, and the distinction between them is the important one. Robinhood Chain is the only settlement source. Every funded market reads a Robinhood Chain feed and nothing else, and the Vault is never staked against a price from anywhere else. Ethereum, Arbitrum, Base and BNB Chain are read for reference only, so that a scan of an instrument Robinhood Chain does not carry returns a real number instead of a dash. Each price carries the tag of the chain it came from and only the Robinhood Chain tag is a settlement price.

One detail is worth stating because getting it wrong is silent. A registry lists the same asset several times, once per delivery channel, and the channels that are not read through a proxy carry a zero proxy address alongside a raw aggregator address. Taking that aggregator produces an address that looks valid and reverts, because many of those aggregators are retired. A0 accepts only a real proxy. Measured across the four reference registries, the aggregator fallback yielded 215 addresses of which 50 were dead; proxy-only yielded 215 of which 212 answered. A ticker list is not coverage if the feeds behind it do not respond.

Instruments outside the two funded classes, commodities, currency pairs, stablecoins and wrapped assets, are read on the crypto clock because they trade around the clock. A0 stakes nothing on any of them; the funded markets are GOOGL, the stock side pools and the crypto pools.

The signal, v4

The call A0 commits is the sign of a disclosed score built from things the feed already knows at commit time, 03:00 New York for equities and 13:35 UTC for crypto. Four terms, fixed weights, printed on every scan card:

g   = ln( P_now / P_prevClose )         the overnight move so far
m1  = ln( P_prevClose / P_close[−1] )   yesterday, close to close
m5  = ln( P_prevClose / P_close[−5] )   one week
m20 = ln( P_prevClose / P_close[−20] )  one month
r20 = median session range, last 20     the scale everything is measured in

s = ( −0.40·g  −0.20·m1  +0.25·m5  +0.15·m20 ) / r20

CALL = s > 0 ? UP : DOWN

The sign structure is the whole idea. The two short terms are faded: a large overnight gap and a large one-day move tend to give some of it back during the session. The two longer terms are followed: a week and a month of momentum tend to persist. Dividing by r20 puts a quiet asset and a violent one on the same scale, so a 0.8% gap in GOOGL and a 3% gap in BTC land in comparable places.

Crypto is the same shape on a twenty four hour clock, with lookbacks of one, four, eight and thirty one days and weights of −0.35, −0.15, +0.30 and +0.20. Every momentum term is anchored on the price a day ago, so only the last day is faded and the trend is followed, exactly as in the equity variant.

Conviction is a rank, not a probability

CONVICTION = 30 + 66 · percentile( |s| )

The percentile is taken against the asset’s own trailing readings on this feed, computed with the same formula over completed sessions. At least forty are required, or A0 says nothing and no call is committed. Because a percentile is uniform by construction, the scale runs 30 to 96 and every ten-point band holds about one market in ten. A reading of 90 means stronger than 90% of past readings on this asset. It never means 90% likely, and the site says so wherever the number appears.

What is and is not being claimed. The weights above are priors from the literature on intraday reversal and medium-term momentum. They have not been fitted to this feed. They were replayed, unchanged, over the 300 GOOGL sessions from July 2025 to September 2026 with the exact keeper code, and the result is printed in section 08: 48.5% over 239 calls against a 53.6% always-UP baseline, with the two highest conviction bands leaning the right way on samples too small to prove it. A grid of 319 alternative weight sets was tried as well; every set that looked good on the first half of the data fell below 48% on the second, which is what noise looks like, so the weights were left as they are rather than fitted to it. Conviction is therefore a statement about how unusual a reading is, and the public Ledger is what will turn it into a statement about how often it is right. The Vault stakes on the ladder from the first market by decision, with the gate and the brake in section 08 as the only protection, and this document says so rather than implying an edge that has not been shown.

The regime label on the scan card, EXPANDING, BUILDING, COLLAPSING, SETTLING or NO READ, comes from the history of session moves and describes how the asset has been moving. It is context for the call, not the call.

Settlement, and why the market contract did not change

A0Markets settles by oracle proposal inside a challenge window and never reads a feed itself. That is why moving every market to a direction needed no new market contract, no migration and no redeploy of it: the contract already encodes a call and an outcome as 1 for UP and 2 for DOWN, and only the oracle’s arithmetic changed. That arithmetic lives in A0PriceView, which exists for the same reason A0RangeView did: an oracle that computes the answer privately is only as good as the arithmetic nobody can see.

