Track every update to the Liquid Edge platform
Fixed an issue where Studio bot trades could show zero profit and incomplete history. Closed trades now record their exit price and profit correctly.
Trade views now show a dash instead of 0.00% when a trade's return percentage can't be determined, and combined hedge figures no longer count those trades as zero.
Smart filters now clearly mark features that aren't live yet, explain when a filter needs market history before it starts acting, and saving a strategy reminds you that edits only apply to newly deployed bots.
Order size checks now use each exchange's actual minimum requirements, reducing unnecessarily rejected orders.
Fixed a resource buildup that could occur when live data connections restarted, improving stability over long sessions.
The hedge bot dialog now shows live progress for position recovery in both directions, so you can see how your bot is working back toward balance.
Strengthened how the platform identifies the true source of incoming requests, preventing forged headers from bypassing security protections, and fixed a related logging overload issue.
Fixed several issues found during live paper testing of Studio strategies, including trades that could be recorded with the wrong direction and exits that could repeat incorrectly.
Made real-time market data and balance connections more resilient, fixing a reconnection issue that could freeze live prices and reducing memory use in long-running processes.
Hedge bots can now recover positions in both directions and keep complete internal records of every recovery step, improving resilience during difficult market conditions.
The trading engine can now backtest saved Strategy Studio strategies and run them as paper trading bots, with live behavior designed to match backtest results.
Fixed an issue where the public site could crash or display incorrectly when viewed with browser auto-translation enabled.
Locked down internal database functions and tables so they can only be reached by trusted server processes, adding another layer of protection around account and trading data.
Rolled out a redesigned, cinematic public website across the home, walkthrough, pricing, FAQ, checkout and legal pages, with a consistent look and smoother navigation feedback. Headline platform metrics and total volume now reflect live data.
Resolved an issue that prevented referral links from working correctly, so referrals are recognized again.
Resolved an error that could cause the checkout page to fail to load, so plans and pricing display reliably.
The dashboard now shows bot counts with clickable status chips, letting you filter your bots by status at a glance.
Fixed an issue where the live-mode dashboard could get stuck on 'reconnecting' over time. Real-time connections are now released and restored cleanly, keeping live data flowing without a restart.
You can now start a new trading bot directly from a saved strategy, with the strategy's exact configuration captured at launch. The New Trading Bot dialog is also reachable from anywhere in the app.
Your Strategy Studio strategies now save to your account. The new My Library lets you load, rename, duplicate, and delete saved strategies, with clear saving and loading feedback throughout.
Hardened live bots against slowdowns and stalls. Heavy data lookups were moved off the main processing path, positions now recover automatically after interruptions, and a watchdog restores activity if processing ever stalls.
Resolved a database error that could interrupt trade recording, so completed trades are saved reliably.
Reduced redundant database reads and streamlined bot startup so the platform stays responsive as trading activity grows.
Centralized how a bot's status changes are applied, keeping each bot's state accurate and in sync across the whole system.
Every trade is now recorded through a single consistent path, so paper and live trades are handled identically and each trade has a complete, traceable record for diagnostics.
If one side of a hedge position ends up stranded, the bot no longer shuts down entirely; the leftover position is now settled automatically while the bot keeps running.
Improved capture of close confirmations for hedge strategies, with automatic recovery if a confirmation is briefly missed, so closed positions are recorded accurately.
Launched Strategy Studio, a guided builder for creating trading strategies from a library of ready-made templates. Pick a template, preview the expected behavior, and a built-in completeness check guides you to a finished, ready-to-save strategy.
The dashboard now breaks down how much capital is deployed on each connected exchange, giving a clearer picture of where your funds are working.
Win rate now counts only winning and losing trades, excluding break-even ones, and drawdown, net profit/loss, and related figures were corrected for a more honest view of performance.
The live trading engine now detects when it has stalled and recovers on its own, with more truthful health reporting and safeguards against runaway resource use, keeping live trading running without manual intervention.
The dashboard no longer presents delayed account data as if it were live. It clearly indicates when displayed equity may be out of date, and shows a smooth "connecting to live" transition when switching from paper to live trading.
Removing a bot now reliably closes any open positions through a single shared path, preventing orphaned trades from being left behind.
