AI Auto-Trading Experiment — Full Timeline

Built the full system structure: KIS API integration, rule engine, Telegram alerts, and scheduler. Closed this phase waiting for the first automated execution.

• 2026-04-07 v1.0

  KIS API, Rule Engine, Telegram, Scheduler — MVP Complete

  The system first became capable of placing real orders. Loss limit safeguard, Telegram alerts, and 3×/day scheduler included.

  View Log →
• 2026-04-08 v1.1

  Dual Market Regime Filter / Volume Filter / Same-Day Re-Entry Block / Log Overhaul

  Configuration complete to ensure the system doesn't buy when it shouldn't. The core is a structure that does nothing when conditions aren't met.

  View Log →
• 2026-04-09

  Python Module Cache Issue Identified — Buy Cycle Restored

  Buy cycle dropped on scheduler restart — found and fixed. First SOP written.

  View Log →
• 2026-04-10

  Cash Balance Field Fix / API Retry / Duplicate Buy Prevention / IS_PAPER Safeguard

  Production-grade stabilization. Cash balance API field error found and replaced. Final stability work before live switch.

  View Log →

One week of paper trading to validate strategy and debug. Trailing stop fired for the first time, universe expanded, scan cadence optimized. Ended with the decision to switch to a live account.

• 2026-04-13

  Paper Trade Week 1 — Regime Filter and 5-Slot Management First Validated

  First week with all 5 slots filled. Experienced the regime filter blocking buys on a down day.

  View Log →
• 2026-04-15

  Trailing Stop Fires for the First Time — Samsung Electronics +3.57% Auto-Sold

  The moment the rules worked. The system made a sell decision without me. Realized P&L: +₩15,000.

• 2026-04-16

  Buy Scans 4→3/day / Universe 5→20 Tickers / Deviation 5%→7% / Max Holdings 3→5

  Final parameters for live trading confirmed. Widened the universe, relaxed conditions to be realistic.

• 2026-04-17 v1.3

  Paper Trade Complete — 4 Bugs Fixed, Decision to Go Live

  Weekly profit: +₩45,060 (+0.58%). Trailing stop validated. 4 bugs fixed. Live account switch set for Monday at 08:50.

  View Log →

Real money moved for the first time. First order, first stop-losses, and bugs that didn't surface in paper trading appeared live.

• 2026-04-14

  First Real Order — Kakao 20 Shares @ ₩48,350

  DRY_RUN turned off. Trailing stop refinement, balance fallback, post-market Telegram summary added. Real money moved for the first time.

  View Log →
• 2026-04-20

  Live Day 1 — 2 Bugs Fixed, 5 Positions Opened

  Stale token cache and MA60=None bugs fixed before first buy at 10:33. Kakao, Kia, KB Financial, Shinhan, NAVER. ₩4,690,300 deployed. End-of-day: –₩54,030 (–1.08%).

  View Log →
• 2026-04-21

  MA60 API Replaced — inquire-daily-price → inquire-daily-itemchartprice (65 candles)

  Root cause of MA60 data shortage resolved. Fixed 30 candles → 65 candles. Validated in live trading the next day.

• 2026-04-22

  Live Day 3 — 2 Stop-Losses, MA60 Filter Validated

  KB Financial and Kakao stopped out: –₩61,800 total. MA60 filter correctly blocked 3 downtrend stocks — API replacement from 4/21 confirmed working. Closing balance ₩4,906,290.

  View Log →

Strategy redesigned based on live trading data. The biggest shift was from "a system that fills slots" to "a system that selectively holds only strong candidates." The current stage adds observation structure before trusting backtest-driven strategy changes.

• 2026-04-23

  3 Live-Account Bugs Fixed Same Day

  3 bugs found during live operation — identified, fixed, and deployed the same day. Operational stability improved.

  View Log →
• 2026-04-24 v1.4

  5 Bugs Fixed · 6 Strategy Improvements

  Wrong cash field (ord_psbl_cash_amt) corrected — orders restored. 18 duplicate KT sell reservations removed. Regime detection switched to 5-condition scoring + 2-day confirmation + total-asset-linked cash ratios. Closing balance ₩4,890,000.

  View Log →
• 2026-04-27 v1.5 BUY_DRY_RUN

  RS Filter Added — Stock Selection Philosophy Overhauled

  KOSPI hit an all-time high; 4 of 5 positions were down. Root cause: slot-filling structure. KOSPI200 excess return below -3% disqualifies candidates; forced slot-filling removed. Files: config.py / rule_engine.py / main.py.

  View Log →
• 2026-05-01 v2.0 Plan

  v2.0 Upgrade Direction — Designing a Seed-Scalable Structure

  Reviewed the high-priced blue-chip constraint in the ₩5M structure and set the direction toward total-asset-ratio-based position sizing instead of ticker-specific exceptions. RS, MA20 deviation, stop-loss, and holding-period rules were not relaxed.

  View Log → View Insight →
• 2026-05-04 v2.0 Check

  First Trading Day Check After the Weekend Upgrade

  Under a roughly ₩10M live-account structure, high-priced stocks can now be evaluated inside the portfolio, but the system still does not buy when conditions fail. LG Electronics trailing sell: +₩12,000, new buys: 0, and SK Hynix was blocked by the MA20 deviation rule.

  View Log → View Insight →
• 2026-05-08 Engine Refinement

  Strategy Engine First Refinement — Capital Structure, Stop-Loss, Target Profit, Sector Limits, RSI Momentum

  Based on live-account operating data: ₩10M baseline / ₩2M per position, stop-loss -2.5%, target profit +5.0%, 20% cash buffer, sector overlap limit, RSI momentum. Profit-risk log moved to the beginning of candidate review. Zero DRY RUN buys interpreted as normal defensive behavior under MA20 deviation filter.

  View Log → View Insight →
• 2026-05-11 Regime Upgrade

  KOSPI Regime Detection Added — Dynamic MA20 Deviation Threshold by BULL / NEUTRAL / BEAR

  Root cause of repeated zero-buy days (fixed 5% MA20 ceiling) analyzed and resolved with KOSPI 5-day return regime detection: BULL 8% / NEUTRAL 5% / BEAR 3%. Universe expanded 20→40, 11:00 BULL-only buy cycle added, pullback filter (OR structure), RSI ceiling 65→70 (75 chase-block kept), partial profit-taking (+6% = 50% exit), tiered trailing relaxation, daily buy limit 2→3, settlement-price pnl tracking, consolidated sell alerts.

  View Log → View Insight →
• 2026-05-22 RS Validation

  RS Benchmark Skew Checked and Post-Rejection Tracking Added

  Investigated why live operation produced zero orders. Six BUY signals fired, but no stock passed RS. Compared ETF, universe simple average, market-cap weighted, and candidate-weighted benchmarks. Benchmark replacement was held back, and D+1/D+3/D+5 tracking for RS-rejected stocks was added first.

  View Log → View Insight →
• 2026-05-28 2-Week Observation

  Backtest Bias Checked, Current Rules Kept, shadow_report Prepared

  Removed look-ahead bias by switching from same-day close execution to T+1 open execution, and confirmed that applying today's 40-stock universe backward creates bias. Compared old-15, loosen-30, and MA20 deviation ceilings of 8%, 12%, and 15%, but kept the live rules unchanged and moved into two weeks of live logging plus shadow_report observation.

  View Log → View Insight →