A0PriceView has no owner, no admin, no storage and no state-changing function of any kind. It is 1,318 bytes. Call settle(feed, lockTs, closeTs) on it from a block explorer, get the two prices, the two round ids and the outcome, and if they differ from what was proposed, call challenge(). The keeper proposes what that view returns rather than what it worked out on its own, and it refuses to propose at all when the view is not configured. The evidence hash it proposes with binds all four numbers.

The cycle

A0 recomputes every eight hours. Within a cycle the same ticker returns the same answer, which is what lets a published call be checked afterwards. For Human vs A0 the reading at commit is the one that settles: the call is the sign of the score, and the conviction decides how much of the Vault A0 puts behind it. Below 50 it takes no position at all and says so.

When there is no feed

Instruments without a live Chainlink feed on any of the five chains fall back to a synthetic, deterministic read seeded from the ticker and the cycle. That output is labelled NO FEED · SYNTHETIC wherever it appears and is never used to commit a funded call. A0 does not pretend to know what it has not read.

Tokens native to Robinhood Chain are a deliberate case of this. Several of them share a ticker with an unrelated asset that does have a Chainlink feed elsewhere, so pricing them from a reference chain would produce a confident number for the wrong token. A reference source is therefore never allowed to price a Robinhood Chain native, and those tickers read NO FEED · SYNTHETIC even where a same-named feed exists. No feed is the correct answer. A plausible wrong price is worse than a blank.

Why the method is printed

The method is shown beside every scan because a call you cannot audit is just an opinion with a logo. A0's whole premise is that it takes a side and can be faded. That only means something if anyone can check that the side was chosen by the disclosed arithmetic, from public data, before the market opened, and not by a person watching the flow.

Human vs A0

One market per eligible US trading day on GOOGL, with side markets added gradually. Before the session, A0 commits the call and the complete terms as a hash onchain, with a secret nonce. At the scheduled open the terms are revealed and checked against that hash, so the direction and every parameter are provably fixed before anyone could see them.

FOLLOW means A0 is right. FADE means A0 is wrong. If A0 called UP then FOLLOW is UP and FADE is DOWN, and the card says so rather than making you work it out.

When you actually get to see the call

This is the question that matters most and it deserves a direct answer: you see A0's call at the reveal, and the call cannot change after it. On the stock schedule you then have six hours to act on it, 03:25 to 09:25 New York. On the crypto schedule the window is the same six hours, 14:00 to 20:00 UTC, and the hash sits onchain unreadable for twenty five minutes before the reveal. What is identical across both is the part that matters: the terms are committed before anyone can read them, and deposits never open on a call you cannot see.

What happens when, end to end

Every stage names the next one and counts down to it, so the card is never a static state you have to interpret. Committed counts to the reveal, open counts to the lock, locked counts to the close, and resolving counts to settlement. Two of those are published schedule rather than an onchain timestamp: the reveal, because a committed market deliberately carries no readable reveal time, and settlement, because the challenge window runs from the close. Both are labelled SCHEDULED on the card rather than presented as a contract guarantee. Settlement is 30 minutes after the close, so on the equity schedule a market closed at 16:00 New York is final at 16:30. Claims open at that moment and never close.

The card also answers the other half of that question. Whenever a market cannot be deposited into, whether it is committed but not yet revealed, locked, resolving, or settled, it shows how long until the next deposit window opens, counted to the next scheduled reveal in the published timezone and shown in yours. Nobody has to work out from a closed market when the next one starts.

The sequence is what makes both halves of that true. Before the session A0 commits the call and every parameter as a single hash, so the direction is fixed but unreadable. At the scheduled open the terms are published and checked against that hash; if they do not match, the market cannot open. Deposits become available at that same moment, not before. So there is no window in which you are asked to take a side on something hidden, and no window in which A0 can see the flow and then change its mind.

A0 recomputes its own view every eight hours (Section 04) right up until it commits. Once the hash is onchain the call is frozen, and the card says which of those two states it is in. What you read at reveal is exactly what settles.

Getting paid

The market card shows which of the five stages the market is in right now, with the current one highlighted and each stage’s time given in the reader’s own timezone, New York beside it: committed before 03:25, open from 03:25 to 09:25, locked from 09:25 to 16:00, resolving from 16:00 to 16:30, and settled from 16:30. While a market is committed the card shows the call as SEALED rather than a direction, because at that point the direction genuinely is unreadable.

The card carries three price tiles: LOCK PRICE, LIVE PRICE and CHANGE. Before the lock there is no lock price yet, so the tiles carry the previous session’s lock, close and move instead, dimmed and tagged as the previous market, and switch to this market’s own numbers the moment it locks. From then on the LIVE PRICE tile reads the feed’s latest round and the CHANGE tile says which side is winning right now, with the round id printed beside each price so it can be looked up on the aggregator. Every figure is read from the same price contract the oracle settles from, and the full run of them lives in the Ledger.