Live account balances now update in real time instead of being polled periodically, so your equity reflects the latest state faster and more consistently.
Real-time connections now detect and recover from frozen or stalled states, keeping live updates and order handling flowing without manual intervention.
Hedge strategy trade history now correctly shows positions that don't yet have a matching hedge, so the history reflects the full picture of open exposure.
The platform no longer saves zeroed or partial account equity when a balance read temporarily fails, preventing brief glitches from corrupting your displayed account value.
Fixed inflated profit-percentage figures shown on some closed trades and corrected affected historical records, so reported returns are accurate.
Hedge and multi-position strategies can now run live on Hyperliquid, expanding the set of strategies available on that exchange.
Eliminated a request spike on first page load and stopped temporary rate-limits from being misread as outages. The dashboard now loads faster and shows fewer false connection errors.
Resolved a crash in credential handling that could prevent bots from starting under certain conditions.
Sign-in now waits for a server-confirmed session before redirecting, and all login paths use one unified, consistent flow — closing a timing gap.
Live DCA Hedge is now limited to one running bot per account. Pause or stop your current live hedge bot to start another one.
The Hedge strategy can now run in live mode on supported exchanges, behind a controlled access gate. Configuration is locked to a validated setup so the strategy runs exactly as tested.
Eliminated duplicate internal trade records that could overstate reported trading volume, so volume and related metrics now reflect real activity.
Applied upgrades resolving seven flagged dependency advisories, keeping the platform's third-party components current and patched.
The Trade History dialog now loads with a skeleton placeholder, fetches one page at a time, and no longer errors or reload-loops on bots with very large trade counts.
Hedge-strategy bots now have a dedicated Trade History view showing open positions alongside completed cycles, with cards reframed around banked profit, live net value, and cycle counts.
Browser password autofill now works on the login page, and the beta welcome notice shows only once per browser instead of on every visit.
Optimization results now report the historical date ranges and conditions behind each result, and run faster on large windows.
Added safeguards that clean up stray orders, confirm uncertain order outcomes before acting, and recover positions more reliably so bots stay consistent during edge cases and restarts.
The main portfolio chart now follows the LIVE/7D/30D/ALL selector, including a real-time live view and clearer paper-balance labeling.
Paper trading now runs on a fixed virtual wallet with a shared capital cap across bots and clearer idle-capital labeling, so simulated results better reflect real allocation limits.
Fixed issues that could interrupt the balance refresh cycle and understate total account equity, so live balances and equity now report accurately.
The Remember me option on the login page now works — your email is saved and prefilled the next time you sign in.
Live market data connections now recover automatically from more disconnect scenarios, and live P&L pricing is more reliable with faster startup and better continuity after reconnects.
Profit & loss and volume now come from one consistent source across the dashboard, and live P&L correctly includes unrealized gains.
The Trading Setup screen now waits for your saved settings to load, so leverage values display correctly instead of defaults.
Fixed issues where certain position closes could be rejected by an exchange and where failed orders could be mistaken for successful ones. Bots now detect these cases and pause or recover safely.
Closing a paper bot no longer contacts the exchange, and unlinking an exchange no longer affects paper bots. Portfolio activation also now resolves correctly in all environments.
The strategy in controlled rollout is now available in paper trading for an expanded set of accounts, allowing safe evaluation before live release.
Updated the Profit Calculator and hero section styling, added a sound cue when switching between paper and live modes, and removed a non-functional chart control.
Each bot's saved leverage setting is now applied consistently throughout the live trading pipeline.
Reduced processing delays under heavy load and fixed a production startup crash caused by a dependency update.
Extensive hardening of an advanced strategy's internal accounting and exit handling, validated by a greatly expanded automated test suite covering recovery and edge cases.
Began development of a new strategy variant with initial validation passing, plus improved pre-launch market condition checks for live testing.
Fixed internal wiring so encrypted exchange credentials are always fetched through the secure provider path.
Fixed a data-retrieval limit that could cause some open positions to be missed when operating at scale.
Expanded the backtesting toolkit with additional indicators and analysis modes, made long runs significantly faster, and improved job handling so failures are surfaced clearly instead of silently.
