Research notes
Ongoing work on Polymarket bots, BTC five-minute markets, backtest reality, risk and automated execution.
Find answers by topic
The full chronological archive remains available; you can also start with a question or research topic.
What Must Not Change During a Trading Bot Upgrade? Production Replay Invariants
A practical look at invariants in automated trading upgrades: decision inputs, state, replay consistency, independent scoring, auditability, and what historical verification cannot prove.
More Trading Rules, Fewer Layers: Why I Simplified the Decision Architecture
A research-driven trading system can accumulate too many decision layers. Here is why a unified architecture, scoped dependencies, and auditable outcomes matter more than simply adding rules.
One Missing Data Dependency Should Not Break Every Trading Decision
How to scope optional data dependencies in a trading automation system without confusing local feature gaps with critical stale feeds or unsafe execution states.
64 Positive Monthly Scores, Yet a Weak Recent Week: How to Read Both
All 64 historical calendar-month segments scored positively, yet the recent seven-day window was weak. A guide to aggregation, rolling windows, drawdown, and forward validation.
619 More Trades, Yet a Smaller Drawdown: What Does That Tell Us?
In a same-context BTC 5m replay, v2.4 executed 619 more decisions while maximum standardized drawdown improved from -58 to -52. What can—and cannot—be inferred?
How Do You Know a Trading Bot Upgrade Is Actually Better?
A same-context BTC 5-minute replay shows why win rate alone cannot validate a bot upgrade: evaluate score, drawdown, time distribution, code consistency, and out-of-sample limits.
Random SIGSEGV After Nuitka Compilation: Tracing a Python 3.11+ Frame Use-After-Free
A long-running Python trading client started crashing after an SDK upgrade—but only as a Nuitka binary. We trace the evidence from concurrency and SSL to an upstream compiled-frame use-after-free and outline a controlled fix and validation plan.
From a Trading Bot to a Maintainable Trading System
An architecture upgrade focused on deployment, online updates, and long-term maintainability. A trading system is not only about strategy quality; it also has to be easy to install, update, recover, and keep running.
Why I Rejected a Strategy Change That Made the Backtest More Profitable
Higher historical return does not necessarily mean structural improvement. This article explains why several stronger-looking exit and filtering candidates were rejected.
Why I Abandoned a Risk-Control Method That Made the Equity Curve Smoother
A smooth leverage-reduction method reduced drawdowns in BTC futures research, but it was ultimately rejected because it did not distinguish trade quality.
Why I Use Binance for BTC Signals and OKX for Execution
Why a BTC futures trading system can use Binance for signal research and OKX for real execution, and why separating research stability from execution realism can produce a cleaner architecture.
A Futures Trading Bot Is Not Just a Signal Strategy
A real automated trading system is more than BUY/SELL signals. It also needs execution gates, dynamic sizing, account risk state, and position management.
High Leverage Is Not Just Multiplying Position Size by 10
Why 10x or 20x in a BTC futures system should be treated as a leverage ceiling rather than a fixed setting, and how trade quality, account state, and path risk interact.
Why the Same BTC Strategy Behaves Differently on Binance and OKX
A cross-market replay study of using Binance BTCUSDT 5-minute data for signals and OKX BTC-USDT-SWAP for execution, and why small price differences can change drawdowns, exits, and the equity path.
61 Positive Months — Why We Still Weren’t Satisfied With This Polymarket Bot Strategy
A Polymarket BTC 5-minute research candidate produced 61 positive months, yet weekly, daily, and out-of-sample results showed why no losing months does not automatically mean full-cycle stability.
Why You Can’t Just Run a Polymarket Bot Strategy Earlier
Research on why a Polymarket bot cannot simply shift its final strategy earlier, and how direction, participation, and strict time ordering differ.
Why Does the Polymarket Order Book Suddenly Jump? BTC, TWAP, and Execution Timing
Research on short-horizon Polymarket order-book repricing, BTC spot/TWAP dynamics, and why this information may be more useful for execution than standalone direction prediction.
Does a High Polymarket Bot Win Rate Guarantee Profit? Entry Price Matters
Why a Polymarket bot win rate cannot be interpreted as profitability by itself, and how entry price, fillability, and standardized replay fit together.
Nearly 80% for Two Days—What Happened Over Five Years?
A Polymarket bot research case study on how a near-80% short-sample result changed after nearly five years of validation, and why robustness matters more than a headline win rate.