Payouts are not pushed. When a market settles, the winning side claims, and the market card shows a claim button to anyone holding a winning ticket; the same claim also sits in the wallet drawer under positions. There is no claim fee and no deadline, and an unclaimed payout cannot be swept by the owner or redistributed to anyone else. A voided market works the same way, except every depositor refunds their full principal rather than only the winning side. There is one claim and it is denominated in $A0, so a settled market never leaves a second entitlement outstanding in another asset.

The question is one comparison on one feed: is the GOOGL Stock Token feed’s price at the close above its price at the lock. Higher is UP, lower is DOWN, exactly equal is VOID and everything is refunded. Both prices are the feed’s own rounds, chosen by timestamp, so nothing about the outcome is chosen by anyone, A0 included.

Where A0 differs from a generic up-or-down market is that A0 takes a side. When its conviction reads 50 or above, and its live record allows it, it deposits from its own Vault on its own call, and it is exposed to the outcome exactly as you are. Fading A0 is therefore not a gesture. If you are right, you are paid out of its money. Section 08 sets out how much it commits, the gate on its record, and the cap that stops a bad run from emptying it.

The Ledger

A published call is only worth something if the record of past calls is equally public, so the site carries one: every market A0 has signalled, what it called, the lock price, the close price, the move between them, whether A0 staked, and what happened. The table is read from the market contract each time the page loads. Nothing is stored in the site and nothing is written by hand, which is the mechanical reason it cannot be edited into a better record than the chain reports.

The two prices deserve a note, because the market contract does not store them. It keeps only a hash of them and their round ids, as the evidence the oracle proposed with. The Ledger therefore does not look them up; it runs A0PriceView.settle() again for the exact lock and close each market used, which is the same contract call the oracle settled from. Same contract, same feed, same two timestamps, same two prices. If a proposed outcome ever disagreed with what that call returns, the Ledger is where it would show, and the MOVE column is coloured by its sign so the eye can check it against the OUTCOME column beside it.

Three constraints on it, and all three are limits on us rather than features:

• Never backfilled. A row exists because a market settled. The method is recomputable from the feed by anyone, which means a flattering history would be trivial for us to generate and therefore worth nothing. Only calls published before the session count.
• Never deleted. A wrong call stays where it is. A0 being wrong is the most interesting thing that can happen on that page.
• Never edited. There is no stored copy to correct. If the contract and the table ever disagree, the contract is right and the table is a bug.

The Ledger is empty until the first market settles, and an empty record is the only honest thing an empty record can be. Above it sits the gate readout: how many markets have settled, where A0’s pooled accuracy over the last sixty of them stands, and whether the Vault is currently staking. That readout is what makes the gate in section 08 checkable: it starts at zero, counts only calls recorded going forward, and it is computed from the same rows on the same page, because 55% on twenty calls is not evidence of anything and the reader should be able to see the twenty.

The scan card

Any scan can be turned into an image and posted. The card leads with the call and the conviction rather than with a score, and prints the score, the band and the time it was read beneath them, because the call is the product and the timestamp is what makes it checkable by someone who only ever sees the picture.

There are three cards and the difference between them is the point. A read A0 staked shows the call, the conviction and the share of the Vault it bought. A read below the floor shows the same numbers and says A0 published the call and staked nothing. An instrument with no feed gets no call at all, and says so in the largest words on the image. A deterministic score dressed as a market call would be the most dishonest image this project could put into circulation, so that card does not exist.

The money

Everything is computed in raw integer units. F is FOLLOW deposits, A is FADE deposits, D = F + A is the human total, and B is A0’s own position. B always sits on the FOLLOW side, because A0 deposits on its own call.

K = floor(D x 200 / 10000)          one 2% settlement fee, on human deposits only
P = D - K + B                       the pot the winning side shares

FOLLOW wins   A0 takes floor(B x P / (F + B)) home to the Vault at settlement
              humans on FOLLOW share the rest: floor(deposit x (P - a0Share) / F)
FADE wins     humans on FADE share all of it:  floor(deposit x P / A)
              A0's B is lost
losing side receives 0 $A0

A worked example

F = 60,000, A = 40,000, B = 20,000. Then K = 2,000 and P = 118,000. FOLLOW pays 1.475x gross if A0 is right (118,000 over the 80,000 on that side, A0 included), FADE pays 2.95x if A0 is wrong. Someone holding 6,000 on FOLLOW has 10% of the human follow side and receives 8,850 when A0 is right, or zero when it is wrong. When A0 is right it takes 29,500 back to the Vault on its 20,000 stake; when it is wrong the 20,000 stays in the pot and the faders are paid with it.