Added a pre-entry safety filter to the Polymarket strategy that screens out unfavorable conditions before committing, with improved run-time observability.
Reduced dashboard load times through smarter data caching and lighter page rendering, and unified the dark theme background for a more consistent look.
An advanced trading strategy is now available to a limited set of accounts during a controlled rollout before wider release.
Revamped the referrals dashboard, and the default referral code now appears in Campaign Management with a clear Default badge.
Redesigned the help center and ticket detail views for a clearer, more consistent support experience.
Rebuilt the billing experience with a cleaner, more consistent layout while keeping all existing subscription and payment functionality intact.
Abandoned checkouts now cancel their pending invoices automatically, and invoice matching was tightened to prevent plans with similar names from being confused.
Refreshed the points earning page with an updated layout that makes available rewards and actions easier to find.
Restyled the exchange API connection page with a streamlined card layout and a guided setup flow for adding new connections.
Closed several live-only issues in an advanced multi-position strategy that could affect order handling and stored results, improving its reliability on real accounts.
Improved the historical backtesting and optimization service so jobs no longer hang when cancelled, progress reporting stays accurate for all job sizes, data fetching is bounded, and errors surface clearly instead of being hidden.
Updated the crypto dashboard with a cleaner top bar, a Connect Wallet action, and general visual polish.
The backtesting service now authenticates with the platform's managed API keys and handles malformed credentials safely instead of failing unexpectedly.
Strengthened a sensitive database operation with additional safeguards and added regression coverage to protect stored balances.
Improved how an advanced strategy persists and restores its internal state across restarts, so it resumes more accurately after downtime.
Made the real-time market-data feed more resilient to temporary disruptions, with steadier reconnection behavior under unstable network conditions.
Corrected a fee-calculation error in one of the trading strategies, ensuring net results are computed accurately.
Resolved an internal data-routing issue that was preventing bots from opening new positions on both live and simulated accounts. Affected bots now enter trades normally.
Reduced overhead in the live processing loop and added diagnostics that pinpoint sources of delay, helping keep real-time trading responsive.
Completed a major refactor of internal backend components to improve maintainability and reliability and to support future capabilities.
Added an admin interface to mint, list, and revoke API keys, with access controls and rate limiting.
Improved filtering of malicious request patterns and added regression coverage to prevent legitimate pages from being blocked.
Corrected an internal tracking mismatch that could wrongly flag a live position as missing its protective order and trigger unnecessary corrective action. Protective-order state is now tracked consistently from one source.
Position reconciliation now verifies order status directly with the exchange instead of relying on local state, improving the accuracy of paired-position tracking.
Unified how simulated (paper) accounts are detected across the system, so paper bots behave consistently across order routing, reconciliation, and credential handling.
Refreshed the vault enrollment experience with featured options, a clearer enrolled-state display, and updated balance cards.
Restyled the crypto purchase and escrow flows for a cleaner, more consistent experience.
Added a settings panel and quick search to the top navigation for faster access to common actions.
Every position state change now records its reason to durable storage, so operators can trace exactly why a position moved between states even after logs rotate.
Hardened the market-data feed in lightweight consumer mode, preventing crashes during sessions that only need price ticks and improving startup safety.
Corrected database upsert semantics and guarded against a realtime-client reconnection bug, reducing the risk of inconsistent stored state.
Strengthened isolation of the background worker processes used for heavy computation, with restricted deserialization, a scrubbed environment, and resource limits.
Moved portfolio charts, activity history, and drawdown calculations to the server for faster, paginated loading, with results that match across live and simulated accounts.
Rebuilt the dashboard with live-updating portfolio metrics and a 1d / 7d / 30d / all time-period selector, plus refreshed login and navigation.
Fixed reading of pending two-factor authentication factors during setup so enrollment completes reliably.
Trading symbols are now canonicalized at the construction boundary, with a safety-critical entry guard test preventing future format drift from causing mismatched lookups.
Unified the internal trade query API around a single bot-scoped lookup, removing a dual API surface and reducing risk of inconsistent reads.
Tightened Hyperliquid credential validation at startup. Bots missing required credentials are now excluded with a clear error at boot, rather than failing later in trade flows.
Reconciliation now correctly skips paper-mode bots, preventing unnecessary work and confusing diagnostics on simulated accounts.
