# Day 5. How to Honestly Measure Your Bot's Performance — and What to Do Next > Today's goal is to learn how to **read** what your bot has accomplished over the past few days, distinguish **signal from noise** in the results, and understand what the **discipline of iteration** looks like. This is the most important day of the five. Programmers skip it most often, and that's precisely why most bots die prematurely. --- ## Why Bother Measuring Anything Today? By now, you have a bot running on a test account. It has made some trades. Some were profitable, some were not. It's very tempting to look at the number in the top right corner and say "it's working!" or "it's not working!" This is the **worst possible** way to make a decision. A small experiment. If you flip a coin 10 times, you might get 7 heads and 3 tails. Does this mean the coin is **broken**? No. It means the **sample size is too small** to draw a conclusion. If you flip it 1000 times, you'll get something close to 500/500. It's the same with a bot. **Ten trades prove nothing**, neither positive nor negative. **One hundred trades** start to show something. **A thousand trades** are statistically significant. But programmers—especially beginners—are constantly tempted to make decisions after **three days**. "The bot is losing—rebuild it!" "The bot won—leave it as is!" Both decisions are equally wrong. This day is about not being that programmer. Today, we'll cover: 1. Which metrics to watch (a short, honest list) 2. What the metrics **don't** mean (common illusions) 3. Walk-forward—the concept of a fair **on-paper** check, as an idea 4. The iteration cycle: one hypothesis → one test → one change 5. When to take a **break**—psychological discipline 6. Criteria for deciding whether to "continue or quit" after 6 months And at the very end—**what's next**, for those who seriously want to go deeper. This will be **at the very end**, with no pressure. If you decide that the five-day challenge is enough for you, that's perfectly fine, and **thank you for going through it with me**. --- ## Which Metrics to Watch Most people only look at one number—**profit or loss**. This is too crude. You need several metrics that, **in combination**, show whether the bot is viable. ### Metric 1. Win-rate (percentage of winning trades) The percentage of trades that closed in profit. For example: out of 50 trades, 28 were profitable—the win-rate is 56%. **How to read it:** - 70-80% and higher is almost always **suspicious**. It's likely the result of backtest overfitting, or the strategy will soon break. Real, good strategies are rarely this successful. - 50-60% is a healthy range for most types of strategies. Not ideal, but normal. - 35-50% can be normal for **trend-following** strategies, where there are fewer wins, but they are larger. - Less than 35% is usually bad. Either the entry logic is broken, or the stops are too tight. **A common misconception:** win-rate by itself **does not indicate profitability**. You can have a 90% win-rate and still lose money if the infrequent losing trades are large. Conversely, a 40% win-rate with large wins can be profitable. ### Metric 2. Risk-Reward ratio (average profit to average loss) Divide the average profit by the average loss. - If the average profit is $20 and the average loss is $10 → RR = 2.0 - If the average profit is $5 and the average loss is $15 → RR = 0.33 **How to read it:** - RR < 1 — each winning trade brings in **less** than a losing trade loses. To be profitable, you need a **very high** win-rate. A difficult combination. - RR = 1.0-1.5 — normal for counter-trend strategies with a high win-rate. - RR = 1.5-2.5 — healthy for trend-following strategies. - RR > 3.0 — sounds great, but in reality, it often means the **stop-loss is too tight**, and you're taking many small losses in a row. ### Metric 3. Max drawdown The **deepest** drop in capital from a peak to a subsequent low, in percentage terms. For example: the bot grew from $1000 to $1500, then fell to $1200—a 20% drawdown from the peak. **How to read it:** - 5-15% — an excellent result for an active strategy over a month. - 15-30% — normal; many professionals operate in this range. - 30-50% — painful, but not a death sentence. - More than 50% — usually a sign that the strategy is **flawed** and needs to be reworked. To recover from a 50% drawdown, you need to grow by 100%. **The main rule:** the drawdown you **can psychologically tolerate** is usually half of what you thought you could when you were calm. If a 15% drawdown is already causing you to lose sleep, your real tolerance is 7-8%, not 15%. Remember this. ### Metric 4. Profit factor Divide the sum of all winning trades by the sum of all losing trades. - If the total profit is $200 and the total loss is $150 → profit factor = 1.33 **How to read it:** - Less than 1.0 — the strategy is losing money. - 1.0-1.3 — weak, on the edge of being noise. - 1.3-1.7 — normal. - Greater than 2.0 — excellent (if not overfitted). ### Metric 5. Average time in position How many hours/days the bot holds a position open on average. **Why look at this:** It shows whether the bot's actual behavior **matches its intended design**. If you designed a trend-following strategy on a 4-hour timeframe, but the average time in position is 30 minutes, something is wrong. The bot is exiting **earlier** than it should. ### What to write down each week A template