Polymarket Bot Execution Research: Why Does Entry Get More Expensive Near Market Open?
Research on why effective entry cost often rises near the BTC 5-minute market boundary, and why a Polymarket bot must balance prediction quality with execution price.
Should a Polymarket Bot Have Only One Entry Time?
A high-level research discussion on whether a Polymarket bot may benefit from multi-stage decisions instead of forcing every trade into one fixed entry time.
Why We Archived Five Years of BTC 1-Second Data for Polymarket Research
Why serious Polymarket BTC 5-minute research benefits from a long-span second-level archive, and how reproducible, validated, time-ordered data reduces research error.
Polymarket Bot Five-Year Replay: What Happened After Execution Coverage Expanded Nearly 8×?
A nearly five-year replay study of a Polymarket BTC 5-minute trading bot, examining what happened to win rate, drawdown and long-term stability after execution coverage expanded by almost 8×.
Polymarket BTC 5m Research: Can a Binance-Derived TWAP60 Be Reliable Enough for Historical Research?
A real-data calibration of Binance 1-second Close and OHLC4 TWAP60 proxies against Chainlink BTC/USD TWAP 60s, followed by a five-year reconstruction robustness test.
Polymarket Bot Research: Can Binance–TWAP60 Divergence Predict the Next BTC 5-Minute Direction?
Using Binance BTCUSDT 1-second data to reconstruct a TWAP60 proxy and study whether spot–TWAP divergence predicts the next Polymarket BTC 5-minute direction.
Polymarket BTC 5-Minute Trading Bot Research: Why Strategy Stability Matters More Than Win Rate
A research note on Polymarket BTC 5-minute trading bots, using a strict five-year production-code replay to examine win rate, monthly and weekly stability, drawdown, multi-branch strategy coverage, dynamic position sizing, and why long-term stability matters more than a single backtest number.
Why Chainlink Makes Polymarket BTC 5-Minute Bots Harder to Research
A research note on the three-layer data structure behind Binance, Chainlink BTC/USD TWAP 60s, and Polymarket—and why this settlement design raises the difficulty of backtesting and live strategy research.
Polymarket Bot Backtesting: What Does a 63.75% Win Rate Actually Mean?
A deeper look at Polymarket bot backtesting through 18,633 historical signals: win rate, payoff structure, STOPs, drawdown, losing streaks, overfitting, and the gap between replay and live trading.
How a Polymarket BTC 5-Minute Bot Actually Works: From Signal to Automated Execution
A full, strategy-safe walkthrough of a Polymarket BTC 5-minute bot: market discovery, data continuity, signals, regime context, no-trade decisions, sizing, risk and execution.
Why a Polymarket Trading Bot Must Learn Not to Trade
A Polymarket bot should not only choose Up or Down. Using 348 public historical STOP events, this article explains why NO TRADE, regime filters, and candidate rejection belong inside the strategy.
No Look-Ahead Isn't Enough: The Other Side of Polymarket Bot Overfitting
A Polymarket bot can avoid future-data leakage and still overfit. This article separates causal replay from research overfitting and explains time splits, rule selection, parameter neighborhoods, and rejected hypotheses.
A New Rule Looked Great Recently. Why I Still Rejected It.
One of the most dangerous moments in Polymarket bot research is when a new rule perfectly fixes recent losses. This is how we test whether a candidate is real—or just a clean explanation of known history.
A Good Backtest Is Not Enough: Why Production Code Must Reproduce It Trade by Trade
Matching total PnL is not enough for a Polymarket trading bot. This article explains why we validate per-event timing, direction, STOP decisions, state, and unchanged historical regions.
Polymarket BTC 5m Is Not a Binance Candle: Why Resolution Sources Matter
Many BTC 5-minute strategies use Binance data for research, but Polymarket resolves markets under its own rules. This article explains Chainlink, TWAP, label definition, and backtest mismatch.
61 Positive Monthly Scores — Why I Still Won't Call It “61 Profitable Months”
How should monthly stability be interpreted in a Polymarket bot backtest? This article separates standardized monthly score, live profit, sample distribution, and research overfitting.
Why Do So Many Polymarket Bot Strategies Die in Live Trading? Three Public Failure Patterns
Public Polymarket bot research logs reveal recurring strategy failures: confusing high probability with edge, paper fills that cannot execute, and repeated parameter tuning that fits known history.
No matching articles. Try different keywords.