A test result is a claim about a particular input and an expected answer. Our software tests can check that the engine keeps two capital bases apart, recognizes changing position sizes, and withholds an output when the history is too thin. Those checks matter because a plausible number can still answer the wrong question.
One constructed survival example uses a run capital of 500 and an evaluation deposit of 250. The test expects the engine to retain both values. On that fixture, 40% of its modeled starting points exhaust the smaller deposit. Evaluating the same constructed paths at 500 changes that share to 20%.
This checks whether the calculation follows the declared evaluation capital. Neither percentage estimates the risk of an EA. The paths were written to exercise the calculation, and their expected answers were worked out in advance. Reproducing those answers does not establish how often a similar path will occur in a market.
Our position-size drift tests construct a history with a stable base lot and larger progression trades. That example must return consistent. Another example raises the base lot across years and must return contradicted. A history with only one qualifying year must return insufficient.
The last result has a separate consequence. An integration test sends a small tester-report fixture through the audit with a declared fixed base lot policy. The survival calculation must stay gated when the drift check lacks enough evidence. Declaring a policy does not itself supply the history needed to test it.
These tests establish behavior for the supplied examples. They do not establish that every future export has a familiar layout, that a vendor will keep the same trading logic, or that a simulated trade sequence will repeat live. They also cannot demonstrate that a missing test would pass. The useful question is which failure an example was designed to expose.
When reading an audit, keep the declared capital basis and the reason for any withheld section beside the result. A passing software check helps explain why the engine produced that result. It does not enlarge the market evidence contained in the original run.