# Launch runners, round 2: what the coins that reach $500k to millions look like (5 Oct 2026, Bangkok)

## Read this first (for the launch session)
- This was written by a separate session that only reads your data. Nothing live was changed. Everything here is a proposal with receipts.
- Ben's question: which launches go to $500k, $1M or multi-millions right after launch, and what can we see at 30 s?
- **The big finding: one launch type makes 72% of the runners.** Native-ETH pool, 0.01% fee, 10 ETH starting cap (feat.db `pad` = `0x0000000000000000000000000000000000000000|100`).
  - Only 1,476 of 15,149 launches were this type, but 7.25% of them became runners, versus 0.3% for everything else.
  - Every day of the sample was between 3% and 13%.
- **The 10x label mixes two very different things.**
  - Runners: 10x+ and $500k+.
  - Pops: 10x from a tiny 2.5 ETH start that tops out under $500k.
  - Starting cap alone tells them apart (AUC 0.86).
  - Training on the runner label instead of the 10x label made the top 1% better for Ben's goal (details below).
- **Inside that one launch type, the 30 s trade data adds little** (best single AUC about 0.6). The extra edge inside it has to come from other information.

## Data (read only)
- `/opt/hist/analysis/mcap.db` gives USD outcomes: entry cap, peak cap, sellable cap.
- Joined with `feat.db` (your 30 s features) and `catbase.db`.
- 15,149 clean launches, 21 Sep to 3 Oct.
- **Runner** = sellable cap of $500k or more AND at least 10x from the 30 to 40 s entry.

| Group | Count | Note |
|---|---|---|
| Sellable $500k+ | 520 | only 148 got there as a 10x from our entry; the rest STARTED high (median entry $126k) |
| Sellable $1M+ | 137 | only 50 as a 10x; the rest started near $1M (median entry $1.0M) |
| Sellable $3M+ | 35 | only 9 as a 10x |
| Runners | 148 (1.0%) | median entry $53k, median sellable peak $846k, median 0.97 h from entry to that peak |
| Pops | 111 | 10x but stayed under $500k |

- Trap: "reached $1M" mostly means "launched big", the same trap as the old $100k label. Always require the 10x from our entry too.

## Findings
1. **Starting cap decides runner or pop** (runner rate, then pop rate):

| Starting cap | Launches | Runner | Pop |
|---|---|---|---|
| up to 1 ETH | 2,998 | 0.30% | 1.63% |
| 1 to 3 ETH | 4,008 | 0.27% | 0.92% |
| 3 to 9.5 ETH | 418 | 2.1% | 2.4% |
| **9.5 to 10.5 ETH** | **1,476** | **7.25%** | 0.88% |
| 10.5 to 15 ETH | 3,438 | 0.03% | 0.03% |
| 15 to 25 ETH | 2,653 | 0.41% | 0.04% |

2. **Runners vs pops at 30 s** (AUC, runner vs pop):
   - starting cap 0.86
   - more distinct buyers 0.78 (median 14 vs 4)
   - more dust buys 0.79
   - more ETH in 0.75
   - no single wallet holding the ETH 0.31 (top buyer share 0.34 vs 0.87)
   - minter sends 0.28
   - wallet-to-wallet transfers 0.38

   Pops look like one wallet pumping a tiny pool. Runners look like a real crowd arriving in the 10 ETH type.
3. **Training on the runner label works better for $500k+ goals.** Walk forward, 9 test days, one-day gap, top 1% of each day (106 picks each):

| Model trained on | Runner hit | 10x hit | $500k+ hit | $1M+ hit | Median sellable peak |
|---|---|---|---|---|---|
| runner label | **13.6%** | 15.4% | **17.4%** | 5.4% | **$141k** |
| 10x label | 8.9% | 17.0% | 12.0% | 4.9% | $69k |

   - Base rates: runner 0.98%, 10x 1.7%.
   - Better on 5 days, equal on 2, worse on 2. Small sample.
4. **Inside the 10 ETH type, 30 s tape features barely help.** Best AUC about 0.6. A small model's top 10% gave 9.4% runners versus 8.1% for the whole type.

## Proposals (ordered)
1. **Add a runner book.** Same features, label = sellable $500k+ AND 10x from entry, kept apart from the 10x books.
2. **Treat the 10 ETH, 0.01% fee native-ETH type as its own pond.** Score every coin in it.
   - Test the honest result of a fixed small ticket on EVERY coin of this type: chain-priced fills and exits, 24 h.
   - The mcap.db "last" price is not an exit price (see the warning below).
3. **Look for edge inside the pond beyond the 30 s tape:**
   - contract family and name kind (your id5 features)
   - who deployed it before
   - X or text at launch
   - the flow from minute 1 to 10

   Runners peak about 1 h after entry, so a 2 to 5 min confirmation may cost less for runners than for quick 10x pops. Test it inside the pond.
4. **Exit for runners.** The median runner went about 16x from our entry ($53k to $846k). Selling everything at 10x caps the very coins Ben wants. Test, inside the pond only: half out at 10x, trail the rest.
5. **Explain the pond.** Find which factory or launchpad makes the 10 ETH, 0.01% type. Track whether its runner rate holds week by week, or fades as more people farm it. Alert if it drops under 3% in a week.

## Data warnings
- mcap.db holds a median of only about 7 min of data per coin. The "last" cap is the end of that window, not a real exit price. Any expected value built on it is wrong; use the chain-priced 24 h tools.
- A few rows have absurd caps (1e46): upside-down prices, the same family problem as the 777 coins. Filter entry caps outside $1k to $5M before any study.
- catbase `has_text` and `l1` labels are biased (only coins that later traded got labels). Do not use them for the past.

Scripts (coin brain session, /root/brain): caps study run inline. Export: http://159.223.88.168/hx-c9ois/brx/caps.jsonl.gz

Version 1, 5 Oct 2026. Latest: http://159.223.88.168/hx-c9ois/brx/launch-runners-round2.md