TL;DR
Start with a complete entry, exit and size, then keep every variant and test the selected rules unchanged on later dates. In the recorded NVDA experiment, the genetic winner returned 22.06% versus NVDA’s 36.00% on April through June 2024. It bought once, stayed invested overnight and failed four of five requirements. A profitable result therefore missed the intended intraday strategy. The published dates are reproduction material, so reserve genuinely unseen dates for your own research.
Follow the recorded NVDA experiment
Start with the actual morning rules
NVDA enters from 30 to fewer than 60 minutes after open, above SMA20, with an empty holding. A separate midday exit sells the full position. Read the complete draft before searching.
Two rules with explicit entry, exit and sizing. Reviewing this draft does not save it or submit a test.
Set the search before seeing its winner
The completed Parameter genetic search used six candidates per generation and three generations. It changed timing, comparisons, sizing and exits. It was broader than a three-value moving-average sweep.
- Development dates
- Jan 2 to Mar 28, 2024
- Population / generations
- 6 / 3
- Training / windows
- 75% / 1
- Final leaderboard
- 6 candidates; unique-trial count unavailable
A saved search plan, every final candidate and an inspection of what changed in the chosen rules.
- Genetic winner 2024-06-28, 20:00:00 EDT
- 12,205.98
- Original morning strategy 2024-06-28, 20:00:00 EDT
- 8,980.94
- NVDA buy and hold 2024-06-28, 20:00:00 EDT
- 13,600.01
- SPY buy and hold 2024-06-28, 20:00:00 EDT
- 10,371.61
Read the later result and the failed goal
On April 1 to June 28, the winner bought once, never sold and stayed invested across sessions. Its return was positive, but four of five requirements failed. The original morning rules also weakened on these later dates.
- Genetic winner / NVDA
- 22.06% / 36.00%
- Original later return
- -10.19%
- Winner drawdown / limit
- 12.73% / 10%
- Closed round trips / minimum
- 0 / 10
- Winner fees paid
- $2.98
- Acceptance
- 1 passed; 4 failed
Keep the failed result. Inspect the full selected exit before calling this an intraday strategy.
Open the fully configured request
Aurora receives the complete seed, search settings, costs, later dates and acceptance rules. Review and edit the request. Ask for a plan and cost before deciding whether to run your own reproduction.
- Capital / observations
- $10,000 / Minute
- Shares / dividends
- Fractional / cash
- Stock fee per fill
- 0.05%
- Current task
- Review and plan; do not launch
A review of the supplied rules and a research plan. Engine validation requires a later approved authoring step; this request stops before saving or starting research.
This completed genetic experiment is intentionally selected to teach the workflow. Its profitable winner missed the original intraday requirements. Published historical dates are available for reproduction; they are not an untouched test for a new reader.
Start with one rule you can explain
The recorded NVDA seed asks a small question: does a morning move above its recent average justify holding until midday? It buys only when no NVDA is held, uses 95% of buying power, and has a separate full-position exit. Read those two rules before introducing a search.
Choose your next step: inspect the original rule, understand what the optimizer changed, or plan a later test. The completed optimization below is a worked research experiment; you do not need to start there.
Start with the question you want to test
Write one hypothesis, a tradable universe and an explicit entry, exit and size rule. Choose a benchmark, costs and observation interval. A trend idea, a reported-company filter and a disclosure-following idea need different evidence, even when they trade the same stock.
Ask Aurora to inspect the backtest corpus first
Ask Aurora to inspect the backtest corpus first
Start with existing experiments. Compare the whole rule and its test inputs, including weaker examples, before choosing a signal. Save the dates you inspected in your research notes so you can reserve a later test.
Before creating a trading strategy, inspect the existing backtest corpus. Search for completed prior examples using these indicator types: Price, SimpleMovingAverage, Fundamental, with comparable tickers, cadence and asset class. Examine both stronger and weaker completed examples and their actual rules, dates, capital, fees, turnover and benchmark. Screen the corpus for a more specific cohort or named strategy when needed. Show the returned records and any filters you widened. Explain which ideas recur and which trades account for the results. Record the historical periods you inspected. Propose three bounded variants and a development period; keep a later unseen test outside the selection. Stop at the research plan.Choose the periods before comparing variants
Show the table · 4 rows, 3 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Development | 2022-01-01 through 2023-12-31 | Create variants, inspect trades and revise rules here. |
| Unseen test | 2024-01-01 through 2024-12-31 | Evaluate the frozen candidate once after choosing the objective. |
| Optional final lockbox | 2025-01-01 through 2025-12-31 | Reserve an additional untouched window before all selection. |
| Paper observation | Later dates after the design is frozen | Observe signals, scheduling, orders and positions as they occur. |
No matching rows. Clear the filter to see all records.
This calendar illustrates a research plan. Use dates suited to your data and record any periods already inspected through corpus research, charts or earlier experiments.
Create variants and keep a trial ledger
Begin with a small set of changes that test the hypothesis. Record the exact rule, dates, starting capital, fees, benchmark, result and decision for every attempt. Use development trades to explain what changed. Once you have selected a candidate, freeze it before opening the unseen period.
Show the table · 7 rows, 2 columns
| Stage | Deliverable |
|---|---|
| Corpus research | Comparable records, rules and dates already inspected |
| Variants | A bounded list with a reason for each change |
| Development | Saved results and a choice made using a preselected objective |
| Unseen evaluation | The unchanged selected rule and its later-period result |
| Optional lockbox | An additional untouched assessment of the final selection |
| Paper trading | A dated observation log of signals, orders and positions |
| Live trading | A selected broker, reviewed allocation and explicit automation setting |
No matching rows. Clear the filter to see all records.
Give Aurora the whole research sequence
First inspect comparable completed records from the backtest corpus. Show rules, dates, fees, benchmark, stronger and weaker examples, and periods already exposed. Propose three variants for my hypothesis. Agree one development period and a later unseen test before launching anything. Iterate only on development dates and keep a trial ledger. Freeze the selected configuration and objective, then test the unseen period unchanged. Reserve an optional final lockbox if requested. Prepare paper observation steps and a separate live deployment review; show the exact target and automation settings.Explore completed examples
Use the options and crypto examples to practice reading a full curve, a signal and its recorded orders. Crypto pages include SPY and BTC buy-and-hold lines over the same period.
No comments yet.