Tuned an internal memory-pressure threshold to reduce overhead from emergency cleanup cycles during normal operation.
Improved dry-run simulation parity with live and added monitoring passthrough for long-form validation runs.
Fixed an issue where persistence failures while recording paired positions could be silently swallowed, in rare cases allowing the same pair to be re-evaluated. Failures now raise explicitly so they are addressed at first occurrence.
Internal error-tracker cache cleanup is now aborted when downstream status updates fail, preventing the system from holding inconsistent state.
Consolidated trading symbol comparison and normalization into a single source of truth, eliminating subtle bugs where the same symbol could be treated differently by different components.
Reconciliation and recovery flows now surface failures explicitly instead of degrading quietly, so issues during restart or recovery are caught immediately.
Made internal data-stall warnings adaptive so transient slowdowns don't flood logs while genuine stalls remain clearly visible.
Generalized the internal trade repository so multiple strategy types cleanly share the same storage layer, reducing duplication and risk of drift.
Added observability into bundled execution flows so operators can diagnose partial-execution issues quickly.
Improved how quotes are managed on paired positions to close gaps that could otherwise cause execution drift.
Hardened database connection handling with idle cleanup, graceful shutdown drain, and per-phase timeouts to eliminate long-running connection leaks under sustained load.
Removed two diagnostic log statements that could write sensitive credential values into server log files. No user action required.
Closed several race conditions in the internal trade cache so per-bot lookups stay consistent during concurrent processing.
Routed high-frequency bot lookups through a cached service, reducing redundant database calls during per-candle work.
Added regression test gates that catch file-descriptor leaks and unexpected call-rate growth during extended runs, protecting against regressions in long-lived processes.
Standardized how strategy code handles task cancellation so shutdowns and restarts no longer surface misleading errors from in-flight work.
Added a defense layer that limits combined exposure across paired positions to reduce risk during adverse scenarios.
Eliminated a memory leak that grew steadily over multi-hour live runs, significantly improving stability on long-running sessions.
Multi-phase refactor reducing event-loop contention and improving real-time responsiveness during heavy concurrent activity.
Tightened how protective exit orders are placed and updated in live trading, including stricter validation and safer rollback behavior on partial failures.
Improved internal performance diagnostics so the system can pinpoint stalls more accurately and avoid flagging its own diagnostic activity as a culprit.
Refactored exchange real-time connection recovery to remove a fragile background task pattern; reconnects are now handled by a single resilient loop.
Improved order-handling and dry-run accuracy in the polymarket runtime, supporting cleaner evaluation of candidate strategy variants.
Consolidated the live market data path into a single source of truth, eliminating a class of bugs where different parts of the system could see different views for the same symbol.
Strategies now run in parallel across symbols within each batch, reducing end-to-end processing latency on accounts with many active bots.
Replaced several silent-failure paths with explicit errors so that misconfiguration and unexpected states surface immediately instead of degrading quietly.
Added automated monitoring and parity-check tooling for polymarket strategy candidates, enabling safer multi-day observation runs with fail-fast on regression.
Hedge strategy now sizes positions based on per-bot capital allocation and supports independent leverage settings for each side of the pair, with a revamped backtest summary.
Resolved a multi-round incident affecting real-time market data delivery to live bots. Bots now consistently receive the latest market updates without dropped or stalled events.
Refined live polymarket strategy behavior based on detailed analysis of prior run data.
Removed a duplicate startup path that could spawn reconciliation tasks twice when the live trader launched.
Bot lookups now distinguish 'no bot found' from 'database unreachable', so operational failure modes can be triaged accurately instead of being swallowed by a catch-all.
Fixed a cache key collision that could surface stale trade state when a bot ran multiple strategies on the same symbol. Cache entries are now partitioned per bot and strategy.
Live market data is now fanned out per strategy across the dispatch chain, reducing the risk of cross-strategy interference when multiple strategies run on the same bot.
Users can now create hedge-style trading bots directly from the dashboard, with a leverage selector and a clear pairing badge in the trades view that shows which positions are paired together.
Trade history now groups paired hedge trades into a collapsible parent row, making it easier to see the relationship between paired positions at a glance.