for your weekly self-report: ``` Week N (dates) Total trades: ___ Profitable: ___ Unprofitable: ___ Win-rate: ___% RR: ___ Profit factor: ___ Max drawdown for the week: ___% Average time in position: ___ Largest profitable trade: $___ Largest unprofitable trade: $___ What seemed strange: ___ What I want to change (one thing): ___ ``` Write a report like this every Sunday. After 12 weeks, you'll have a **journal** more valuable than any course. --- ## What the Metrics **DON'T** Mean Typical illusions of beginner algorithmic traders: ### Illusion 1. "A high win-rate = a good strategy" As we saw above—no. You can have a 90% win-rate and lose money. ### Illusion 2. "If it's profitable for a month, the strategy works" One month is too short. Every strategy has **periods of luck and periods of pain**. If you draw conclusions after a month, you will **keep** randomly successful strategies and **discard** randomly unsuccessful ones. Both are paths to disaster. The minimum for evaluation is **3 months on a test account**. 6 is better. ### Illusion 3. "If the bot is better than my manual trading, it works" Most people trade **worse than random**. Comparing to "better than I do manually" is a weak standard. You should compare it to **buy-and-hold** on the same pair over the same period. If the bot **doesn't beat** simple buy-and-hold, it might be easier to just hold. ### Illusion 4. "Test account metrics = real account metrics" They will differ. On a real account, there's **slippage** (your order executes at a different price than you expected), sometimes **higher commissions**, and **liquidity** that isn't the same as in Testnet (where there's always an "infinite buyer"). Real results are usually **20-40% worse** than test results. ### Illusion 5. "Backtest = future profit" The most dangerous illusion. A backtest is a **check against the past**. The future does not repeat the past exactly. If a backtest shows an 80% annual return, you'll be lucky to get 20-30% in reality. Often, you'll get 0%. Sometimes, a loss. --- ## Walk-forward — A Concept Without Code The most correct technique for validating a strategy is **walk-forward validation**. You don't need to code it by hand—an AI will do it for you when needed. But you need to keep the **concept** in mind because it defines everything. The idea is simple. You have a **year** of historical data. You: 1. Take the first 6 months to tune your strategy's parameters (this is called the training set). 2. Run the strategy **with these parameters** on the next 1 month **that you haven't seen yet** (the test set). 3. See how it performs. 4. Slide the window forward by one month: 6 months for training, the next month for testing. 5. Repeat. This gives you a **series** of small, realistic tests on data the strategy didn't see during tuning. If the strategy consistently performs well on each test window, there's a chance it will work in the live future. If it only shows results on its "own" training windows, it's overfitted. **Why you need to know this:** When you see someone claim "200% annual return on my bot," ask: was that a **walk-forward** test or just a backtest on a single chunk of data? If it's just a backtest, it's a **picture**, not proof. When it's time to move to real money, you'll ask Claude / Cursor: "run my strategy through a walk-forward validation with N=12 windows of 1 month each." It will write the code. The important thing is **knowing what to ask for**. --- ## The Iteration Cycle After a week of running the bot, you'll feel the urge to **change** something. This is normal. The danger is changing **too much** at once. The correct cycle: ``` 1. I observe something strange. (Example: "The bot closes on the time-based stop too often, without hitting the take-profit.") 2. I formulate one hypothesis. (Example: "Perhaps a 24-hour time-based stop is too short for a 4-hour timeframe.") 3. I make one change. (Example: "I'm increasing the time-based stop to 48 hours, leaving everything else the same.") 4. I run it for the next week. (Minimum 50-100 trades.) 5. I check if the behavior has changed in the desired direction. - Yes → The hypothesis is confirmed. I can look for the next one. - No → I revert the change and formulate a different hypothesis. 6. I record what I tried and the outcome in my journal. ``` Six steps. One cycle = one week, minimum. Not five minutes. Not one day. **What NOT to do:** - Change three parameters at once—it's impossible to know what had an effect. - Change parameters **in the middle** of an observation cycle—it invalidates the observation. - Completely change the strategy after a series of losses—that's not iteration, it's panic. **Good iteration is slow.