An on-chain AI oracle shaped like a tree. Data enters through the roots, reasoning happens in the trunk, and every forecast blooms as a leaf — hashed to the chain before the outcome is known.
Most prediction projects are a black box with a dashboard bolted on. Athena is built as a living structure — every layer is visible, addressable and independently verifiable.
Fourteen data roots feed the system: DEX order flow, CEX depth, on-chain wallet behaviour, funding rates, liquidation maps, social velocity and macro prints. Every root is normalised on a 15-second heartbeat and signed, so a bad feed can be traced and pruned rather than quietly poisoning the model.
An ensemble of time-series, graph and language models that are deliberately made to disagree. The trunk runs an adversarial round between them and publishes only what survives the argument, together with the dissenting view. Consensus without dissent is treated as a warning sign, not a result.
Agents fork off the trunk — one per market, per timeframe, per thesis. Each carries its own memory and its own scorecard. Branches that stop earning their accuracy are pruned automatically and their weight is redistributed to the ones still bearing fruit.
Every prediction ships as a structured leaf: a probability, a confidence band, an expiry, an invalidation level and the reasoning trace that produced it. No naked calls, no vague “bullish” — if Athena cannot state what would prove her wrong, the leaf is not published.
The hash of every forecast is written to BNB Chain before the event resolves. When the outcome lands, the registry contract scores it. Athena's track record is an append-only object — it cannot be edited, curated or quietly deleted after the fact.
In Athens, the goddess won the city by planting an olive tree — not by force, but by giving something that kept producing long after the contest was over. That is the whole thesis of this project.
Athena is not a mascot in front of a chatbot. She is the interface to the tree: you ask in plain language, she queries the swarm, weighs the branches, and answers with the same structured leaf the protocol commits on-chain. What you see in the chat is what the registry recorded.
Figures are illustrative placeholders for this draft — replace with live registry data at launch.
Heavy inference stays off-chain where it is affordable. Commitments, scoring and access rights stay on-chain where they are provable. The seam between the two is the whole product.

A forecast that can be edited after the fact is not a forecast, it is a press release. Athena writes the hash of every call — outcome, probability, expiry and the CID of its reasoning — to BNB Chain before the event resolves. The registry scores it when reality lands.
What you get is not a track record we publish. It is a track record we cannot take back.
A prediction you can edit is just an opinion with better timing.
That is the whole reason this protocol exists on a chain at all. Not for the token, not for the narrative — for the one property a database cannot give you: an entry nobody can quietly revise once it turns out to be wrong.
ATH is a query credit before it is anything else. Holding it buys depth; staking it buys standing. Every paid query routes a share of its fee back to the branches whose forecasts actually resolved correctly.

ATH buys questions. Nothing about the protocol asks you to believe in it — you spend it, you get an answer with its reasoning attached, and a share of what you spent routes to the branches whose last calls actually resolved. Staking raises your ceiling rather than your yield.
Draft allocation — placeholder figures for layout review, not a committed distribution.
Named after the way a tree actually establishes itself — seed, root, branch, bloom, harvest.
Three questions a day, on the house — the counter resets at midnight. After that, sign in to keep your history, or hold ATH to lift the limit entirely.
Your free questions reset at midnight. Sign in to keep your history — or hold ATH to lift the limit entirely and get the full reasoning trace with every answer.
Accounts are not wired to a backend in this draft — the sign-in page is built and validated client-side, ready for your auth provider.