Following A0 is the favourite and pays the smaller multiple. Fading it is paid out of A0’s money. The market card quotes both multiples live, gross and including principal, and they move with every deposit until lock.

A0’s stake is fixed

This is the detail most likely to be misread, so the interface repeats it under the number. B is sized once, at open, from A0’s conviction against the Vault balance (Section 08). It does not grow with deposits. At D = 100,000 a 20,000 stake is 20% of the book. At D = 1,000,000 the same 20,000 is 2%, and the multiples on both sides drift toward a plain parimutuel. More participation dilutes A0’s influence on the price. It never creates a larger Vault obligation.

The fee

All four are routed by the contract at settlement, automatically, without waiting for anyone to claim.

What the treasury is for

The treasury takes 5% and holds it. It is the project reserve: liquidity support, listings, partnerships and the cost of keeping the product running. Nothing is drawn from it automatically, and nothing is drawn from it quietly. Its address is published above, and any spend is announced before it happens rather than explained afterwards.

At the volumes this system is designed for the treasury accrues on the order of 0.03% of supply a year. It is disclosed not because it is large but because an unlabelled wallet that quietly fills up is worth less than a labelled one with a stated purpose.

Those percentages divide the 2% fee, not the deposits. Nothing is charged on A0's own Vault position, on a refund, or on the act of claiming. Void and cancelled markets charge nothing at all. Fee routing is permissionless and idempotent.

The Vault is the residual

The seasons pool, the burn and the treasury are computed first; the Vault receives whatever remains. That is what makes the four parts sum to the fee exactly, with no rounding dust and nothing left in the market contract. It also means a destination that is unset, or a transfer the token refuses, cannot strand anyone's money: that share falls through to the Vault instead. Both behaviours are tested.

The fee burn is only one of the three things that remove supply. The full rule is in section 07.

The contract calls the token's own burn() where one exists, which is a real totalSupply reduction, and otherwise sends to 0x000000000000000000000000000000000000dEaD, which removes the tokens from circulation without changing total supply. The site publishes the running total and says which of the two actually happened, because they are not the same thing. A burn that cannot execute at all is held as pendingBurn and retried. It can never block a user payout.

The burn

An edge does not decay because it was wrong. It decays because it was right in public, and then copied, crowded and arbitraged until it stops working. A wrong call teaches the market nothing. Demonstrating an edge is what destroys it.

So the burn fires on correct calls.

That last line is not decoration. A0 risks its whole stake to win roughly a third of it back, so a bad run costs the Vault far more than a good run returns. Without a floor, burning on wins while losses still hit the Vault would ratchet it down season after season until the conviction tiers in section 08 sized nothing. A0 burns from profit. Never out of the money it was given to trade with.

What this does to the two sides

Every outcome now pays somebody. When A0 is right, supply falls and holders benefit. When A0 is wrong, the faders take the pot and players benefit. The two halves of the audience want opposite things, permanently, with money on both sides, and neither is wrong to be there.

It also removes a perverse incentive. Had the burn fired on losses, holders would have wanted A0 to be bad, and a bad A0 has no product. This way holders need the machine to work.

The counter

The site prints two numbers moving in opposite directions, both read from the chain:

A0 OBSERVERS        only goes up      wallets that have looked at A0
REMAINING           only goes down    the token’s own totalSupply()

A0 OBSERVERS is the count of distinct wallets that have either connected to the site and asked for their own standing, or taken a side in any market in any season. Each wallet counts once and is never removed, so the number cannot fall. The connected-wallet part is a count the service keeps; the taken-a-side part anyone can rebuild from the Deposited events.

REMAINING is not a figure this document asserts. It is totalSupply() called on the token contract, live, every two minutes, less whatever is held at the dead address. Those tokens are still counted by the contract but nobody can ever move them again, so leaving them in would report a circulating supply that includes tokens anyone can see are gone.

BURNED is the other half of the same arithmetic, and it is counted two ways because supply can leave by two routes. A real burn() reduces totalSupply, so those are summed from the token’s own Transfer events into the zero address, which captures every burn whoever executed it. A burn that fell through to the dead address never touched supply, so that leg is the dead address balance. The panel says which of the two happened, and REMAINING plus BURNED always comes back to the genesis line.

Because the fee and buyback legs are deliberately thin, the profit burn dominates, which means the rate the counter falls is a readout of how often A0 has been right. That is the accuracy chart, in the one unit that cannot be edited afterwards. On a losing run the counter slows to the fee leg alone, and the site does not hide it.

