Transparency
Transparency
Method, limitations and conflicts of interest, stated openly.
How the lab works
A scanner reads the chain, a scoring engine ranks what it finds inside comparable cohorts, and agents turn a narrative into a token. Every transaction is built from an explicit intent, verified against that intent by a separate signing service, and written to an append-only audit log.
- 01
Ingest
Every trade and migration on the tracked launchpads is decoded straight from the chain.
- 02
Features
Trades become rolling windows: flow, buyers, holders, liquidity and who is trading.
- 03
Cohorts
Each token is compared with tokens of the same launchpad, age and time of day, as percentiles.
- 04
Score and state
Percentiles feed a capped score with explicit risk penalties, and a state machine tracks acceleration, breakout and exhaustion.
- 05
Outcomes
Every alert is measured at 15 minutes, 1, 6 and 24 hours, and the hit rates are published here.
Launch record
30 days
Creator fees
Accuracy of published calls
Every call the scanner raised, measured at fixed horizons. The misses are counted too.
- WatchCalls 88
- 15 min
- 19%77 measured
- 1 hour
- 31%78 measured
- 6 hours
- 45%76 measured
- 24 hours
- 44%71 measured
- HotCalls 12
- 15 min
- 11%9 measured
- 1 hour
- 11%9 measured
- 6 hours
- 22%9 measured
- 24 hours
- 44%9 measured
| Level | Calls | 15 min | 1 hour | 6 hours | 24 hours |
|---|---|---|---|---|---|
| Watch | 88 | 19%77 measured | 31%78 measured | 45%76 measured | 44%71 measured |
| Hot | 12 | 11%9 measured | 11%9 measured | 22%9 measured | 44%9 measured |
A call counts as a hit when the token rises at least 50% within the horizon. Window: the last 30 days. Cells show hit rate and, below it, how many calls were measured.
Our wallets
The only addresses we operate from. Anything else claiming to be us is not.
- Launch walletView on explorer
- Fees walletView on explorer
Launch Wallet holdings
Holdings per token are not published at the moment.
What we will not do
No fake engagement
We do not buy followers, likes, replies or views, and we do not run bot networks on our own posts. Social numbers you see on our accounts are whatever they actually are — including when that is embarrassing.
No wash trading
We do not trade against ourselves to inflate volume, we do not run market-making bots on our own tokens to manufacture activity, and we do not coordinate buys to fake a breakout. Our initial buy, if any, is disclosed on the token page.
Conflicts stated up front
We create the tokens we write about and we earn creator fees on them. That conflict is real, it is not resolvable by a disclaimer, and it is why the raw numbers are published next to the commentary.
Public data is delayed
The public site lags the internal platform and publishes bands instead of scores. This is deliberate: a live scanner attached to our own positions would be a signal service, and we are not running one.
Disclosures
Signing keys are held by a separate signing service. No AI model ever sees or uses them.
Public data is delayed by 15 minutes.
We create and hold the tokens published in this section. Our analysis is not investment advice.
Data sources
Solana RPC and websocket streams for on-chain events, launchpad programs and APIs for curve and migration state, market-data providers for prices and liquidity, and public social and trend feeds for narrative signals.
Known limitations
Holder counts come from swap and transfer activity, not from a full token-account crawl, so they are an estimate. Bot classification is a heuristic and gets fooled. Percentiles depend on the cohort, and thin cohorts are noisy. Anything derived from social platforms is subject to their rate limits and gaps.