Every strategy in AlgoLift lives in a project with a goal. The project puts it through stages that each add a kind of evidence the one before could not, from one backtest on history to days that did not exist when you built it. The Scorecard weighs that evidence and answers your goal's question.
One backtest is one history
A backtest replays your rules over past market data and reports what would have happened. That is useful, and it is also a single path. The trades arrived in one order, the market had one character, and you probably tried several versions before this one. Any of those can make luck look like an edge.
So AlgoLift never stops at the first result. Each stage asks the question the backtest cannot answer about itself.
A project and its goal
You start by creating a project and choosing its goal. The goal decides which question the project answers and which number its verdict leads with. A prop project and a Standard project run the same stages; what they ask of the evidence differs.
| Goal | The question it answers | Who it is for |
|---|---|---|
| Pass an evaluation | Will I pass the eval? | You're not funded yet. Optimize your strategy for the highest chance of passing. |
| Maximize withdrawals | How much can I actually pull out? | You're already funded. Pull out as much as you can without losing the account. |
| Full prop lifecycle | What does the whole grind net me? | One strategy, every stage: eval, funded, payouts, resets. What a year of it nets. |
| Standard strategy | Is this edge real? | No prop firm. Build it, validate it, prove the edge is real. |
The stages
A project's page has a rail of stages across the top. Overview and Journal are the project's own pages. The stages after them run from the weakest proof to the strongest.
| Stage | What it is for |
|---|---|
| Overview | Project scorecard, what each stage has proven so far, and a table of every result. |
| Journal | The project's written memory. Capture your thesis, ideas, and what you learned here. The copilot will add to it as you go. |
| Build | Assemble your strategy and run it against historical data. |
| Optimize | Tune the settings, and test whether the tuning holds on data it never saw. Walk-forward, a sweep, or keep your settings. |
| Forward Test | Prove the edge on data that did not exist when you built it. The reserved holdout first, then new bars as they arrive each night. |
| Live Trading | Run your champion from your own computer with AlgoLift Desk. Unlocks after launch. |
What happens after every backtest
When a backtest finishes, two more analyses start without you asking:
- Robustness re-deals the run's trades thousands of ways, to measure how much of the result depended on their order and on chance.
- The prop simulation runs the same trades through the rules of every prop firm account AlgoLift models, so a prop project can read its pass rate from the run you just made.
Their results land in the run's tabs a little after the backtest's own.
Tuning, graded on months it never saw
Optimize searches for better settings, and a setting picked from many tries is suspect until it holds somewhere the search did not look. Walk-forward tunes on one stretch of months and tests the pick on the next, again and again. A sweep is scored for how much its search overfit.
The holdout, then new days
When you create a project, AlgoLift sets its most recent months aside as the holdout. A run whose dates reach into those months is refused unless you unlock them, and the project permanently records any run that used them.
In Forward Test you spend the holdout once. After that, the strategy keeps running on new days as they arrive each night, judged against the range it predicted before any of them happened.
The Scorecard
The project's Overview opens on the Scorecard. It reads each stage's verdict, weighs them by how strong their evidence is, and answers your goal's question with a tier: Robust, Watch or Fragile.
The weights follow the evidence ladder. A single backtest is the anchor and counts for nothing on its own. Results on new days count the most.
Some findings cap the tier however good the headline looks: a run that used the holdout, a pick from a sweep with a high risk of selection bias, a forward test that broke the range it predicted, or evidence your goal needs that does not exist yet. Until a strategy has results on data it never saw, the project cannot read Robust. The Scorecard names the finding that caps it, so you know what would lift it.
The copilot
The copilot reads the same evidence the Scorecard does. It can build and edit a strategy, explain a result, and propose the next test worth running. Anything that would spend credits waits for your approval on a card. The copilot can frame a finding in its own words, but it can never change the numbers.
What it means for your results
- Read a backtest as a starting point. Its equity curve is the most flattering view of the strategy you will get.
- Expect the verdict to move. Each new kind of evidence can strengthen it or overturn it, and a forward test can overturn everything before it.
- Keep the holdout clean. It is the one test you cannot run twice on fresh data.