Bots now validate exchange credentials before going live, surfacing configuration issues at startup instead of failing mid-trade.
Fixed two issues that could destabilize the live trading process — malformed market data rows can no longer crash the trader, and a class of long-running operations no longer blocks live processing.
Completed live-execution parity for the hedge strategy — paired positions now open and close atomically with reconciled state, and a retry path covers transient close failures so live behavior matches simulation.
Sensitive tables and database functions are no longer accessible from the browser; all reads now flow through authenticated server routes. Removed unused public access grants and closed multiple database security audit findings.
Backend-only tables now have explicit deny-all access policies, ensuring that operational data is never reachable through public APIs.
Each bot now stores and applies its own leverage setting end-to-end, from the dashboard form through validation to live execution.
Updated the whitepaper to v4 and added a walkthrough section to the navigation for new users.
Strengthened active-position safety checks, added continuous health monitoring of protective orders, and improved post-restart recovery so live state is reconciled more reliably. Added a circuit breaker on exchange API calls to prevent cascading failures during outages.
Introduced a new hedge-style strategy that can open partial offsetting positions to reduce directional risk under specific conditions.
Live trading now verifies protective orders are correctly registered after entry and self-heals if any are missing, reducing the risk of unprotected positions.
Tightened parity between backtest and live execution for strategy decisions, including pre-entry order queuing and per-symbol exclusion controls.
Closed multiple detection gaps so the monthly retraining run more reliably catches genuine issues with deployed strategies, with new alerting and conservative defaults that prevent silent escalations.
Strengthened the Polymarket execution path with idempotency on close operations, additional event telemetry, and stricter canonicalization of internal state schemas.
New monthly workflow surfaces strategy candidates and validates them across multiple historical years before promotion to live, raising the bar for what gets deployed.
Added an automatic router that selects between incremental and full retraining paths based on detected drift, cutting unnecessary work while ensuring complete refreshes when needed.
Corrected how warmup history is propagated during backtests so simulation conditions more closely match live runtime.
Added safety circuit breakers that automatically pause Polymarket activity and force-close exposed positions when abnormal conditions are detected.
Redesigned how exit prices are recorded during live fills so trade history matches actual execution outcomes across all event-timing edge cases.
Added an automatic daily safety floor that pauses Polymarket activity if cumulative daily losses cross a configured threshold, providing a hard guardrail against bad-day cascades.
Refined entry filters and added cooldowns between consecutive entry events to reduce noise and avoid stacking similar exposures.
Fixed a condition where overlapping on-chain submissions could fail. The system now detects collisions and retries with a fresh sequence number automatically.
Added exponential backoff to real-time data reconnection so transient outages don't repeatedly hammer the upstream service.
Major hardening of how live bots close positions. Unified close handling with stricter intent contracts and a more atomic protective-order protocol prevents inconsistent state during disconnects, restarts, or partial fills.
Position reconciliation now correctly distinguishes liquidation events from manual closes and withdrawals, preventing mislabeling that could affect performance reporting.
Resolved multiple root causes of a stall that could prevent scale-in entries from progressing under specific conditions. Live DCA bots now advance through entry stages reliably.
Strengthened the Polymarket V2 integration with on-chain verification that rescues falsely reported transaction failures, additional RPC failover, more robust order-cancel handling, and broader exception coverage in the runtime watchdog.
Improved how backtests account for unresolved scenarios so simulation results more closely track what would have happened in live execution.
Added new opt-in safety mechanisms for the Polymarket runtime that limit exposure under unusual conditions and pause activity when health signals degrade. Defaults are conservative — existing behavior is unchanged unless explicitly enabled.
Multi-phase hardening of the live trading runtime: in-process recovery (no longer requires external supervisor restart), heartbeat and pipeline-progress watchdogs that detect stalled bots faster, bounded background pauses to prevent runtime lag, and more resilient WebSocket reconnect handling.
Added support for Polymarket's V2 protocol, including new collateral handling, contract approvals, and a V2-aware launcher. Opt-in via flag; V1 remains the default.
Multiple improvements so backtest results more closely track live execution, including better cross-session reconciliation and a calibration framework for tuning simulation parameters.
