# Brief: gMAK "rug tiers" post as tests (Ben, 10 Oct 2026)

Source: x.com/gmakcrypto/status/2108497193640345984 (9 Oct 2026, 16k views). Claims are big ("99.9% rug", "10x more in a month", "without fail") and come from one trader's experience. Treat every line as a hypothesis to measure on our chain, not as truth. Survivor bias warning: he describes what worked for him.

## The claims, mapped to things we can measure on Robinhood Chain

**Tier 1 rug (he says 99.9% rug):** socials 1-5 days old · 0 mutuals · anon team · no revenue.
**Tier 2 rug:** tweets "DEX paid" · beta play the day after a main runner · template ("slop") website · GitHub repo 1 day old, no history · launched on a pump.fun-style launchpad.
**Keep only:** doxxed dev · revenue from real use · established GitHub · socials with mutuals / smart followers · already pumped, retraced, and the dev keeps shipping.

| Signal | How we measure it here | Data owner |
|---|---|---|
| Social age, followers, smart/known followers | X handle from token metadata or GMGN/DexScreener token info; account creation date, follower count, overlap with a list of known trench/KOL accounts | dataeng (spec below) |
| Anon vs doxxed, revenue | Not measurable at scale. Proxy for revenue: contract has fee/treasury flows from users other than the deployer. Skip doxx. | analysts decide |
| "DEX paid" | DexScreener paid profile/boost flag per token (public API: orders + boosts endpoints), time it was bought vs launch | dataeng |
| Beta play after a runner | On-chain, already in our data: launched within 24 h after a $1M+ runner, with a name/ticker/meme similar to it | analysts (have the data) |
| Template website | Website URL from metadata; fetch once, hash the HTML structure, cluster identical templates; site/domain age | dataeng |
| New GitHub | GitHub link from metadata; repo created date and commit count (public API) | dataeng |
| Pump.fun-style launchpad | Our launcher/kit label (we already have it per pool) | analysts (have it) |
| Pumped, retraced, dev still working | On-chain: peak ≥5x, then retrace 60-85%, dev wallet has NOT sold, and dev keeps acting (LP adds, contract calls, metadata or socials updates) | analysts + dataeng |

## Tests to build

1. **Rug filter as an arm on existing tests (Analyst 2 + Test Manager).** Score every coin Tier 1 / Tier 2 / clean. Re-run the 5-Min Cap replay and the paper tests with a "skip Tier 1" and a "skip Tier 1+2" arm. Question: does skipping them cut losers more than it cuts runners? Our entries are in the first minutes, so many signals (website, GitHub, DEX paid) may not exist yet at entry: record WHEN each signal becomes known and use only what was known at entry (no look-ahead).
2. **Beta-play avoid rule (Analyst 2, now, data exists).** Coins launched within 24 h of a runner with a similar ticker/name/meme: their rug rate, runner rate and mean x vs all coins. If clearly worse: a skip rule, straight into paper.
3. **Second-wave entries (Analyst 1, new family).** Not first-minute: buy coins that already pumped ≥5x, retraced 60-85%, dev has not sold and is still active; hold hours to days with a wider stop. Measure on the last 30 days of pools. This is his "keep only" list turned into a strategy and is different from everything we run today.
4. **Quality score (later).** Once dataeng has the social/website/GitHub fields, build a 0-5 quality score and test it as an entry filter and as a size multiplier.

## Data spec for dataeng (one shared table, all new tokens since 13 Sep)
token, pool, created_at, x_handle, x_created_at, x_followers, known_followers (from our KOL list), dexscreener_paid (bool, time), boosts, website_url, website_template_hash, domain_age_days, github_url, repo_created_at, repo_commits, telegram_url, first_seen_time_of_each_field. Free APIs only (DexScreener public, GitHub public, GMGN where we have it). X data: only what is reachable without paid API; say what is not.

## Done means
Each analyst posts a verdict per claim (holds / does not hold on our chain, with numbers). Anything that holds goes to the Test Manager the same day as a new arm or test, and gets a tab on the tests page.