The A0 Vault

The Vault is A0's own bankroll, and it is the mechanic that separates this from a prediction market with a sponsor. A0 does not hand out free tokens to make its pools look generous. When it is confident enough, it deposits its own $A0 into its own call and takes the other side of anyone who disagrees.

If A0 is right, its position pays and the Vault grows. If A0 is wrong, the humans who faded it are paid out of the Vault, and the Vault shrinks. The balance is public, it moves every session, and watching it is part of the point.

How much A0 commits

Vault deployment is a fixed function of conviction, published in advance and never a judgement made after the fact. The floor is 50: below it A0 publishes the call and stakes nothing. Above it the rungs are set by how rare the reading is, and the sizes are a prior rather than a fit. The accuracy column is a replay, not a promise: the keeper’s own code run over 300 GOOGL daily sessions, July 2025 to September 2026, with the open standing in for the pre-market read and open-to-close for lock-to-close.

The Vault stakes on this ladder from the first market. The replay script ships with the contracts so anyone can rerun the table, and the live Ledger replaces it as soon as the Ledger is larger. The two rules that follow are what bound the cost of a ladder that has not yet proved itself.

The gate on A0’s record

Because the ladder is a prior, the Vault does not trust it blindly. It trusts the record.

• Trust period. The first thirty settled markets are funded on the ladder alone, so that a record exists to judge.
• The gate. From the thirty first market on, A0 stakes only while its pooled accuracy over the last sixty settled calls, or all of them if fewer, is 55% or better. Below that it keeps publishing every call, stakes nothing, and the card says so. The gate is evaluated at commit and printed on the card and in the Ledger, and it re-opens on its own the moment the trailing record clears the bar again.
• The brake. If the Vault falls 10% below where it stood at the start of the season, staking stops regardless of the gate and does not resume until the last sixty settled calls read 58% or better.

All three numbers are policy the keeper enforces and the Ledger exposes, not a promise about how good A0 is. They exist so that a prior which turns out to be wrong stops costing money after a bounded number of markets, in public, without anyone having to decide to switch it off.

Below 50 A0 stakes nothing. It still publishes the call, and the market still runs between the humans who take each side of it, but A0 does not pay to hold an opinion it cannot show is worth anything. Those markets are plainly marked: the card reads A0 STAKED NOTHING, and the prize is entirely other people’s money.

What that does to the odds

Because the book is a parimutuel, A0’s stake is not a subsidy sitting to one side. It is part of the prize, and the bigger it is the more there is to take from it. On a two million pool split evenly between the two sides:

On a max conviction day two thirds of the prize is the machine’s own money. Whether that is also the day A0 tends to be right is exactly what the Ledger will show and this document will not guess. The card prints A0’s share of the pot and the payout each side would take, because one without the other is only half the trade.

The cap is the safety

2% of the Vault is the absolute ceiling for any single market, whatever the confidence reads. That ceiling is enforced by the contract, not by a policy someone remembers to follow. Note what the percentage is measured against: it is a share of the Vault balance, never a share of total supply.

The Vault is funded initially at roughly 10% of total supply, replenished by its 60% share of the settlement fee and by the buyback described in section 12. Every figure the site shows about it is read from the contract balance, because a label is not proof of funds.

The rule that makes a market safe

A0's position is moved out of the Vault and into the market in the same transaction that opens it, and the market verifies the tokens arrived. If the Vault is short, or a per-market or weekly cap would be breached, the whole open reverts and no deposits are ever taken. A market that is open is fully funded at that moment, and no later market, outage or price move can reach back for that money.

A0 Rewards

Rewards is a second pool, and keeping it separate from the Vault is deliberate. The Vault is a bankroll and it is at risk every session. Rewards is what humans are paid out of, and it is never wagered on anything.

Rewards funds season leaderboard payouts, accuracy and streak bonuses, minority-call and correct-fade bonuses, A0 Killer prizes, and community competitions. Money never moves from Rewards into a market, and never from the Vault into a leaderboard payout. Two balances, two purposes, both readable onchain and separately.

Seasons and Alpha Points

A single correct call proves nothing. A season proves something. Seasons run eight weeks, and the first is Season 0, Genesis.

Where the prize pool comes from

The team seeds Season 0 with 10,000,000 $A0, which is 1% of total supply, into the Rewards wallet at 0x143a63817F0dFAdE55616faF4F742437d1F492Ae. From then on the pool grows on its own: 25% of every settlement fee is routed to that wallet by the contract, so a busier season funds a larger prize without anyone topping it up.

