product updates: march 8, 2026
positions no longer vanish on refresh, sells that fail now retry themselves, the scanner is 8x faster, and a brand new strategy was derived directly from 2,000 real trade outcomes. here's everything that shipped.
---
position recovery via database
the most significant reliability change we've shipped.
previously, open positions were tracked in the browser's localStorage. if you refreshed the page, the app had to reconstruct your positions from local state — and if that state had drifted (a common occurrence after a session with a lot of activity), positions could disappear or reappear as ghost entries.
positions are now backed by the database. on every page load, the agent fetches open positions directly from supabase and reconciles them with what's in memory. what you see on screen now reflects ground truth.
as part of this change:
---
auto-retry on failed sells
when a sell fails — network error, slippage too tight, wallet momentarily unavailable — the agent now queues a retry automatically. the retry fires after a 2-minute cooldown.
three specific failure modes now have dedicated handling:
these three cases were responsible for the majority of stuck positions. they're now handled deterministically.
---
scan speed: 8x faster
the token scanner was running up to 100 sequential requests to DexScreener to enrich candidate tokens with liquidity, volume, and price data. at ~300ms per request, this was the main source of scan timeouts — the 30-second API gateway limit was hit before results came back.
the rewrite runs 12 parallel batches instead. real-world scan time dropped from 25–35 seconds to under 5 seconds on the same token set. timeouts are now rare. a latent bug in the fallback path was also fixed — it was returning 0 tokens when the primary path hit the timeout wall.
rug check results are now cached per token address. tokens that reappear in consecutive scans — common for tokens that linger near the filter threshold — no longer trigger a fresh rug check every time. this cut Helius credit usage significantly.
---
PRESSURE_ZONE strategy
a new entry strategy derived from the analysis of 2,000 paper trade outcomes across all agents.
the finding: tokens that experience a sharp spike followed by a pullback to a support band — specifically to the 40–60% retracement of the initial move — have a statistically higher probability of launching a second leg. the market has found value. weak hands have been shaken. the setup is cleaner than catching the initial pump.
PRESSURE_ZONE specifically looks for:
this strategy is now assigned to Yuki (whale watch mode). yuki was already the best-performing agent for catching second-leg moves; PRESSURE_ZONE formalises why that's true and makes the entry logic explicit rather than emergent.
---
agent exit level tuning
sienna — take-profit levels widened. sienna's previous TP structure was optimised for average moves. for tokens that run hard, the trailing stop was firing too early — capturing a strong return but missing the continuation. TP2 and TP3 have been pushed out to give genuine runners more room.
claudia — rebuilt with two wide take-profit levels: 100% and 420%. trailing stop activates at 25% gain and trails at 15%. claudia now explicitly hunts the outlier trade, accepting more volatility between entry and the first partial exit in exchange for keeping a meaningful position alive through the move.
both changes were driven by the same trade outcome data that produced PRESSURE_ZONE.
---
token chart overhaul
the chart header was showing a price where market cap is more useful, and was missing context that traders check immediately when they open a new chart.
changes shipped:
---
leaderboard accuracy fix
the leaderboard was ranking users by their period-filtered points (correct) but displaying their all-time total in the Points column (incorrect). this produced situations where a user ranked #29 could display more points than the user ranked #2, because the all-time totals weren't what the sort was based on.
the Points column and "Your Rank" card now both display the same metricValue that the ranking is sorted by. the number you see is the number that determines your position on the board, for any time period selected.
trade counts in the leaderboard also now filter by the selected period — previously they always showed all-time totals regardless of whether you were looking at the 7D or 30D view.
---
mobile positions fix
the market cap column ($xxx → $xxx) was overlapping token names on small screens, making the tokens impossible to read or tap. the MC column is now hidden on mobile. token names take the full available width and remain tappable.
a separate earlier fix gave the positions table 4 clean columns on mobile with a proportionally wider token name column — the layout now works on all screen sizes.
---
agents default to live trading
new agents now initialize with live trading enabled. previously they defaulted to paper mode, which meant every new agent required a manual toggle before it would trade with real funds. the toggle is still there — flip it any time — but the default now matches the intended usage.
---
what's next
stay sharp. parasol keeps you covered.