Lagias is one of the six products below, and it is ours. It is also not available yet, so every claim about it describes how it is built rather than a result anyone outside the team can check. It is listed last, and it fails the availability row that all five others pass. If that is disqualifying, the other five are scored the same way and the rubric is published in full.
Why there is no number one
Because we could not produce one honestly. Across these six products, five of the eight rubric answers come back as “not stated” — their public pages simply do not say. Averaging a set with that many unknowns into a single score manufactures precision that does not exist, and the result would be decided by how we chose to weight the gaps.
The other reason is that these products are not doing the same job. Ranking a research platform above a charting tool says more about which one we happened to be thinking of than about either of them. So they are grouped by what they are for, and the grid is published so you can weight it yourself.
The eight questions
Each one is a checkable property of a system rather than a judgement about quality. Stated before the table, so the weighting is yours rather than ours.
Out-of-sample data held back by the system
Validation the user has to remember to run is validation that gets skipped exactly when a result already looks good.
Counts the strategy variants tried
The number of attempts behind a result is the single largest input to whether it is real. It can only be reported by a system that kept them.
Reports selection-corrected performance
A deflated Sharpe ratio, or equivalent. Without it, a raw figure is presented as a measurement when it is the winner of a search.
Reports probability of backtest overfitting
PBO is a property of the whole search rather than of one strategy, so reporting it requires the trials matrix to exist.
Walk-forward analysis by default
Offered as an option is not the same as applied to everything. The distinction only matters when you would rather not run it.
Cost model enforced, not user-set
A slippage assumption you can lower until the strategy passes is a preference, not an assumption.
Grading separated from proposing
An optimizer told to maximize a backtest result will maximize the backtest result. Whether the grader shares that objective is architecture, and architecture is checkable.
Available to use today
Included because it is where Lagias currently loses, and a rubric that omits the author’s weakness is not a rubric.
QuantConnect
A full quant research platform: point-in-time data, a research environment, an optimizer, an MCP server and a route to live trading.
Best for: People who write strategies in code and need serious data and execution infrastructure behind them.
| What is checked | Verdict | What was observed |
|---|---|---|
| Out-of-sample data held back by the system | Not stated | Out-of-sample handling is left to the algorithm author; nothing read describes the platform holding data back for you. |
| Counts the strategy variants tried | Not stated | Optimization runs are stored and listable, so the raw material exists, but nothing read reports a count alongside a result. |
| Reports selection-corrected performance | Not stated | Not named on the homepage or in the MCP server README, which documents an includeStatistics flag without listing the statistics. |
| Reports probability of backtest overfitting | Not stated | Not named anywhere in the documentation read, and no equivalent overfitting probability is described. |
| Walk-forward analysis by default | Not stated | Not named on the pages read. The marketed workflow is parameter sensitivity testing across thousands of backtests. |
| Cost model enforced, not user-set | Yes | Their own words: "point-in-time, fee, slippage, and spread-adjusted backtesting", with realistic margin modelling. Best documented in this set. |
| Grading separated from proposing | No | You write the strategy and the same platform runs and scores it, with the optimizer targeting whatever objective you set. |
| Available to use today | Yes | Shipping for years, with a free backtesting tier and a large public user base. |
Read from www.quantconnect.com on · Full comparison
StrategyQuant X
A no-code strategy generator — "create your own algo-strategies without programming" — with a named battery of robustness tests.
Best for: People who want strategies generated for them and want the standard robustness battery applied without writing it themselves.
| What is checked | Verdict | What was observed |
|---|---|---|
| Out-of-sample data held back by the system | Partly | Walk-forward optimization and a walk-forward matrix are named features, which implies out-of-sample segments, but they are tools you choose to run. |
| Counts the strategy variants tried | Not stated | The product generates large numbers of candidates; nothing read describes the count being reported alongside a surviving strategy. |
| Reports selection-corrected performance | Not stated | Not named on the pages read, and no selection-corrected performance figure is described. |
| Reports probability of backtest overfitting | Not stated | Not named on the pages read, and no equivalent overfitting probability is described. |
| Walk-forward analysis by default | Partly | Named explicitly alongside Monte Carlo simulations and System Parameter Permutations. Offered as a tool rather than described as applied to everything. |
| Cost model enforced, not user-set | Not stated | No transaction cost or slippage modelling is described on the page read. |
| Grading separated from proposing | No | The same product generates the strategies and runs the tests on them, which is the conventional design for a strategy generator. |
| Available to use today | Yes | Shipping, with an established user base in the retail algo segment. |
Read from strategyquant.com on
Build Alpha
"Create Thousands of Algorithmic Trading Strategies at the Click of a Button", with robustness tests described as professional and proprietary.
Best for: People who want strategy generation at volume and are comfortable with the validation being described rather than specified.
| What is checked | Verdict | What was observed |
|---|---|---|
| Out-of-sample data held back by the system | Not stated | Not described on the homepage, which refers to validation and robustness without naming the procedures. |
| Counts the strategy variants tried | Not stated | The headline offer is creating thousands of strategies; nothing read describes that count feeding into how a survivor should be read. |
| Reports selection-corrected performance | Not stated | Not named on the homepage, and no selection-corrected performance figure is described. |
| Reports probability of backtest overfitting | Not stated | Not named on the homepage, and no equivalent overfitting probability is described. |
| Walk-forward analysis by default | Not stated | Not named on the homepage. Robustness testing is called "proprietary", so the methods are not public. |
| Cost model enforced, not user-set | Not stated | No transaction cost or slippage modelling is described on the homepage read. |
| Grading separated from proposing | No | One product generates the strategies and applies its own robustness tests to them. |
| Available to use today | Yes | Shipping, sold as a one-time licence, established in the retail algo segment. |
Read from www.buildalpha.com on
LuxAlgo
A charting and indicator platform with an AI agent that "builds limitless strategies", backtests them and "optimizes live on the chart".
