Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/alphaleaks60-maker/docs2/llms.txt

Use this file to discover all available pages before exploring further.

Wallet intelligence is the foundation of Alpha Leak’s edge. A raw buy event on Pump.fun is noise. That same event, attached to a wallet with a 40% graduation rate, a 0.8 capture efficiency, and a consistent history of top-10 entries, is a meaningful signal. Phase 1 exists to build and maintain that context for every wallet the system encounters — continuously, at scale, without lookahead.

Alpha scoring

7-component composite score (0–100) summarising a wallet’s historical edge across graduation rate, win rate, timing, and consistency.

Adversarial guards

Bot-classified wallets are hard-capped at alpha 30, preventing inflated scores from high-volume mechanical activity.

Decay profiling

Per-wallet alpha decay curves across 8 delay buckets reveal the optimal follow delay and signal half-life for each wallet.

​
Wallet discovery

Wallets enter the tracking system through one of three paths, each carrying a different level of initial confidence.
1

Graduation-based discovery

When a token graduates (crosses 85 SOL on the bonding curve), WalletDiscovery processes all buyers of that token. Wallets that bought early and held through graduation are promoted to is_active = true status. This is the highest-quality discovery path — the wallet demonstrated precisely the behaviour that predicts future success.
2

Signal-based discovery

Any wallet that executes a trade alongside a known tracked wallet triggers an implicit discovery event. The new wallet is written to the wallets table and queued for enrichment. This path surfaces co-buyers who may themselves be high-quality wallets.
3

Manual seeding

Wallets can be added directly to the tracked_wallets Redis set for immediate prioritisation. This is useful for pre-seeding the system with known high-quality wallets before live data accumulates.

​
Wallet enrichment

The WalletEnrichment service continuously processes newly discovered wallets, fetching their historical Birdeye data (where available) to bootstrap feature profiles without waiting for live trades. A wallet that was active on Pump.fun before the pipeline started can arrive in the system with a complete historical record — graduation counts, win rates, position history — from its first appearance.
Enrichment runs continuously in the background. A newly discovered wallet may have its full historical profile available within minutes of discovery, even if it has not traded since the pipeline started.

​
The alpha score

The alpha score (0–100) is a single number summarising a wallet’s historical edge. It is designed to reward the behaviours that actually predict future token success, weighted by how reliable each signal has proven to be.
Score formula
alpha_score =
  graduation_score   × 0.20   +  // Share of bought tokens that graduated
  win_score          × 0.20   +  // Share of positions closed in profit
  capture_score      × 0.15   +  // Ratio of realised to peak return
  consistency_score  × 0.15   +  // Inverse of return standard deviation
  rank_score         × 0.10   +  // How early the wallet typically enters
  recency_score      × 0.10   +  // Exponential decay for inactive wallets
  discovery_score    × 0.10      // Average peak multiple of tokens bought
The fraction of tokens bought by the wallet that subsequently graduated (crossed the 85 SOL bonding curve threshold). Graduation is the strongest single predictor of token quality — it requires sustained demand over minutes to hours, not just initial bot activity. A wallet with a 40% graduation rate across 100 tokens is a highly reliable signal source.
The fraction of positions the wallet closed in profit. This is distinct from graduation rate: a wallet can buy a token that graduates but still sell at a loss if it bought near the top. Win score measures realised trading performance specifically.
The ratio of the wallet’s average realised return to its average peak return. A capture score of 0.8 means the wallet typically exits at 80% of the maximum price it could have achieved. High capture scores indicate disciplined exit behaviour — the wallet does not hold through peaks and give back gains.
The inverse of the standard deviation of the wallet’s returns across positions. A wallet that consistently returns 2–3× is scored higher than one that occasionally hits 10× but loses on most trades. Consistency is critical for signal reliability: a signal from an inconsistent wallet is harder to act on with confidence.
A measure of how early the wallet typically enters relative to other buyers. Rank is computed as the wallet’s buy order position among all buyers of a token, normalised to 0–1. A rank score of 0.9 means the wallet consistently buys in the first 10% of buyers. Early entry is a prerequisite for large returns on Pump.fun.
An exponential decay applied to wallets that have been inactive. A wallet that last traded 30 days ago has its score penalised regardless of its historical quality — market conditions change, and stale wallets may no longer be active participants. The decay ensures that only currently active wallets receive full score credit.
The average peak price multiple of tokens the wallet has bought, measured at their 24-hour peak. This rewards wallets that consistently find tokens with high upside, independent of whether the wallet itself captured those gains. A wallet that buys tokens that 10× is more useful as a discovery signal than one that buys tokens that peak at 1.5×.