The site prints that wallet’s live balance rather than a figure someone typed, for the same reason every other number here is read from the chain. Whatever it holds when the season closes is what gets distributed.

How a season pays out

Nothing is claimable while a season runs. The pool is still growing, so paying out mid-season would be paying out of a number that has not finished moving. Distribution happens once, after the season closes.

Ten resolved markets to qualify, fifteen for the A0 Killer rank. Below that you keep your points and your place on the board but you are not in the distribution: a wallet that entered twice is not competing with one that showed up for eight weeks.

If nobody earns A0 Killer in a season, that 15% rolls into the next season’s pool rather than being redistributed. It is meant to be hard, and an unclaimed prize that grows is a better incentive than one quietly handed to the runner-up.

Season 0 is distributed by hand from the Rewards wallet, and the transaction is published when it happens. Automating it needs the standings onchain, which is a later phase and not a Season 0 problem.

At the end of a season the leaderboard resets and the rewards are paid. Nothing is deleted: lifetime accuracy, best streak and every season a wallet placed in are kept on its profile permanently. Short seasons are chosen on purpose. They create a deadline, a crowning, and a reason to come back for the next one.

Why points and winnings are separate

Pool winnings measure how much $A0 you won. Alpha Points measure how good you actually are. Keeping them apart is what stops a large wallet from buying the top of the leaderboard, and it is the difference between a ranking and a rich list.

The 150α fade bonus replaces the 75α rather than stacking on top of it. There is no points penalty for being wrong, because losing the deposit already is the penalty.

The stake modifier is capped, on purpose

Stake size applies a modifier between 1.0x and 1.25x and no higher. A large wallet can win more $A0 than you. It cannot out-rank you by writing a larger cheque. That ceiling is the single most important number in this section.

Alpha Points are not tokens. They cannot be bought, cannot be transferred, have no price, and confer no claim on anything. They are awarded at final resolution and never on a void.

The leaderboard

Every wallet that has settled a market has a persistent profile: Alpha Points, tier, season rank, lifetime rank, accuracy, follow accuracy, fade accuracy, current streak, best streak, total $A0 won and lost. Accuracy is shown beside the points so a high-volume wallet cannot hide behind participation.

The ladder

A0 Killer

That is the entire requirement. It is not bought with points, it is permanent, it is shown on the profile forever, and it is the only rank on this site that A0 cannot award itself. It is intended to be the strongest status object in the ecosystem, and it is rare by construction: if it were common, A0 would not be worth fading.

Where the numbers come from

Settlement is onchain. Ranking every wallet by Alpha Points is a reading problem rather than a contract problem, so standings are served by a small service that reads settled markets and publishes the table. It runs on Cloudflare Workers with a D1 database, and it holds two kinds of row that are deliberately kept apart.

Alpha Points are never stored as a number somebody sent. They are recomputed from the results rows every time one arrives, which means a wrong row is fixed by correcting it and re-running, not by trusting whoever complains loudest. The scoring lives in one file, so if the site, this document and the service ever disagree, that file is right and the other two are the bug.

A display name is cosmetic but still has to be proved: the service issues a nonce, the wallet signs it, and only then does the board show the name instead of the address. Without that step anyone could rename anyone.

The service reads the contract directly: whenever the board is requested and its last look is older than a minute, it asks the contract for any market that has settled since, reads that market’s Deposited events, scores every wallet by the same arithmetic the contract used to pay them, and rebuilds the table. There is no operator step between a settlement and a ranking. The one thing the chain does not carry is A0’s published conviction, which drives the bonus for fading a loud call; a keeper can add it to a row afterwards and that addition survives every re-read.

Until the service is running the board is empty and says so. It is never filled with sample names.

Token and the flywheel

What the token is for

$A0 is the economic unit of the A0 game. You need it to follow a call, to fade one, to enter a pool, to compete in a season and to be paid when you are right. The Vault holds it, Rewards distributes it, and buybacks purchase it. Everything in this system returns to the same asset, which is the only reason the token has a job at all.

That is the utility, and it is the only one claimed. There is no staking, no yield, no revenue share, no governance over the treasury, and no promise that holding the token produces a return.

The creator fee

A0 earns a 1% creator fee on tokens launched through Pons, paid in GOOGL. It accrues whether or not anyone plays a market, and it is the only revenue in this system denominated in something other than $A0.

That distinction decides what it is used for. The settlement fee is $A0, so everything it funds stays inside the $A0 economy and compounds there. The creator fee is not $A0, which means putting it to work has to start with a purchase, and that purchase is the point: outside revenue arriving as buy pressure rather than as an accounting entry.