Best for: People who work primarily on charts and want indicators and plain-language strategy building in the same place.
| What is checked | Verdict | What was observed |
|---|---|---|
| Out-of-sample data held back by the system | Not stated | Not described on the homepage; the marketed workflow is testing against years of history in seconds. |
| Counts the strategy variants tried | Not stated | The agent is described as building limitless strategies; nothing read reports how many were tried. |
| Reports selection-corrected performance | Not stated | Not named on the pages read, and no selection-corrected performance figure appears in the copy. |
| Reports probability of backtest overfitting | Not stated | Not named on the pages read, and no equivalent overfitting probability is described. |
| Walk-forward analysis by default | Not stated | Not named. The homepage does not use the words overfitting, validation or walk-forward at all. |
| Cost model enforced, not user-set | Not stated | No cost, slippage or spread modelling is described on the homepage read. |
| Grading separated from proposing | No | Their own description has the agent building, backtesting and optimizing in one loop before running what you approve. |
| Available to use today | Yes | Shipping, with free and paid tiers and one of the largest user bases in the category. |
Read from www.luxalgo.com on · Full comparison
The Honest Quant
A no-code lab that assembles strategies and returns a statistical "honesty score" from five named tests, rather than a profit estimate.
Best for: People who want a verdict on backtest quality today, from a product that already ships and already counts parameter changes.
| What is checked | Verdict | What was observed |
|---|---|---|
| Out-of-sample data held back by the system | Yes | Out-of-sample decay is one of five named tests feeding the honesty score, applied as part of the product rather than as an option. |
| Counts the strategy variants tried | Yes | Stated directly: "Each parameter change you try is recorded", with an example verdict reading "50 parameter variations tried". |
| Reports selection-corrected performance | Not stated | Not named on the page read; the trial count feeds a data-snooping judgement rather than a stated selection-corrected figure. |
| Reports probability of backtest overfitting | Not stated | Not named on the page read, and no equivalent probability is described. |
| Walk-forward analysis by default | Not stated | Not named on the page read, though out-of-sample decay and Monte Carlo luck testing are. |
| Cost model enforced, not user-set | Yes | Real-cost survival is one of the five named tests, so costs are part of the scoring rather than a user-set field. |
| Grading separated from proposing | Partly | You assemble the strategy in their lab and the same platform scores it, but the scoring is a fixed set of named tests rather than an optimizer targeting the result. |
| Available to use today | Yes | Shipping, and the verdict is stated to be free. |
Read from thehonestquant.com on
LagiasThis is us
This site. A validation harness that evolves plain-English ideas into candidates and grades them with a pipeline the proposing model cannot influence.
Best for: Nobody yet — it is not available. When it is: people whose question is whether an idea holds up, rather than how to build one.
| What is checked | Verdict | What was observed |
|---|---|---|
| Out-of-sample data held back by the system | Yes | The evaluation pipeline runs on data the proposing model never saw. Enforced by the system rather than chosen per run. |
| Counts the strategy variants tried | Yes | The harness generates the candidates, so the count is a by-product of the search rather than something anybody reconstructs afterwards. |
| Reports selection-corrected performance | Yes | A deflated Sharpe ratio is reported alongside the raw one, built from that trial count. |
| Reports probability of backtest overfitting | Yes | Probability of backtest overfitting is one of four checks in the evaluation pipeline. |
| Walk-forward analysis by default | Yes | Applied to every candidate. There is no setting that turns it off, because a setting is something you switch off when the first result looks good. |
| Cost model enforced, not user-set | Yes | Costs are applied by the pipeline on identical terms for every candidate, so no strategy can win by having been given kinder assumptions. |
| Grading separated from proposing | Yes | The evaluator is deterministic code and runs separately from the model that proposed the strategy. Nothing downstream retries until it passes. |
| Available to use today | No | Waitlist only, as of July 2026. Everything above describes how the system is built, not a result anyone outside the team can check yet. This is the row where every product on this page beats us. |
Read from lagias.com on
What is not on this list, and why
A roundup that silently omits products implies it considered them. These were not read directly, so they are not scored.
- TradingView Strategy Tester
- Almost certainly the most-used backtester in retail. Its documentation pages returned 404 to our reader, and we will not score a product from memory.
- AmiBroker
- The long-standing retail walk-forward standard. Not yet read directly.
- Composer, TradeTest.ai, Nexera, TrendSpider, Backtrex
- Adjacent products in the category. Not yet read directly, so not scored.
- Backtrader, VectorBT, Zipline and other open-source libraries
- A different shape of thing: they give you the machinery and every rubric answer becomes "whatever you implement". Scoring them against a product rubric would be misleading in both directions.
How to read “not stated”
It means we read the product’s public pages and the answer was not there. It does not mean the product cannot do the thing. Several of these tools may implement checks they simply do not market, and a user can implement most of them by hand on any platform that runs code.
What the pattern does show is what the category considers worth advertising. Across six products, transaction cost modelling is named twice, walk-forward once, trial counting once, and selection-corrected performance never. Those are the checks that decide whether a backtest means anything, and they are largely absent from how these tools describe themselves.
Backtested and hypothetical results are not indicative of future performance. Nothing here is investment advice. This page compares what systems measure and report, which is verifiable from public documentation — not what anybody earns, which is not.