​
Confidence scaling

The alpha score alone is not sufficient — a score of 80 from a wallet with 10 trades is very different from a score of 80 from one with 200 trades. Confidence scales from 0 to 1 based on trade count, reaching full confidence at 50 tokens. The ML model receives both the raw alpha score and the confidence value as independent features.

​
Bot penalty

Wallets classified as bots by the BotDetector are hard-capped at a maximum alpha score of 30, regardless of their apparent metrics. Bots can accumulate impressive-looking statistics through sheer volume, but their behaviour is mechanical and not meaningful for signal purposes. The is_bot field is also a direct feature in the ML model.
A bot wallet with a high apparent graduation rate does not indicate trading skill — it indicates that the bot buys large numbers of tokens and some fraction happen to graduate. The 30-point cap and the is_bot ML feature together prevent this pattern from polluting signal quality.

​
Feature computation cadence

The FeatureComputer runs on a 30-minute rolling cycle, processing up to 5,000 active wallets per pass. For each wallet it computes the full feature set from the trades, signals, and pnl_records tables, then writes the results back to the wallet_features table.
CadenceScopeOutput
Every 30 minutesUp to 5,000 active walletsFull feature vector per wallet
On signal firePoint-in-time snapshotwallet_score_at_entry captured for ML training
HourlyAll wallets with signal historyAlpha decay curve (8 delay buckets)

​
Point-in-time correctness

A critical property of the alpha score is that it is computed at the time of scoring using only data available up to that point. When the ML model records wallet_score_at_entry, it captures the score exactly as it was when the signal fired — not the wallet’s current score. This prevents lookahead bias in model training and ensures that historical backtests reflect what was actually knowable at decision time. A wallet’s score as computed on day 30 does not contaminate the signal record from day 1.
Point-in-time correctness is enforced by design: the signal writing pipeline captures wallet stats at the moment of emission, not at query time. This makes the training dataset causally valid without any additional filtering.

​
Alpha decay tracking

Not all smart money signals age equally. Some wallets move markets — when they buy, price moves within seconds. Others have a longer information half-life where their signal remains valid for minutes or more. The AlphaDecayTracker runs hourly and computes a per-wallet decay curve: the average return achievable if you had bought at each of 8 delay intervals after the wallet’s signal.
Delay bucketDescription
1 secondEffectively simultaneous — requires bot execution
5 secondsVery tight — requires sub-second latency stack
10 secondsTight — achievable with optimised execution
30 secondsModerate — achievable with alert-based execution
60 secondsComfortable — manual execution viable
120 seconds2-minute delay — signal likely fading
300 seconds5-minute delay — significant decay expected
600 seconds10-minute delay — mostly informational
From the decay curve, two values are derived:
  • Alpha half-life — the delay at which the expected return drops to 50% of the instant return.
  • Optimal follow delay — the delay that maximises expected return. Buying instantly is not always optimal; in some cases a short delay filters out false starts.
Wallets with sub-5-second half-lives are effectively unmatchable manually. Wallets with half-lives above 60 seconds are highly actionable for alert-based or manual execution strategies.

Adversarial detection

How bot detection, bundle detection, and copy-trade classification protect the alpha score from manipulation.

ML features

How wallet alpha scores, decay curves, and feature vectors feed into the 68-feature ML model input.

Phase 1: scoring

The full Phase 1 pipeline — PeakTracker, PnlCalculator, FeatureComputer, and WalletScorer in context.

Live trader strategies

How wallet alpha scores influence position sizing and strategy selection in the live trader.

Build docs developers (and LLMs) love

Get started for freeTalk to us