Added test coverage for emergency-lockout and canary-watchdog safety mechanisms in the Polymarket runtime.
Migrated the Binance perpetuals price feed to the new endpoint after Binance retired the legacy URL on 2026-04-23. Restores reliable real-time market data for affected bots.
Added a new strategy to the platform with full backtest plumbing and supporting analysis tooling.
Wrapped remaining synchronous calls in the Hyperliquid integration so they no longer block the live event loop, preventing intermittent stalls during heavy activity.
New 7-step public guide page that walks visitors through getting started — no sign-in required.
Corrected the price-data preload so live bots no longer use stale candles for the first several minutes after startup.
The automatic liquidation detection service now requires confirmed close-event evidence from the exchange before marking a bot as liquidated, replacing a heuristic that could misread transient sync gaps as real closes. Includes normalized symbol matching across exchanges and removal of temporary workarounds in favor of the root-cause fix.
Exit prices for closed positions are now read from the actual fill event rather than the protective-order trigger price, correcting small discrepancies in realized P&L reporting on closed bots.
Reconciliation now uses on-chain balance as the authoritative source for executed position size, closing a gap where simulated fills were over-reported during live sessions. Also adds configurable reconciliation cadence and a periodic order-book refresh for longer-running sessions.
Completed removal of a legacy caching layer that was no longer in use. Simplifies deployment by removing an external service requirement and reduces operational surface area.
Fixed a response-parsing bug where submission results were read from the wrong field, causing successfully placed orders to be reported as failures. Submission reliability reporting is now accurate.
Large batches of simultaneous order submissions are now chunked and paced to respect exchange limits, eliminating rejections seen under bursty submission activity.
Added a reconciler that resolves submitted-but-unacknowledged orders after a transient connection drop, preventing orphan orders from lingering in exchange state.
Reduced how long the system waits before treating a silent real-time feed as stale, enabling faster recovery when a connection degrades without an explicit disconnect.
At boundary events, position accounting now defers to on-chain balance as the source of truth rather than an internal estimate, eliminating small cumulative drift.
Added parallel scaffolding and a migration runbook for an upcoming exchange protocol upgrade. The new path runs side-by-side with the current implementation in a dry-run mode to validate behavior before cutover.
Addressed multiple audit-identified gaps between backtest simulation and live execution accounting, including label mismatches, resolution-credit handling, and phase-transition behavior. Backtest numbers are now more faithful to live results.
Added persistence for internal state that previously reset on restart, so the system resumes from where it left off after a redeploy or crash.
Bots that hit liquidation or are auto-closed by the system now display a clear dedicated status instead of a generic terminal state, and action buttons that don't apply to those states are hidden to prevent invalid operations.
Fixed an internal tracking mismatch that caused the system to fall back to a slower path when recording protective orders on new positions. Protection is now captured on the first attempt, reducing setup latency.
Added a background service that automatically detects when a bot position has been liquidated by the exchange and cleanly closes the bot state, recovering trapped resources. Feature-flagged and off by default pending broader rollout.
The public landing page was rebuilt on a lighter, more consistent design system. Faster first load and cleaner presentation on mobile.
Cleared an internal flag that could linger after a profit-taking order was confirmed live, preventing a rare edge case that could cause unintended duplicate orders under specific timing conditions.
Extended the timeout for processing real-time exchange messages during active periods, reducing spurious disconnects when activity is bursty.
Fixed a condition where an internal state tracker could persist across cycles if a prior cycle stalled, leaving live bots frozen with pending entries permanently rejected. Added defensive cleanup and new observability so any recurrence is surfaced immediately.
Introduced an opt-in entry guard for ETH that filters out entries matching a condition pattern that has underperformed historically and worsened under 2026 market conditions. Default-off for all other assets.
Added a new backtest reporting layer that models the live operator workflow of withdrawing profits and topping up losses at month boundaries. Runs alongside the standard backtest with no impact on core results.
The monthly retraining workflow now works for any trading pair with sufficient history, not just one specific asset. Added a diagnostic that automatically surfaces potential regime shifts during routine maintenance, so parameter drift is caught before it materially affects performance.
