Smarter Strategy, Sharper Reporting, Safer Operations
A May 2026 product update on UniswapAutoBot.
Over the past two weeks the bot graduated from "automated rebalancer" to a genuine decision-making system: it now reasons about market regime, verifies its own ideas against historical data before acting on them, and reports performance with the honesty of a fund statement.
This update walks through what changed and the reasoning behind each choice — written for users and operators, not engineers.
Table of Contents
- A strategy engine that explains itself
- Auto-tuning, with a safety net
- A realistic backtest harness
- Live configuration, zero downtime
- Reporting that reflects reality
- Tighter risk controls
- A more transparent dashboard
- A richer assistant interface
- What this means for you
A strategy engine that explains itself
The strategy evaluator now combines several proven signals into a single, auditable decision — and every decision tells you why it happened.
Trend filters that respect the bigger picture
A higher-timeframe EMA-slope filter now acts as a hard veto on closing decisions that would fight the prevailing trend. Combined with ADX as a regime gauge, the bot is less likely to exit a winning position into a shallow pullback. The strategy stays aligned with the dominant market context instead of overreacting to local noise.
Chandelier exits that wait for proof
The new chandelier-exit gate only arms after a position has been open long enough and is showing real profit. The result is fewer early stop-outs on positions that just need a moment to breathe — protective behavior triggers when there is actually something to protect.
Stable-riding and hysteresis refinements
Range tightness and rebalance hysteresis were re-tuned so that ranging markets no longer trigger unnecessary churn. The channel refits adaptively while price walks within its band, preserving fee capture instead of cycling gas.
Auto-tuning, with a safety net
The bot can now propose its own parameter adjustments — but it proves them before it ships them.
LLM proposals, validated by backtest
When the auto-tuner suggests new parameters, those proposals are first replayed against recent historical data inside the keeper itself. Only proposals that improve simulated performance get persisted to live configuration. Bad ideas die quietly in the simulator instead of on production capital.
This is the design principle behind the whole feature: an automated system that adjusts its own behavior must prove that the adjustment is an improvement, not just a change.
Grounded in real fee data
The auto-tuner now reads collected fees from the source of truth in the database rather than estimating them. Its feedback loop measures the same numbers your portfolio statement shows — there is no gap between what the tuner optimizes for and what the user actually receives.
A realistic backtest harness
A first-class simulation environment now powers both research and the guard-railing of automated changes.
Execution simulation, not just price replay
The backtest harness models the things that actually move PnL: rebalance cadence, swap slippage, gas costs, and the asymmetry between quoted and realised execution. Strategy ideas are evaluated under conditions that look like the live market, not an idealised one.
One engine, three use cases
The same harness is exposed as an in-process function, which is what makes the auto-tuner safety net practical. Research, validation, and live decision-making all share one engine — so a result that looks good in backtest carries real weight when it reaches production.
Live configuration, zero downtime
Parameters can now be adjusted without restarting anything.
Two-minute SSM polling
Runtime configuration is centrally stored and refreshed by each keeper every two minutes. Adjustments — whether human-driven or made by the auto-tuner — propagate automatically without operator intervention.
State that survives redeploys
The evaluator's in-flight context (entry markers, arming state, trend memory) now persists across deploys. Releasing a new version no longer resets the bot's situational awareness, so improvements ship without losing the strategy's accumulated context.
Reporting that reflects reality
The performance report is now an honest mirror of what the strategy is actually doing relative to a passive baseline.
HODL baseline tracks deposits and withdrawals
External wallet inflows and outflows are now folded into the passive-hold baseline. Adding funds mid-period no longer makes the strategy look artificially better than just holding — the comparison stays apples-to-apples.
Internal swaps no longer pollute the basket
The bot routinely rebalances between assets to maintain the target range. Those internal legs are now correctly separated from real inflows and outflows, so the basket-change detector reports only genuine portfolio movements.
Valuation consistency across token pairs
Reporting math has been audited and corrected so that whichever token is the stable side of a pair is valued consistently. The dollar number on the dashboard matches the dollar number in your wallet.
Transient close-spikes suppressed
Portfolio time-series no longer show artificial spikes during the brief window when a position is being closed and converted. The chart shows the position, not the plumbing.
Tighter risk controls
Several small-but-important safety improvements landed in the same cycle.
Smarter ETH reserve validation
When closing a position, the bot now correctly anticipates the ETH that the close itself will release, instead of conservatively assuming reserves before the close. Operations that were previously gated by overcautious checks now proceed safely.
Evaluator state aligns with external actions
If a rebalance happens outside the evaluator's normal flow, the evaluator now clears its derived state instead of acting on stale context. The strategy's internal view of the world stays in sync with on-chain reality.
Wallet transfer tracking
Direct wallet transfers (deposits and withdrawals outside the bot's control) are now first-class events. Operators have a full ledger of what moved, when, and why.
A more transparent dashboard
The new Strategy page turns the evaluator from a black box into a glass box.
Wait-status badges and hold reasons
When the bot is intentionally holding rather than acting, the dashboard now shows exactly which conditions are gating the decision — a missing trend confirmation, a not-yet-armed exit, a cooldown window. No more guessing why nothing is happening.
Signal detail modal
Every evaluated signal can be expanded into a full decision-chain modal showing every input the evaluator considered. It's the same view the engineers use to debug strategy logic, now available to everyone.
Parameter legend
The Strategy page now ships with an inline reference for every parameter that drives the engine, so the relationship between configuration and behavior is always one glance away.
A richer assistant interface
The bot's machine-readable interface (MCP) gained parity with the dashboard, which makes conversational and programmatic workflows much more capable.
Full parameter awareness
The assistant interface now understands every tunable parameter in the system. Asking "what would happen if I changed X?" or "why did you do Y?" produces grounded answers because the tool has the same context the strategy does.
Evaluator state, HODL changes, and backtest-vetting status
The current evaluator state, the most recent basket changes, and whether a configuration update has been backtest-vetted are all exposed as first-class queryable surfaces. Operational visibility extends naturally into automated tooling.
What this means for you
Three threads tie this release together:
- The bot thinks more carefully. Multi-signal regime detection, armed exits, and trend-aware vetoes replace simpler heuristics.
- The bot proves its ideas before acting on them. Auto-tuned parameter changes are validated against a realistic execution simulator before reaching production.
- The bot shows its work. Reporting honestly accounts for deposits, withdrawals, and internal swaps; the dashboard exposes every reason behind every decision.
Two weeks. One engine that thinks, validates, adapts — and shows its work.
About UniswapAutoBot
UniswapAutoBot is an automated liquidity management system for Uniswap v4, running concentrated liquidity strategies across Ethereum, Polygon, Arbitrum, Optimism, Base, and Unichain. Each user runs an isolated, non-custodial keeper with a private dashboard and a programmatic assistant interface.
Learn more: ubamm.ai