npm run watch is the same feed as a 120×40 terminal screen: launches on the left, the card on the right, every score logged before its outcome exists.
augur reads pons v2 launches on Robinhood Chain, scores them and ranks them. Today that happens on your own machine: you clone a repository and run a runtime, which rules out almost everybody. A hosted board is the next thing that matters, and it is not built.
Here is what gets built next, in order, with no dates. Anything marked planned above does not exist. If a page on this site ever describes something as working when it is not, that is a bug worth reporting.
The graduation model is validated on time-ordered folds and reports what it actually did, including the folds where it did no better than chance. The peak model ranks well and sizes badly, so it is shown as an 80% band rather than as a number. Both improve with more settled history, which is what the backfills are for.
Curves are read on demand, pools are swept wholesale, and enrichment has only covered the recent window. Every one of those gaps shows up as a dash somewhere on a card. The work is unglamorous and mostly spent waiting on rate limits.
A hosted board, then alerts. The local version stays free and stays the reference: the score you compute yourself is the score the site shows, which is the point of it being open.
Commits show what shipped and when. The model page shows what the model scored on data it had not seen, and the scoreboard grades claims that were recorded before their outcomes existed. If something here goes quiet for weeks, ask about it in public.
git clone https://github.com/Artemoon13/Ponscan-.git && cd Ponscan-npm install && npm run setupnpm run backfill -- --hours 26npm run enrich-window -- --hours 20npm run train && npm run validatenpm run boardnpm run watchnpm run nightlyOnly the official RPC serves eth_getLogs, and it has no websocket. publicnode has the socket and faster state reads but refuses logs. So detection rides the socket, log reads go to the official node, and enrichment reads transactions from publicnode, which keeps the log budget with the watcher.
Both are public and neither needs a key. Measured over a day the whole thing costs about 133,000 requests, roughly 1.5 a second, against endpoints that start refusing near ten. A long bulk read earns a 403 eventually anyway; that is a request to slow down, and the reader waits it out rather than failing.
| Key | Default |
|---|---|
| RPC_HTTP_URL | rpc.mainnet.chain.robinhood.com |
| RPC_STATE_URL | robinhood-rpc.publicnode.com |
| RPC_WS_URL | wss://robinhood-rpc.publicnode.com |
| POLL_MS | 300 |
| RPC_IN_FLIGHT | 3 |
| RPC_SPACING_MS | 60 |
| LOGS_CHUNK_BLOCKS | 60000 |
| DB_PATH | ./data/augur.db |
| BOARD_PORT | 4663 |
| BOARD_HOST | 0.0.0.0 |
| RATE_LIMIT | 240 |
BOARD_HOST=127.0.0.1 with TRUST_PROXY=1.Running npm run telegram puts the board in a chat. You create the bot yourself with @BotFather, put its token in .env, and it talks to nobody else: /watch 8 sets a floor, and every launch that clears it arrives with its score, its rank and its three reasons. /token 0x… answers with the card, /top with the highest scoring launches right now, /status with whether the watcher is still alive.
This is the one part of the project that speaks to a third party, so it is worth saying exactly what leaves: launch addresses and a number this machine computed, both public. No key, no seed, nothing about a wallet. It is a separate command you start on purpose, never part of the watcher.
The plan is a single room that acts as the tape: every $AUGUR buy lands in it the moment it clears the pool, with the wallet, the size and the price. The bot answers questions in the room, and in a direct message it opens Pro for anyone who can prove the holding.
The holder check is a signature over a plain sentence, the kind every wallet can produce without spending gas, and the sentence says in words that it moves nothing.
Nobody will ever message you first. Nobody will ever ask for your seed. No links to other tokens. There will be exactly one contract, published on the $AUGUR page, and every other address carrying this name is not ours.
The plan is that holding $AUGUR opens the hosted site up: more watches at once, more live feeds, the heavy reads without a challenge, and creator history over any window. Proving the holding is a signature over a plain sentence. A signature has no chain, so the wallet can sit anywhere, and the site never asks for a key, a seed or an approval.
Everything you are looking at. The board, the card behind every launch, the model and its validation, the prediction log. You run it yourself, it talks to a public RPC and nothing else, and the score you compute is the score the hosted site shows. That is the point of it being open, and Pro does not change it.
The board runs on a node and a model that costs money to retrain every night. Rather than sell a subscription, the plan is a coin launched on the same factory the board watches, with its creator fee pointed at the cost of running it. Trading it is how the board stays free.
That is the whole arrangement. No roadmap for the token itself, no unlock, no share of anything, and nobody holding a reserved allocation.