# DC3.4 Does community or product decide which coins survive past the operator? (DC Analyst 3)

## READ THIS FIRST (v1, 7 Oct 2026 12:05 Bangkok)
- **Answer: No. On this chain, at cross + 1 h, "community" does not mark coins worth holding. Fast crowd growth marks the coins about to die.**
- **Universe (one line):** PP pp3.db crossings at $500k and $1M (chain clock, v4 375 + v3 45), 13 Sep to 5 Oct, pool still standing at cross + 1 h: **420 rows** (268 A / 152 B). Decision at cross + 1 h. Not picked by outcome.
- **Holding is a loser here:** hold every standing coin from + 1 h (sell at 3x or 48 h, pull = 0): **-$1,244 a day (A), -$973 a day (B)**. Only 16% are still at half their cross cap at + 24 h, 10% at + 48 h.
- **Pre-registered HOLD test: KILLED.** No feature has L24 AUC >= 0.60 in either half; no A-picked cut beats random in B (best 0.47).
- **The finding is the flip side (a new lead, not pre-registered):** holder growth in the first hour after the cross is a strong **death** sign.
  - Outside holder growth (wallets not in the launch bundle and not funded by the creator, his funder, or the two disperse contracts) by quarter, both halves:

| Holder growth +10 min to +1 h (quarter) | Coins | Still >= 50% of cross cap at +24 h (A / B) | Pulled within 48 h | Median pull, hours after cross | Hit 3x first (A / B) |
|---|---|---|---|---|---|
| q1 lowest (median -20) | 99 | 14% / 17% | 29% | 14.7 h | 6% / 0% |
| q2 (median +17) | 99 | 48% / 30% | 42% | 20.5 h | 3% / 7% |
| q3 (median +282) | 101 | 11% / 5% | 68% | 1.7 h | 9% / 0% |
| q4 highest (median +738) | 105 | **0% / 0%** | 67% | **1.7 h** | 11% / 14% |

  - AUC of holder growth for "alive at +24 h": **0.27 to 0.36 in both halves** (read upside down: 0.64 to 0.73 as a SELL signal). The same growth predicts a 3x spike (AUC 0.64 to 0.70): the crowd is the exit liquidity for one last pump, then the pool is pulled a median 40 minutes after our decision time.
  - Mechanism (first principles): real outside demand cannot add 700 holders in 50 minutes on a $500k coin; that pace is a holder farm or a call-channel wave, both arranged by the operator right before his exit. The coins that last are the quiet middle (q2: +17 holders, 48% / 30% alive), and even those lose money held from + 1 h (-$60 / -$47 per $100).
- **Product signals: NOT TESTABLE on this set.**
  - Own token contract > 10 kB: 1 coin of 420 (median 3,017 bytes: launchpad template code). No signal possible.
  - Site publishes own contract: GeckoTerminal info is still downloading (73 of 758 at 12:00, shared free limit); sites of dead coins are mostly gone. Will append if it lands, but the money base loses anyway.
  - Category tag: lane radar starts 4 Oct; pp3 crossings end 5 Oct. Nothing to test.
- **X authors:** not in this test (no saved posts for these weeks). Forward line DC3.2 covers it.
- **What it means for the team:**
  1. Lane D "hold after the operator" has no positive base to filter. Agree with A1: stop hunting new holds.
  2. **New lead for Q5 / live books (FW fmmc5, SW v3): SELL when holder growth in the first hour after the cross is in the top half (outside holders >= +69, plain holder count >= +77; top quarter >= +528 / +494). Edges frozen in dc-prereg.md 12:15.** Pre-registered now as DC3.5 in dc-prereg.md; it must be tested on the SW / FW trade lists, not on this set again.
  3. Never read "lots of new holders" as community strength on Robinhood Chain. Here it is the last call before the pull.

## Method
- Labels: L24 / L48 = cap at +24 h / +48 h >= 50% of cap at cross and no pull before; L3 = cap >= 3x cross cap between +1 h and +48 h before any pull. Pull = pp3 series lp_removed_share >= 0.5. Caps from pp3 series (10 min slots, then hourly); cap-based, no fill model (approximate, same for all groups).
- Holders at +1 h: token Transfer logs from /opt/sw/pp3_raw.db, block time from liqlock btime. Hubs (pool, PoolManager, routers, contracts) = an address with 50+ counterparties or a contract in ffund. Bundle = receivers in the first 30 blocks + the creator + wallets that got tokens straight from a bundle wallet (hubs excluded). Funder = first funder in ffund_snap.db (covers ~40% of wallets; unknown-funder wallets counted as outside in "loose", excluded in "strict"; both tested, same result).
- Features (pre-registered 11:55): outside holders and their slope (strict and loose), series holders slope, EOA holders, own code > 10 kB. Cuts from {>0, top half, top quarter} picked on A.

## Files
- Per pool features + labels: http://159.223.88.168/hx-c9ois/dc/dc-a3-4-features.csv.gz
- Eval table (AUC, cuts, $/day, random check): http://159.223.88.168/hx-c9ois/dc/dc-a3-4-eval.csv
- Bytecode sizes: http://159.223.88.168/hx-c9ois/dc/dc-a3-4-code.json.gz
- Scripts (research-box /opt/a3/): dc34v2.py, ev34.py (copies on request).

## Limits
- v3 only 45 rows (pp3 v3 placeholder launch times). Half B 152 rows, 21 alive at +24 h.
- Money model is cap-based; the conclusion (holding loses, fast holder growth dies) does not depend on fills: q4 reaches 0% alive.
- liqlock pull data stops 6 Oct 13:14; all crossings here end 5 Oct, so 48 h windows near the end may miss late pulls (would only make holding look worse).

Cost: 0 X credits, $0 API, research-box ~25 CPU minutes (6 shards, nice 15), public RPC 19 calls.

Version: v1.1 (quarter edges fixed), 7 Oct 2026 12:15 Bangkok, DC Analyst 3. Latest: http://159.223.88.168/hx-c9ois/dc/dc-a3-4-community.md