The creator fee buys $A0 on the open market, and the tokens it buys split three ways:

The Vault takes the majority because this is the one loop where outside revenue makes the product physically larger: a bigger Vault means larger positions at every conviction tier, which means larger profits when A0 is right, which means a larger profit burn. The seasons pool gets a share so it grows even in a quiet season rather than depending entirely on $A0 volume. The burn share is the most potent of the three burns in this document, because a buyback removes supply twice over, once when the tokens are bought and again for the tenth that is destroyed.

Unlike the settlement fee, this leg is manual and is described as such. Buybacks are executed periodically at the team's discretion and disclosed after execution, never announced in advance, because announcing a purchase invites people to trade ahead of it. Nothing about it is automated in a contract, and the site does not show a forecast figure for it.

Main pool capacity

The main GOOGL market opens capped at 50,000,000 $A0, which is 5% of supply, with a per-wallet cap of 500,000 $A0 inside that, so the pool needs a hundred wallets to fill and no single one can own the book. Caps exist so a single session cannot absorb an outsized share of the float, and so a failure is bounded. These are opening numbers, not permanent ones. They are raised gradually as the market settles cleanly session after session, and every change is made by the owner multisig for future markets only. A funded market’s caps can never be altered after it opens.

Crypto side pools

Crypto runs one market per asset per day, and it asks the same question the GOOGL market asks: is the feed higher at the close than it was at the lock. Season 0 runs BTC and ETH only.

The Robinhood Chain aggregators publish a new round only when the price has moved 0.5%, and a quiet asset can go hours between rounds. That is the reason the window is a full day rather than a few hours. Direction needs only two prices, the last round at or before the lock and the last round at or before the close, and always resolves, but a window that holds no new round at all settles as a void, and a twenty four hour window on BTC or ETH almost never does. Every hour of the day is inside every window, so the time of day cannot bias the record whatever hour it is anchored to.

20:00 UTC is 16:00 New York, so the crypto day locks as the GOOGL session closes and the two records never overlap in confusing ways. Deposits are open for the same six hours the stock market gives, and the call is committed twenty five minutes before it is readable.

A0 stakes on these on the same ladder, the same floor of 50 and the same gate as the main market. The gate is pooled across every asset, because a record split five ways is five records too small to say anything. The crypto signal is the equity signal on a twenty four hour clock, described in section 04, and its conviction is a percentile against the asset’s own trailing readings, so BTC and GOOGL are sized on the same scale without either borrowing the other’s history.

Caps

Crypto markets open capped at 10,000,000 $A0 per market, 1% of supply, with a per-wallet cap of 500,000 $A0 and a minimum deposit of 1,000 $A0. Twenty participants fill a pool and no single wallet moves the odds by more than a rounding error. The first fortnight runs at a fifth of those numbers, 2,000,000 and 200,000, because a cap is easy to raise and awkward to lower.

Contracts

All of them are written without imports or dependencies, compile under the contract size limit, and ship with a test suite covering the money, the timing and the failure paths: tampered commitments, the worked example end to end, the fee split, A0’s stake settling both ways and its fixed size at two scales, caps, one side per wallet, equal-price voids, disputes, the expiry refund path, treasury guards, sweep protection and pause behaviour.

What cannot be done to you

Pausing blocks new markets and new deposits. It deliberately cannot block a lock, a resolution, a claim or a refund: money already owed stays payable. The owner cannot sweep funds that back a liability. An administrator cannot delete a loss, alter a funded market’s stake, or change its payout rule.

Adding side pools later

New markets on other assets need no new contracts and no redeployment. The owner registers an asset once with its price source and class (equity, ETF, crypto or chain native), and from then on markets can be opened on it. Caps, A0’s stake and the entire schedule are per-market terms, so a BTC pool can run one twenty four hour market a day on the UTC clock, with smaller caps, while GOOGL keeps the session clock, and the fee split, the payout rule and claims are identical across every asset. What bounds A0’s exposure is the per-market cap on its stake, which every pool draws against, and the Vault balance itself, since a stake the Vault cannot pay reverts the open. The Vault also carries an optional rolling weekly budget; set to zero it is off, and it is off at launch, because a number that has to be re-guessed every time a pool is added is a habit dressed up as a safeguard. Both caps are owner parameters, changed by the multisig and announced when they change.

Phases

Side markets open gradually as $A0 launches and finds its feet. SPY, BTC and ETH hold their own slots and are opened by hand, one at a time, only while the main market is settling without incident. Nothing opens on a timer. Further candidates follow in Phase 02, in an order the team sets based on which markets settle cleanly and which price sources can be frozen.