** If you've done **20 iterations** in 6 months, that's a lot. Most amateurs do 200 iterations in a month and wonder why nothing works. --- ## When to Take a Break This is a part almost no one talks about, and it kills more algorithmic trading careers than bad strategies. The bot will annoy you. That's a guarantee. In the third week, the sixth week, the fourth month—there will inevitably be periods when: - The logic is correct, but the bot loses money for ten days straight. - A change you made made things worse, not better. - You see an **obvious** signal on the chart, but the bot ignores it. - The strategy works, but painfully slowly. In these moments, the "I need to do something urgently" mode kicks in. If you give in, you'll lose the discipline of iteration you've been building for months. **The rule:** when you feel the urge to change the logic urgently, **do nothing for 24 hours**. Close your laptop. Go to the gym, take a walk, get some sleep. The next day, re-read your journal. In 80% of cases, the urge will pass. In the other 20%, it will transform into a **well-founded** hypothesis that needs to be tested through a proper iteration cycle. This isn't a slogan about "mindset." It's a **practical defense against yourself**. The biggest losses in algorithmic trading come not from bad strategies, but from **impulsive interventions in working ones**. --- ## The Decision: Continue or Quit After 6 months, you will have **enough data** to make one of four decisions. ### Option 1. Continue as is Signs: - The bot has been running for 6 months. - The win-rate is stable within your expected range (±10%). - The drawdown has not exceeded your stated limit. - You haven't gone crazy, aren't losing sleep, and aren't intervening manually. This is a **success**. You can start transitioning to small real money ($200-$500) and continue observing. ### Option 2. Deep strategy review Signs: - The strategy worked for the first 2-3 months, then stopped. - Metrics have deteriorated **across the board**, not just profit. - You see that the **market itself** has changed (it was ranging, now it's strongly trending—or vice versa). Don't **quit**, but **go back to the drawing board**—Day 3 of this course, from scratch. You might have been right about the archetype, but the specific implementation needs an update. ### Option 3. Change archetype Signs: - Over 6 months, the strategy has fundamentally **not worked**—there weren't even periods of profit. - The win-rate is constantly hovering below 30%. - No matter what parameters you change, it doesn't help. You may have chosen the **wrong** archetype for you. Go back to Day 2. Rewrite your profile (time, capital, personality). Perhaps a different approach suits you better. ### Option 4. Quit this niche Signs: - The bot has been losing money for 6 months. - You've lost **interest** in the process itself—it's no longer an interesting engineering challenge, just a chore. - You've realized that the time commitment is too much for you: you're not willing to spend 1-2 hours on this every day. This is **not a failure**. This is **self-knowledge**. Not everyone is suited for algorithmic trading. It's better to realize this after 6 months and move to another area of programming than to bang your head against the wall for another five years. All four options are **respectable** outcomes of the journey you've taken. The worst-case scenario is **making no decision** and continuing "somehow" for another year and a half. --- ## Today's Assignment This is the last assignment of the course. - [ ] **Recorded** all metrics for the past 4 days (win-rate, RR, max drawdown—even if there are few trades). - [ ] **Formulated one hypothesis** you want to test next week. - [ ] **Created a weekly report template** in a `journal.md` file next to `bot.py`. - [ ] **Made a decision** on what you'll do for the next 4 weeks—continue observing / change something / take a break. - [ ] **Read your three answers from Day 1 aloud to yourself** (about capital, time, and Plan B)—has your understanding changed after 5 days? Post a final message in the Telegram channel: ``` Day 5 — challenge complete! What I learned in these 5 days (one sentence): [your conclusion] The most unexpected discovery: [one thing] For the next 4 weeks, I plan to: [observe / make changes / take a break] ``` --- ## What's Next — No Pressure These five days were designed as a **self-contained** path. If you've done all the homework and have a bot running on a test account—**congratulations**, you've gone further than 90% of people who thought, "maybe I'll try algorithmic trading." From here, you **don't have to** buy anything. You can go your own way, find answers in open-source projects, read books, and experiment. This is a completely valid path. However, if you feel that **much was left uncovered in the five-day challenge**, we have a **full 10-module course** on nexus-bot.pro. What's covered there that **wasn't** in the challenge: - Multi-timeframe filtering—how to use 3-4 timeframes at once to improve signal quality. - Backtesting architecture with walk-forward—a working template for strategy validation. - Portfolio-level risk management—for when you have bots on several pairs simultaneously. - Infrastructure: systemd, monitoring, auto-restarts, Telegram alerts, logs that work. - Live deployment—a step-by-step transition from a test account to small real money. - 5 prompt templates for AI assistants for each component of the bot. - Access to a private Telegram community of students. Six tiers depending on the level of support you need: - **$199** — The full course, without the community. - **$299** — Course + private Telegram community. - **$399** — Course + community + ready-made backtest templates. - **$499** — Course + community + templates + a code review of one of your bots by me. - **$599** — Course + community + templates + code review + a one-on-one hour-long meeting on Zoom. - **Custom** — Three-month mentorship, discussed personally. **For those who completed the challenge**—a 30% discount on any tier for 48 hours from the moment you send your final message in