The backtest engine was missing three recently-added live execution features, which biased absolute performance numbers and distorted filter-selection rankings. All three have now been ported with full test coverage. A new candidate configuration was identified through extended parameter search; it awaits a live canary review before deployment.
Trimmed verbose logging in the order execution and credential paths. Removes log lines that could carry references to decrypted credential material, improving operational security.
An audit found the previously-deployed ETH configuration was underperforming its own baseline under recent market conditions. A new configuration was derived from multi-year analysis with an out-of-sample held-out window and deployed.
Dry-run simulation now covers additional live execution mechanics that were previously omitted, improving fidelity between simulated and live performance. Known remaining divergences are documented for reference.
Fixed a bug where positions added manually by users counted toward the automated strategy's entry cap, causing the strategy to stall after manual top-ups. The cap now only counts strategy-driven entries, while position averaging continues to reflect all positions correctly.
Resolved a regression that briefly blocked DCA entries and scale-ins, including both live and paper-mode orders. The internal order contract is now consistent across every execution path, with additional regression tests added to prevent recurrence.
Restored end-to-end dry-run validation against live market data, allowing execution changes to be tested against real flow without placing real orders. Profit accounting in dry-run now matches live behavior so simulated runs are directly comparable.
Added an opt-in safety mechanism that detects sustained one-sided inventory buildup within a window and halts further participation on the imbalanced side. Disabled by default; emits dedicated audit events when triggered for full transparency.
Closed an edge case where protective stops could move during the bar that triggered partial profit-taking. Stops now stay frozen until the triggering bar closes, matching simulation behavior exactly. Also improved exit-price capture on partial closes so trade reports show real fill prices instead of approximations.
Fixed Hyperliquid balance queries to include spot stablecoin holdings, repairing the bot resume balance check. Bots paused on Hyperliquid now resume reliably without false insufficient-balance errors.
Added an opt-in execution behavior that actively closes unpaired inventory toward the end of a trading window, reducing exposure carried into quieter periods. Guarded by independent caps and disabled by default.
Recovered two customer positions that had been stuck open for more than 10 hours after a rare strategy-exit edge case. Added operator tooling to safely detect, close, and reconcile stuck positions going forward, with strict double opt-in safeguards and audit logging.
Fixed a Bitget position-close path that could fail with a generic error message. Failures now surface the exchange's specific error code and details, making issues significantly faster to diagnose and resolve.
Closed three small gaps surfaced from live session auditing: stronger restart safety for protective orders, clearer attribution when orders are cancelled by the system, and more complete event logging during slow-market adjustments.
Recalibrated order sizing to better align with the profile of successful market-maker activity observed in the space, and added an opt-in filter that avoids participating in historically low-quality trading windows. Both behaviors are guarded by independent caps and can be disabled for testing.
Fixed a critical bug that was causing DCA bots to fail on exit and get stuck holding positions. All affected bots now exit cleanly.
Fixed a rare condition where a missing exchange event could be misread as a completed exit, leading to incorrect position sizing. Exits now require confirmed fill evidence before advancing.
DCA order placement is now fully restart-safe across all supported exchanges. Retries and reconnects will no longer result in duplicate orders.
Fixed an issue where the reported exit price could show as unknown on certain protected exits. Real-time exit data is now captured reliably for accurate reporting.
Migrated Bitunix order cancellation to the current supported endpoint. Cancel requests that were previously failing silently now complete successfully.
Added two execution behaviors observed in top-performing market makers: faster capital reuse within a trading window, and automatic replenishment of filled orders. Both paths are guarded by independent safety checks and conservative caps.
Improved reliability of the Polymarket market-making pipeline: detects and recovers from silent data stalls, backs off gracefully on rate limits, reconciles execution state against on-chain truth, and uses more accurate book-diff detection to reduce false drift signals.
Updated the backtest fee calculations to match Polymarket's current published fee schedule. Historical results now reflect real-world economics more accurately.
Tuned the protective stop behavior so that favorable price moves are captured more effectively. Validated through extended historical analysis with consistent positive results across multiple years.
Hardened DCA entry-cap enforcement so concurrent cycles can no longer exceed the configured maximum entries. Also restored correct bot and symbol attribution in audit logs, fixed a memory leak in the decision service, and added explicit exit-confirmation logging.
