What is the Signal Scorecard?

It is a live, out-of-sample backtest of every trading signal MadeOnSol serves. For each signal we publish hit rate, base rate, lift over base, and sample size so you can judge the real edge instead of trusting a black box.

How are signals backtested?

Out-of-sample with no lookahead. Deploy-anchored signals use a rolling train/test split, a minimum 15 early-buyer filter, and peak-timing outcome labels (peaking under 15 minutes after deploy is a dump; 60 minutes or later is a runner). Mid-life signals like KOL coordination use a forward return instead — peak market cap reaching 2x the entry within a fixed horizon. Numbers refresh daily.

Are these signals financial advice?

No. Signal lift and hit rates are backtested statistical indicators of historical patterns, not predictions of future performance. Always do your own research.

Can I access signal efficacy via API?

Yes. The /api/v1/signals/{name}/performance endpoint returns the same per-bucket hit rate, base rate, and lift so bots can weight signals programmatically. Open to any authenticated tier.

Signal Scorecard

Every signal we serve, backtested live and out-of-sample. We show exactly how well each one predicts outcomes — hit rate, lift over base, and sample size. No black boxes.

Dump-cluster detection

classification

How often a token actually dumps (peaks <15 min after deploy) when its top-20 early buyers include known dump-cluster wallets. Higher count → stronger signal.

cluster buyers in top-20

P(dump)

lift

sample

base rate

83%

—

k>=1

92%

1.11×

n=1343

k>=3

97%

1.17×

n=709

k>=5

99%

1.19×

n=366

k≥3 is a strong dump signal — treat it as an avoid / early-exit filter.

14d out-of-sample·2,418 tokens·updated 21h ago· no lookahead· 40d drift

Field dump_cluster_count · GET /api/v1/signals/dump_cluster_count/performance

Recycled early-buyers (clean tokens only)

classification

Among tokens with NO dump-cluster wallets, does a high count of “recycled” early buyers (wallets that keep reappearing in first-20s) lean toward a runner? A mild, contextual signal — near-zero edge on its own, only meaningful once dump-cluster is ruled out.

recycled buyers in top-20

P(runner)

lift

sample

base rate

32%

—

r>=3

33%

1.04×

n=726

r>=5

36%

1.14×

n=640

r>=8

36%

1.12×

n=524

Use as a tiebreaker on already-clean tokens — never as a standalone filter.

14d out-of-sample·1,890 tokens·updated 21h ago· no lookahead· 40d drift

Field recycled_early_buyer_count · GET /api/v1/signals/recycled_early_buyer_count/performance

Deployer runner-rate

calibration

Calibration: when a deployer's as-of runner-rate sits in a bucket, how often do their next tokens actually run (peak ≥60 min)? A trustworthy score should track the diagonal.

deployer as-of runner-rate

P(runner)

lift

sample

base rate

20%

—

[0,0.2)

13%

0.64×

n=412

[0.2,0.4)

23%

1.14×

n=160

[0.4,0.6)

30%

1.51×

n=94

[0.6,0.8)

53%

2.68×

n=17

[0.8,1]

37%

1.87×

n=57

Weight deployers with a higher as-of runner-rate; gate on ≥3 labeled tokens for confidence.

14d out-of-sample·740 tokens·updated 21h ago· no lookahead· 40d drift

Field deployer_runner_rate · GET /api/v1/signals/runner_rate/performance

KOL coordination by count

forward_return

Forward return: when N distinct KOLs buy the same token within a rolling 15-min window, how often does it 2x (peak market cap) within the next 6h, entering at the Nth KOL? Base is a lone-KOL buy. More KOLs → stronger edge.

KOLs in 15-min window

P(2x)

lift

sample

base rate

24%

—

k>=2

26%

1.10×

n=10493

k>=3

27%

1.11×

n=6454

k>=4

27%

1.12×

n=4358

k>=5

28%

1.14×

n=2941

k>=6

28%

1.15×

n=2094

2-3 KOLs barely beat a lone KOL — the real edge starts at k≥5. That's why coordination alerts default to min_kols=5.

14d out-of-sample·26,340 tokens·updated 21h ago· no lookahead· 40d drift

Field kol_count · GET /api/v1/signals/coordination_count/performance

scout_first_touch

calibration

bucket

P(swarm_3plus)

lift

sample

base rate

21%

—

11%

0.55×

n=8059

21%

1.05×

n=6288

30%

1.47×

n=6501

66%

3.21×

n=234

14d out-of-sample·21,082 tokens·updated 21h ago· no lookahead

Field scout_first_touch · GET /api/v1/signals/scout_first_touch/performance

How we measure it (honestly)

Rolling train/test split: signals are derived from older tokens and graded on the most recent settled window — never on data they were built from (no lookahead). We only label tokens with ≥15 early buyers (sparse, barely-traded tokens are unlabelable). Labels are peak-timing based: a token “dumped” if it peaked <15 min after deploy, “ran” if ≥60 min. Mid-life signals like KOL coordination instead use a forward return — peak market cap reaching 2× the entry within a fixed horizon — since there is no deploy to anchor to. Numbers update daily and will drift with the market — that's the point.

Reading the table

base rate: the outcome's rate across all tokens, ignoring the signal — the bar each bucket has to beat.
P(dump) / P(runner): probability of the labelled outcome: dump = peaked <15 min after deploy, runner = peaked ≥60 min.
lift: hit rate ÷ base rate. 1.0× = no edge over chance; 1.5× = 50% more likely than the average token.
bucket: a slice of the signal's value — e.g. ≥3 cluster buyers, or a deployer rate of 0.6–0.8.
sample (n): tokens in that bucket over the test window. Small n = treat the number as noisy.
classification vs calibration: classification asks does the signal's presence predict the outcome; calibration asks does a score's value match reality (does 0.7 really mean ~70%).
out-of-sample: every number is graded on a recent window the signal was NOT built from — forward performance, not a curve fit.

Programmatic access: GET /api/v1/signals/{name}/performance

Add ?history=true for the per-day drift series.

Disclaimer

MadeOnSol provides on-chain data and statistics for informational purposes only. This is not financial advice and not a recommendation to buy or sell any asset. Past performance does not guarantee future results — always do your own research (DYOR).