the Telegram channel. Use the promo code I'll send you in a private message (only for those who check in on the Day 5 chat). And one more thing—if you're **not ready** to buy, but you still have questions, you can continue to hang out in the general Telegram channel for the challenge. I'm not closing it. I'll be there periodically answering questions from new participants, and old ones can read along too. --- ## Thank You You've gone from "I don't know where to start" to "I have my own working bot, I understand every decision, I know how to read metrics and how to iterate with discipline." This is **more** than 95% of the people who call themselves algorithmic traders have. Whatever you decide to do next—continue with me or go your own way—**thank you for spending these five days with me**. And remember: one of the most valuable engineering skills is the **ability to say "stop" when something isn't working** and try something else. Algorithmic trading is not a field where you must "never give up." It's a field where you must **calculate soberly** and **decide with discipline**. Good luck. --- # Homework — Day 5 (Final) > Today is less about "doing" and more about "reading your data and making decisions." ## Checklist - [ ] **Metrics recorded** for the past 4 days (even if there are very few trades, write down what you have) - [ ] **One hypothesis formulated** for the next week of observation - [ ] **Weekly report template** created in a `journal.md` file next to `bot.py` - [ ] **A decision has been made** on what to do for the next 4 weeks: observe / change something / take a break - [ ] **Read your three answers from Day 1 aloud** (capital / time / plan B) — do you notice any change in your understanding? --- ## Template for metrics from the past 4 days > Fill this out even if you have few trades. The main thing is to see the **format** you'll use for future records. ``` PERIOD: from ___ to ___ TRADE STATISTICS Total trades: ___ Profitable: ___ (___%) Unprofitable: ___ (___%) KEY METRICS Win-rate: ___% (good: 50-65%, suspicious: >75%) Risk-Reward: ___ (good: 1.5-2.5) Profit factor: ___ (good: >1.3) Max drawdown: ___% (concerning: >30%) Average time in position: ___ hours EXTREMES Largest profit: $___ Largest loss: $___ ``` --- ## Template for a hypothesis for the next week ``` OBSERVATION What I noticed that was strange about the bot's performance: ________________________________________________ HYPOTHESIS Possible cause: ________________________________________________ CHANGE What I am changing (one specific thing): ________________________________________________ SUCCESS CRITERION I will consider the hypothesis confirmed if: ________________________________________________ WHEN TO CHECK After how many trades (not days): ___ (minimum 50) ``` --- ## Weekly report template (for `journal.md`) ``` # Bot Performance Journal ## Week 1 — dates ___ Total trades: ___ Profitable: ___ / Unprofitable: ___ Win-rate: ___% RR: ___ Max drawdown: ___% What seemed strange: ___ What I want to change (one thing): ___ ## Week 2 — dates ___ [repeat template] ... ``` This will be your main document for the coming months. A new entry every Sunday. --- ## Decision for the next 4 weeks Choose one of the four options and **write down** in your journal why you chose it: ### A. Continue as is Signs: The bot is working, metrics are stable, you are calm. **Don't change anything**, just observe for another month. ### B. Make one change based on a hypothesis Signs: You see a specific anomaly, and you have a hypothesis. Make **one** change and observe for the next week. ### C. Take a one-week break Signs: You're frustrated with the bot and feel an urgent need to change something. **Close your laptop**. Reread your journal in a week — the urge will likely have passed. ### D. Deep review Signs: The bot is **fundamentally** not working after 5 days (no trades / only losses / unpredictable behavior). Go back to Day 2-3 — re-evaluate your choice of archetype and its formulation. --- ## Self-check from Day 1 Open your answers to the three honest questions you wrote down on the first day: 1. How much capital are you truly prepared to lose? 2. How many hours a week can you realistically commit? 3. What will you do in 6 months if the bot isn't working? **Reread them.** Has anything changed in your understanding after these five days? If **yes** — update your answers. This is normal and good. Real experience changes assessments. If **no** — excellent. It means you were thinking clearly from the start. --- ## Final message in Telegram ``` Day 5 — challenge complete! What I understood in these 5 days (one sentence): [your conclusion] The most unexpected discovery: [one thing] For the next 4 weeks, I plan to: [observe / change / take a break / rethink] ``` After this message, I will send you a private message with a **promo code for a 30% discount** on the full nexus-bot.pro course. It's valid for 48 hours. No pressure — if now isn't the right time, you can use it later when you've decided. --- ## What I want to say in parting You've gone further than 95% of people who have ever started the journey of algorithmic trading. Most drop out on the second or third day when they realize it's not about getting rich quick. Whatever you decide to do next — whether you continue with me in the course or go your own way — **thank you for spending these five days with me**. Good luck.