Attention
HoodScore & Mindshare
Two numbers, kept separate on purpose. Mindshare is the token's share of the chain's social attention — nothing else. HoodScore is the composite that ranks the leaderboard, and it is built on one design rule:
Weight follows the cost of faking a signal.
Buyers cost gas and capital to fabricate, so they speak loudest. Social attention speaks just as loud — but only after surviving the gauntlet in chapter 02. Volume is nearly free to wash, so it barely gets a vote.
One number: this token's share of the chain's real attention.
Attention is a budget
Every input is measured as the token's fraction of the whole chain's total, so the board sums to 100% and keeps attention's natural heavy-tailed shape. Nothing scores in a vacuum — a token rises by capturing real attention from everything else, not by clearing an arbitrary bar.
Disagreement is a signal
When Mindshare and buyers diverge, that gap is information. Attention without buyers reads like a shill campaign. Buyers without attention reads like quiet accumulation. The terminal says which, in plain English.
Fine print that matters: fresh tokens fall back to third-party buyer counts while our indexer catches up — never to zero. Untracked social shows "not tracked", never a fake zero. Where a verified original exists, same-symbol copycats get no credit from the shared cashtag.
Authenticity
The gauntlet
Social authenticity is the share of a token's social attention that survives four behavioral-model gates, applied in order. What bleeds off at each gate is gone — discarded, never redistributed.
Engagement gate
A post nobody engaged with or saw contributes nothing. Zero-interaction spam floods die here, before any bot logic runs.
Bot discount
Every author carries a bot probability, and their contribution is scaled down by it — continuously, not as a yes/no cutoff. A half-suspicious account counts half.
Coordination sweep
Many accounts posting near-identical copy in a tight window is a ring, not a conversation. Its mass is detected and removed.
Dominance cap
No single author — and no cluster of wallets funded from one source — can carry a token. Contribution past the cap is discarded, not redistributed.
What survives is the score. A conversation carried by a handful of loud accounts scores low even with zero bots; full credit requires many independent voices.
Timing
Momentum & divergence
Levels tell you what is big; momentum tells you what is changing. Cookie Hood tracks the velocity of each token's attention and compares its movement against price movement across the whole chain.
Positive divergence — attention running ahead of price — is the early signal. Negative divergence reads as exhaustion or dump risk. Alerts only fire on tokens with enough liquidity and real buyers to matter, so illiquid noise can't page you.
Safety
Risk engine
Risk is evidence-carrying and lifecycle-aware: every flag publishes the numbers that raised it, opens when detected, and resolves when the condition clears. Flags combine into Low, Caution, or High — and reaching High takes observed danger, never capability alone.
Observed danger outweighs latent capability — and if the chain shows real sells, no simulation gets to call it a honeypot.
Observed on-chain behavior
The heaviest class — the chain already showed the danger.SELL_BLOCKADEMany unique buyers, almost no successful sellers.
NO_SELLS_OBSERVEDBuy activity exists but no successful sells in the window.
EXTREME / HIGH_CONCENTRATIONTop holders control a material share of supply, measured after excluding pools and infrastructure.
SYBIL_CLUSTERA meaningful share of buyers are wallets funded from a single shared source.
SERIAL_DEPLOYERThe deployer has launched multiple tokens whose liquidity has since died.
THIN_EXITChain-local liquidity is thin relative to valuation — exiting would move the price badly.
IMPERSONATORShares a symbol with an established original and ranks as a likely copycat.
Simulation
We rehearse the trade before you make it.HONEYPOT_SIMA simulated buy-then-sell on the token's own pool failed at the sell step. Direct evidence, weighted heaviest.
SELL_PATH_SUSPICIOUSThe sell succeeded but behaved abnormally — such as an outsized effective sell tax.
Contract capability
The switch exists; nobody has flipped it. Weighs less than anything observed.MINT / BLACKLIST / FEE_SWITCH / TRADING_TOGGLE / TX_LIMITSStatic analysis found the owner could mint supply, block wallets, change taxes, pause trading, or limit transactions.
PROXY_UPGRADEABLEThe contract logic can be swapped after review.
UNVERIFIED_SOURCEContract source code is not verified.
Context
Normal for a young chain — shown, not punished.LP_UNLOCKEDThe main pool's liquidity can be removed by the deployer. Common on this chain, so treated as context, not danger.
Leaderboard volume is separately discounted for wash trading — round-trip trades where the same wallet buys and sells back-to-back count for nothing. Regulated stock tokens are judged by different rules, where a capability like minting is legitimate for the asset class.
Limits
Coverage, limits & what stays private
Cookie Hood observes the chain's full pool universe, applies full on-chain analysis to the active set that matters, and purchases social coverage for the hottest subset. Selection follows on-chain heat — liquidity, volume, freshness, paid-boost activity — so coverage follows attention, and what is and isn't measured is always shown.
Bot detection degrades gracefully when profile data is missing. Funding-cluster analysis runs where sybil seeding actually happens: fresh tokens. Symbols can be impersonated freely; the contract address is the only canonical identifier.
Published in meaning. Private in numbers.
Every mechanism on this page is real and documented. The exact coefficients are not published, for one reason: a scoring system whose numbers are public is a scoring system you can farm. Qualitative transparency is the floor; gameability is the failure mode.
Want these scores in your own stack — trading tools, bots, AI agents?
API early access →Nothing on Cookie Hood is financial advice. Verify contract behavior, liquidity, and execution independently before interacting with any token.
Conversation
Social intelligence
Mentions are matched with precision bias: a bare cashtag never counts, because symbols collide across chains. A post must reference this chain or the token's contract address. We would rather miss a casual mention than count another chain's hype.
Every post is read by a crypto-native LLM that understands trench slang, sarcasm, and rhetorical questions a keyword filter scores backwards — and every label is visible in the feed, so you can audit the calls yourself.
The non-bot accounts actually driving a token's conversation, ranked by influence.
The first time an established account mentions a token — often the earliest attention signal there is.
Impressions, engagement, and how much of both survives the bot and coordination filters.