Redesigned the market-making bot based on competitive analysis of profitable wallets. Widened price range, increased position sizing, and removed an early-exit mechanism that was cutting off a significant portion of late-arriving fills.
Added on-chain split and merge integration so unsold token pairs are automatically recombined back to stablecoins at the end of each trading window, eliminating directional exposure from one-sided fills.
Fixed critical concurrency issues, added resource locking to prevent capital overdraw, corrected profit tracking on failed operations, and added circuit breakers for capital and daily loss limits.
Discovered and fixed a data fetcher bug that silently dropped half the day trading data. All historical analysis and strategy validation prior to this fix was operating on incomplete data.
Exhaustive analysis of real production trades confirmed no profitable parameter configuration exists for the directional prediction strategy. Deprecated in favor of the market-making approach.
Independent security review of the new split and merge contract interaction code. Verified correct contract addresses, proper key handling, no race conditions, and standard approval patterns.
Comprehensive overhaul of the parallel trading pipeline with new checkpoint logic, better queue management, connection resilience, and crash recovery.
Overhauled the backtesting engine to simulate real-world trading conditions more faithfully, accounting for competition from other traders. Predictions now more closely match observed live performance.
Added a minimum movement filter that skips low-confidence predictions during quiet market conditions, reducing noise-driven losses.
Multi-parameter sweeps now run significantly faster by computing shared data once instead of redundantly for each variant.
Built a new market-making simulation tool that tracks orderbook position, fill probability, and economics in real-time using live trade data.
Refined the market-making simulation to account for earlier positioning and dynamic participant behavior, producing more accurate fill estimates.
Added the ability to fetch current-day trade data for same-day analysis and validation, enabling faster iteration on live strategy performance.
Fixed internal fee calculations to match the exchange's actual formula, preventing accounting discrepancies in profit tracking.
Rebuilt the market-making engine to monitor 12 prediction windows simultaneously instead of one at a time, increasing coverage from 24 to 288 windows per day with continuous operation.
Added automatic position adjustment that detects one-sided market activity early and reduces exposure by locking filled positions and keeping only matching orders active, minimizing risk on losing outcomes.
Validated the new position management approach against 35 days of historical data across nearly 10,000 prediction windows, confirming improved risk-adjusted returns.
Added comprehensive documentation covering the market-making pipeline architecture, position management logic, and operational procedures.
Ensured the DCA strategy produces identical behavior in both simulation and live environments, so performance in production reliably matches what testing predicts.
Validated simulated trades against real production results and corrected all discrepancies, so backtest projections now closely match actual trading outcomes.
Launched a public changelog page where users can browse all platform updates organized by date, filter by product area and change type, and expand entries for additional details.
Fixed issues in the order execution flow that could lead to unintended duplicate orders, inaccurate profit reporting, and incorrect position sizing under certain conditions.
Resolved a resource leak that accumulated unnecessary connections over time for idle instruments, improving long-running system stability.
Improved trade entry timing and exit execution reliability based on extended historical analysis, resulting in fewer failed exits and better fill quality.
Resolved an issue where exit orders could fail due to a mismatch between estimated and actual position sizes caused by fee calculation differences.
Strengthened system restart recovery so active orders are properly restored after interruptions, preventing stale orders from remaining on the exchange.
Resolved the root causes of recurring server errors including connection timeouts and background process failures, resulting in a more stable production environment.
Launched a new trading mode that uses limit orders for improved pricing, with real-time monitoring, reliable fill detection, and a complete audit trail.
Replaced unreliable trade fill detection with direct verification from the exchange, eliminating missed or incorrectly reported fills.
Fixed a critical issue that caused the trading server to become unresponsive for extended periods by isolating heavy background operations and preventing cascading recovery failures.
Upgraded the position tracking system to more closely mirror actual exit behavior, so internal tracking stays consistent with real exchange state.
Corrected the live data feed source to match what backtesting uses, eliminating price discrepancies that could cause trades to trigger at different times than expected.
Added safety guards that pause trading activity during abnormal market conditions, reducing exposure to low-liquidity and high-volatility situations.
Resolved an issue where protective orders could fail to activate after trade entry due to an internal tracking mismatch, ensuring positions are always properly guarded.