claudia rebuilt: the best of every strategy, in one agent
we don't hide from our data. we publish it.
13,457 real-money trades across 8 agent strategies over 18 days. every entry, every exit, every win, every loss — logged, measured, ranked. the numbers don't lie, and what they told us about Claudia was uncomfortable.
she was our worst performer. -$1,826 total PnL. 15.8% win rate. 84 out of every 100 entries were losers.
so we killed the old Claudia and rebuilt her from scratch using the DNA of every strategy that actually made money.
---
what the data said
before touching a single line of code, we ranked every agent by real-money performance:
| agent | trades | win rate | total PnL | profit factor |
|---|---|---|---|---|
| sienna oracle | 2,497 | 21.1% | +$2,434 | 1.23 |
| ricky (scalper) | 132 | 67.4% | +$81 | 4.17 |
| baseline | 397 | 46.1% | +$29 | 1.22 |
| james (selective) | 67 | 56.7% | +$12 | 1.46 |
| claudia (old) | 3,360 | 15.8% | -$1,826 | 0.89 |
three things were working. Claudia wasn't one of them.
---
why old claudia failed
the diagnosis was clear once we looked at the data:
too loose on entries. 0.08 composite score threshold let garbage through. for context, Sienna uses 0.14 and she's the profit leader. Claudia was entering tokens that Sienna would reject without a second look.
too narrow on market cap. $5K–$100K is peak rug territory on Solana. 98% of tokens in this range die within hours. Claudia was fishing in a poisoned pond.
pump.fun only. she ignored every DEX-listed token. Sienna's biggest wins came from tokens that had already graduated to Raydium — tokens Claudia couldn't even see.
no stop loss grace period. normal volatility in the first few minutes of a trade would trigger her stop loss. she'd enter a token, watch it dip 12% (completely normal for memecoins), get stopped out, and then watch it run 300% without her.
only 2 take profit levels. she waited until +100% to take her first profit. most winning trades never reach +100%. Ricky's data showed that banking at +80% captures far more winners.
no AI gating. every token that passed the basic composite score got traded. no second opinion. no deeper analysis.
---
the new claudia: "the alpha"
we took the best-performing element from each profitable strategy and combined them into one agent.
from sienna (profit leader, +$2,434):
from ricky (highest efficiency, 4.17 profit factor):
from james (best risk-adjusted, 1.46 profit factor):
---
how she trades now
entry
Claudia scans both pump.fun bonding curve tokens and DEX-listed tokens across a $10K–$50M market cap range. every candidate must pass:
exit ladder
once in a position, Claudia manages it with a 3-tier take profit ladder:
TP1: +80% — bank 35%. this is the sweet spot from Ricky's data. most winners peak between +60% and +120%. banking a third here locks in profit on the majority of winning trades.
TP2: +250% (3.5x) — trim 15%. the position has proven itself. take a small trim, let the rest ride.
TP3: +400% (5x) — trim 15%. confirmed runner. another small trim. the remaining 25% of the original position plus the 10% moon bag continue riding.
moon bag: 10% rides indefinitely. if the token goes to 50x, 100x, or beyond — Claudia is still in.
adaptive trailing stop
the trailing stop widens as gains increase:
| gain level | trail width |
|---|---|
| +20% (activation) | 15% |
| +80% (post-TP1) | 18% |
| +250% (post-TP2) | 22% |
| +400% (post-TP3) | 25% |
this means a token that's up 300% won't get shaken out by a normal 15% pullback. the trail gives confirmed runners room to breathe.
time exits
dead weight gets cut fast:
stop loss
-12% hard stop with a 5–10 minute grace period (randomized per position) and 10-second confirmation. the grace period means normal early-trade volatility doesn't trigger a premature exit. the confirmation window means a brief wick doesn't close the position — the price has to stay below the stop for 10 consecutive seconds.
---
the numbers, side by side
| parameter | old claudia | new claudia |
|---|---|---|
| market cap range | $5K–$100K | $10K–$50M |
| composite score | 0.08 | 0.14 |
| buy ratio | 0.60 | 0.58 |
| data sources | pump.fun only | pump.fun + DEX |
| TP1 | +100%, sell 25% | +80%, sell 35% |
| TP2 | +420%, sell 25% | +250%, sell 15% |
| TP3 | none | +400%, sell 15% |
| moon bag | 10% | 10% |
| stop loss | -15%, no grace | -12%, 5–10min grace |
| SL confirmation | 8s | 10s |
| trailing activation | +25% | +20% |
| time exits | 20min / 45min | 10min / 20min / 40min |
| MC-adaptive exits | no | yes |
| AI gating | no | yes |
| max concurrent | 10 | 25 |
| strategy weighting | EARLY_BIRD dominant | VOLUME_CLIMAX dominant |
the single biggest change: strategy weighting. old Claudia heavily favoured EARLY_BIRD entries — which our data showed is the biggest dollar loser across the entire platform (-$165K total). new Claudia leads with VOLUME_CLIMAX (the only consistently profitable strategy at +$228K) and MOMENTUM, with EARLY_BIRD demoted to neutral weight.
---
what this means for you
if you're running Claudia, the update is automatic. next time she scans, she uses the new parameters.
you'll notice:
---
the philosophy
we don't believe in one perfect strategy. we believe in data.
13,000+ trades told us exactly what works: Sienna's entry quality, Ricky's exit discipline, James's selectivity. old Claudia had none of these. new Claudia has all of them.
she's not a diamond hands runner anymore. she's not a scalper. she's not an oracle.
she's the best of all three.
less noise. more alpha.
---
Claudia is available now on parasol.so. create an agent, select Claudia, and let the data do the work.