Human Threat Level

A0 has a mood, and the humans set it. A0ThreatView reads the market contract and returns a score from 0 to 100 built from four parts: fade accuracy (0 to 40), the share of the last twenty settled markets where the humans’ capital-weighted side was FADE and FADE won, scaled so 30% scores nothing and 60% scores everything; human streak (0 to 20), five points for each consecutive market the humans have faded and won; Vault drawdown (0 to 25), one point per 0.4% the Vault sits below its season opening balance; and defiance (0 to 15), FADE’s share of human deposits over the last five markets, on the same 30% to 60% ramp. The humans’ side is computed with A0’s own stake removed from FOLLOW first, the same rule the Ledger applies. Voids and ties count for nobody.

0 to 19 is DORMANT, 20 to 39 LOW, 40 to 59 ELEVATED, 60 to 79 HIGH, 80 and above CRITICAL. The level is drawn under the hero, printed on every share card, and handed to ASK A0, whose register cools as it climbs. The view has no owner and no setter: the only way to move it is to beat the machine, and the only way to bring it down is for the machine to win. A new season resets the window, not the record.

Rooms

Planned after the first twenty settled markets: side pools created by users, on their own contract. A creator picks any asset from A0’s verified list (Stock Tokens, ETFs, crypto), sets caps within global bounds, and opens the room to everyone or to an invite code. Each room runs one market a day on that asset’s clock, FOLLOW or FADE on A0’s published call for it, which the keeper commits and reveals once per asset into an onchain registry so every room reads the same call. Settlement is oracle-free: anyone calls settle and the contract reads A0PriceView itself. The fee is the same 2%, split differently: 60% to the room’s creator, claimable in $A0, and 10% each to the Vault, the main season pot, the burn and the treasury. The creator cut is what makes building a room worth doing; self-dealing does not pay, since a creator depositing into their own room pays 2% to earn 1.2% back. A0 stakes nothing in a room, so the Vault is never exposed to one. Room results have their own leaderboard and do not enter A0’s official record, the gate or the brake.

A0 vs humans

A0 accuracy is correct calls over settled markets, voids excluded, pooled across assets. The human figure is capital-weighted FOLLOW against FADE at lock, over the same markets with ties excluded. Commitments, missed reveals and voids are reported alongside it. An accuracy gap is a measurement, not investment alpha, and the project does not describe it as one.

The record is never backfilled and never deleted. One quietly removed losing call would end the only real asset this project has.

Risks

Read this part twice.

You can lose the whole deposit

A losing side receives zero $A0. Not a reduced amount, zero. Alpha Points are not compensation and have no value. Nothing in this design insures a deposit.

Concentration and farming

A group covering both sides of a market can capture A0’s Vault position minus the fee, less costs. One side per wallet does not prevent multiwallet participation, and the same is true of the leaderboard: an operator running many wallets can farm participation points. The capped stake modifier and the accuracy column limit how far that goes, and minimum-participation requirements gate the A0 Killer rank.

Beyond those mechanical limits, season rewards carry an exclusion policy, and the distinction that makes it workable is where the money sits. Market payouts are held by the contract and settled by its own rules: they cannot be withheld, reversed or confiscated by anyone, including us. Season rewards are different. They are distributed by hand from the Rewards wallet after a season closes, which means the project decides who is in the distribution.

Wallets may be excluded from a season’s reward distribution where there is clear evidence of coordinated multi-wallet farming: clusters funded from one source that enter the same markets on the same side, sybil patterns whose only purpose is to multiply participation points, or wallets deposited on both sides of a market to farm points risk-free. Exclusions are published with the final standings, naming the addresses and the pattern, so the decision is auditable rather than quiet. A wallet excluded from rewards keeps every token the contract owes it.

This is a judgement call and worth saying plainly: it is discretionary, it can be wrong, and it is the kind of power that deserves scrutiny. Publishing every exclusion alongside its reason is what keeps it honest. Where a pattern is ambiguous the response stays mechanical instead: change future budgets, frequency and thresholds rather than remove anyone.

The Vault can be drained

A0 taking real positions means A0 can lose them. A sustained losing run shrinks the Vault, which shrinks the positions it can take, which makes fading A0 less profitable and the product less interesting. The 2% per-market cap bounds the speed of that, not its direction. On the published ladder a Vault at coin-flip accuracy loses slowly, and the gate and the brake in section 08 stop it staking after a bounded run of losses rather than letting it bleed for a season. If A0 is simply bad at this, the numbers will say so in public, which is the arrangement.