# a2: per-pool rug tier table for #336 (#602), 10 Oct 2026
Script it39/t.py, table pool_rugtier_proxy.csv (27,272 pools, e1x-runners-v2, 21 Sep-9 Oct).

## Verdict: the real Tier 1/2 score cannot be built yet
The tiers need X account age/mutuals, DEX-paid, website template, GitHub age. None of that exists on any box (no dataeng social/metadata table; checked /opt/ppcopy, pub dirs). Beta-play is already tested: does not hold (fw/a2-gmak-betaplay.md). So no skip-Tier1 arm can be run today. Request for the table stays with dataeng (spec in mgr-gmak-rugtiers.md).

## On-chain proxy tried (do NOT use as an arm)
Proxy T1 = token with >=10 pools (fee-ladder spam) or creator cluster with >=3 bundled launches; T2 = bundled launch (n=9 in this set); else clean.
| tier | n | real (>=10 buyers, wash<.5) | reach 1M | reach 5M |
|---|---|---|---|---|
| proxy T1 | 5280 | 27% | 39% | 14% |
| bundled | 9 | 89% | 22% | 11% |
| clean | 21983 | 18% | 7% | 3% |
On the 5,238 pools in bundle_fund `seen` (population, not runner-sampled): T1 47% reach 1M vs clean 4%.
The proxy "rug" tier does BETTER, which means it is a look-ahead / selection artefact: pool count per token and cluster size are only known after the fact, and tokens that run get re-pooled. Not usable at entry. A causal version needs pool count and creator history frozen at entry (+5 min); the creator is NULL on v3 and unknown on most v4 pools, so it needs ppdata's creator/txfrom table (#226).

## Next
Nothing to send to testmgr. When dataeng's social table lands: score at first_seen time of each field, then run skip-T1 / skip-T1+2 on the mc5 replay. Reopen then.