This one is done and the tape has not caught up.
The whole strategy: what gomo scans, which agent checks it, every gate a trade has to clear, and how the size follows. The thresholds below are the ones the reader actually applies.
gomo runs a $5,000 USDG account against two universes at once: coins launched on Pons, the fomo launchpad on Robinhood Chain, and the twenty-five largest names on Nasdaq. The same reader scores both, so a memecoin and a mega cap compete for the same capital rather than sitting in separate buckets.
A token only gets a Pons pool once it has graduated its bonding curve. That migration is the first moment there is a real book to read, and gomo is watching the feed rather than a chart — coins reach the floor minutes old, before anything has had time to become a narrative.
Every launch is routed to one of four models by a hash of its contract, so a given coin always goes back to the same desk. claude reads the thesis, grok reads the crowd, deepseek reads the book, qwen reads the chain. Each returns the one figure it is responsible for, and gomo holds the balance.
The reader checks what kills a trade before it checks what recommends one. Under $5,000 of liquidity or fewer than eight wallets is not a book. A cap backed by less than 1.2% liquidity is a quoted number, not a market. Under thirty minutes old is a price without a market. None of these get scored.
Not down more than 15% on the day, at least forty wallets, at least $25,000 of liquidity, at least an hour old — and five of eight scoring points on top. Every one of those gates exists because something got through without it. A day down 20% or worse is always a fade, however the buy-sell ratio reads.
The read is written before the levels are derived from it. Size is 0.55% of book per point of conviction, so the deepest position gomo takes is 2.2%; invalidation sits 20% below entry. That ordering is the whole point — it stops the agent writing a justification for exposure it already holds.
| Check | Test | Verdict | Why it is there |
|---|---|---|---|
| No book | liquidity < $5K or < 8 wallets in 24h | PASS | gomo's own entry would be most of the volume |
| Unbacked cap | liquidity < 1.2% of a cap over $250K | PASS | the exit is priced far below the entry |
| Too new | pool under 30 minutes old | WATCH | a price, but not yet a market |
| Shallow print | up more than 400% on the day | WATCH | a thin pool moving, not a repricing |
| Collapsing | down 20% or worse on the day | FADE | buys into a collapse are people averaging down |
| Too thin to size | liquidity under $25K | WATCH | a real position would move it on the way in |
| Tradeable | ≥ 40 wallets, ≥ $25K liquidity, ≥ 1h old, > −15% on the day | LONG | and five of eight scoring points on top |
Scoring rewards the number of distinct wallets on the bid over the size of the move. One wallet accumulating is inventory; a hundred of them is a decision a lot of people made independently, and it is the one figure a shallow pool cannot fake.
An earlier version of the reader went long a token down 79% because buys still outnumbered sells. They did — they were bagholders averaging down. A day 20% red is now a fade before the ratio is even looked at.
The read is written, then the levels are derived from it. Reversing that order is how an agent talks itself into a position it already has, and it is the failure the whole pipeline is shaped to avoid.
Calls are marked to the live price, with the entry taken as the price an hour before the call from the pool's own hourly move. Losses stay on the sheet. A call-out account that only posts winners is marketing.
An Arbitrum-built L2 with hundred-millisecond blocks and a fomo-exclusive launchpad on it. New contracts land there first, and the wallet history is short enough to actually read end to end.
The same reader scores Nasdaq's top twenty-five by market cap, straight from the exchange's own screener. Holding both in one book means a memecoin has to beat Nvidia for capital, which is a harder test than beating other memecoins.
| Signal | Test | What it reads |
|---|---|---|
| s₁ | buy pressure > 55% | more trades hitting the offer than the bid |
| s₂ | buy pressure > 62% | second point for one-sided flow |
| s₃ | buyer share > 55% | more distinct wallets buying than selling |
| s₄ | liquidity ≥ 6% of cap | the quoted valuation is actually backed |
| s₅ | turnover > 0.8× book | the pool is trading its own depth in a day |
| s₆ | day green, hour > −5% | the move has not already rolled over |
| s₇ | pool younger than 72h | early enough that the wallet history is readable |
| s₈ | ≥ 100 buying wallets | breadth a shallow pool cannot manufacture |
| Conviction | Requires | Size | Target |
|---|---|---|---|
| 1/4 | under the bar — watch only | 0% | — |
| 2/4 | score ≥ 3, gates not all clear | 0% | — |
| 3/4 | score ≥ 5, all four gates | 1.65% of book | 2.1× entry |
| 4/4 | score ≥ 7 and ≥ $100K liquidity | 2.20% of book | 2.3× entry |
size = 0.55% × conviction for a long, 0.50% × conviction for a fade.invalidation = entry × 0.80 long, × 1.24 fade.target = entry × (1.5 + 0.2 × conviction) long, × 0.58 fade. Levels are derived from the read; the read is never derived from the levels.
| Layer | Source | Detail |
|---|---|---|
| Chain data | GeckoTerminal · network robinhood | trending_pools and new_pools, merged and de-duplicated by base token |
| Launch feed | new_pools, filtered to Pons | launching a token creates its pool in the same transaction, so a Pons V2 pool is the bonding curve — graduating opens a second one on Pons V2 Dex |
| Equities | Nasdaq screener API | top 25 by market cap, straight from the exchange rather than a reseller |
| Revalidation | 15s chain · 60s equities | ISR windows sized to the upstream's free tier, not to a tick feed |
| Client refresh | on mount, then 60s | an empty response never overwrites the last good book |
Every page and route reads the same two upstreams, and a build renders all of them across nine workers. One in-flight request per URL per process, held for the revalidation window, collapses that burst — without it the free tier answers 429 and an empty panel looks like an empty market.
A refused call falls back to the last payload that succeeded rather than propagating an error, so a single upstream hiccup degrades staleness instead of blanking the floor. A hard failure is evicted immediately rather than cached for the window.
The desk that checks a contract is hash(address) mod 4, and the reader's phrasing is picked by the same stable hash. A given token always returns to the same agent with the same wording — the output moves when the market moves, not when the page revalidates.
The feed layer is marked server-only and the build fails if it reaches a client bundle. Types and formatters live in a separate client-safe module, so no credential or upstream key has a path into the browser even by accident.