the number that decides whether Parasol trades
every token that enters Parasol's pipeline gets a single number: the composite score. it ranges from 0 to 1. if the score is above the agent's threshold, the token is a candidate for entry. if it's below, it's ignored. no exceptions, no overrides.
this post explains what goes into that number, how the threshold works, and why it matters.
what the composite score actually is
the composite score is a weighted blend of three signals:
the formula:
``
compositeScore = confidence × wC + min(1, rrScore/10) × wRR + strategyScore × wS
`
the weights (wC, wRR, wS) aren't fixed. they shift based on the current market regime.
regime-adaptive weights
Parasol reads the macro environment and adjusts how much it trusts each signal:
| regime | confidence weight | R:R weight | strategy weight |
|---|---|---|---|
| bullish | 0.30 | 0.25 | 0.45 |
| neutral | 0.40 | 0.30 | 0.30 |
| bearish | 0.50 | 0.35 | 0.15 |
in a bull market, momentum matters more. the engine leans into strategy signals and rides trends. in a bear market, it demands higher quality tokens and better risk/reward before entering. this happens automatically — no manual switching.
what happens after the base score
the base formula produces a number, usually between 0.1 and 0.6. then a series of adjustments push it up or down based on real market signals.
multi-timeframe alignment
if the 5-minute, 1-hour, and 6-hour timeframes all agree on direction, the score gets a +0.20 bonus. if only 5m and 1h agree, +0.10. if 5m is bullish but 1h is bearish (divergence), it gets a -0.10 penalty.
this catches the difference between a token trending on all timeframes vs one that just spiked in the last 5 minutes while the longer trend is down.
token freshness
learner data shows a -94.3 correlation between token age and winning trades. newer tokens with rising volume are far more likely to continue pumping. older tokens have already peaked.
fresh tokens get rewarded. stale tokens get buried.
runner prediction (for diamond-hands agents)
agents configured for runner-catching (like Claudia) get additional bonuses for tokens matching the profile of 4x-16x runners from historical data:
these bonuses stack, but combined with all other bonuses they're capped at +0.25 total to prevent weak base signals from inflating through stacked bonuses alone.
other adjustments
the range in practice
the composite score ranges from 0 to 1, but in practice:
minCompositeScore: the threshold
each agent has a minCompositeScore setting. any token scoring below this is rejected before it can become a trade. the threshold varies by agent personality:
| agent style | minCompositeScore | what it means |
|---|---|---|
| careful (default profile) | 0.28 | only high-conviction entries. few trades, higher quality. |
| aggressive | 0.10 | lower bar. more trades, more variance. |
| degen | 0.08 | nearly everything that's not noise. high volume, high variance. |
| pump.fun runners (Claudia) | 0.03 | almost no filter. designed for super-early tokens where data is sparse. |
| DexScreener selective (James) | 0.15 | demands strong signal from established tokens. |
| session traders | 0.12 | balanced. needs real signal but not as strict as careful. |
raising minCompositeScore means fewer trades but higher average quality. lowering it means more trades but more losers mixed in. there's no universally correct value — it depends on the strategy.
for pump.fun agents scanning bonding curve tokens, data is sparse. a brand new token has no volume history, no multi-timeframe data, no holder distribution to analyze. the composite score for these tokens is inherently low, so the threshold has to be low too. the scoring system compensates with freshness bonuses and runner-prediction filters instead.
for DexScreener agents scanning established tokens, there's rich data available. volume, price changes across timeframes, holder data, smart wallet activity. the composite score can be more meaningful, so a higher threshold makes sense.
how it affects position sizing
the composite score doesn't just gate entry — it also determines how much capital the agent allocates:
in bearish regimes, the maximum conviction multiplier drops from 1.5x to 1.2x. the engine automatically becomes more conservative with sizing when the macro environment is hostile.
the final gate
even after all adjustments, two hard gates remain:
these gates can't be overridden by bonuses. they exist to prevent the engine from entering positions where the math doesn't work.
summary
the composite score is the single number that determines whether Parasol trades a token. it blends confidence, risk/reward, and strategy signal with regime-adaptive weights. it gets adjusted by timeframe alignment, freshness, runner prediction, smart wallet activity, and social signals. it's capped at 1.0, floored at 0 (or 0.03 for pump.fun), and the agent's minCompositeScore` threshold is the gate that decides which tokens pass and which don't.
lower threshold = more trades = more noise.
higher threshold = fewer trades = higher average quality.
the right setting depends on what you're trying to catch.