1
00:00:00,104 --> 00:00:02,956
A student who memorizes an exam may score well

2
00:00:03,177 --> 00:00:06,654
without solving a new problem Selecting robot settings from

3
00:00:06,667 --> 00:00:10,339
the best historical result presents a similar danger We

4
00:00:10,352 --> 00:00:14,948
separate development data from evaluation data Tables are hypothetical

5
00:00:14,961 --> 00:00:18,945
and photographs provide context A profit curve alone cannot

6
00:00:18,958 --> 00:00:22,904
establish behavior under new conditions The central question is

7
00:00:22,917 --> 00:00:26,484
whether the evaluation was truly separated from the decisions

8
00:00:26,497 --> 00:00:30,221
used to select the configuration rather than merely given

9
00:00:30,234 --> 00:00:31,302
an impressive label

10
00:00:32,458 --> 00:00:35,857
The example divides twelve months into eight for development

11
00:00:36,247 --> 00:00:39,919
and four for evaluation This ratio is not universally

12
00:00:39,932 --> 00:00:44,112
appropriate Record the boundary and rationale before inspecting

13
00:00:44,125 --> 00:00:48,643
results Define indicator warm-up and treatment of positions crossing

14
00:00:48,656 --> 00:00:53,161
the boundary Repeatedly consulting evaluation data to tune settings

15
00:00:53,435 --> 00:00:56,560
means it is no longer an untouched final test

16
00:00:56,989 --> 00:01:00,453
The label must reflect actual data use Preserve the

17
00:01:00,466 --> 00:01:04,581
dates assumptions and every revision so the testing history

18
00:01:04,594 --> 00:01:05,375
can be reviewed

19
00:01:06,516 --> 00:01:09,537
Consider A with profit eight and drawdown nine B

20
00:01:09,550 --> 00:01:12,779
with profit six and drawdown three and C with

21
00:01:12,792 --> 00:01:16,464
profit five and drawdown two The prewritten rule rejects

22
00:01:16,477 --> 00:01:20,501
drawdown above four then chooses the highest remaining profit

23
00:01:20,930 --> 00:01:24,381
Reject A and select B Changing the rule after

24
00:01:24,394 --> 00:01:26,907
viewing the table to favor A creates a different

25
00:01:26,920 --> 00:01:30,839
experiment The report should include the criterion and rejected

26
00:01:30,852 --> 00:01:33,925
outcomes not just the number that makes the selected

27
00:01:33,938 --> 00:01:35,123
version look attractive

28
00:01:36,254 --> 00:01:39,913
Evaluate the chosen version with its saved settings If

29
00:01:39,926 --> 00:01:43,520
results disappoint do not quietly replace it with whichever

30
00:01:43,533 --> 00:01:46,957
alternative looks better on that same segment and publish

31
00:01:47,074 --> 00:01:50,929
only that outcome Further research is possible but acknowledge

32
00:01:50,942 --> 00:01:54,913
that this data informed selection Design the next evaluation

33
00:01:54,926 --> 00:02:00,004
separately An unfavorable result is still decision evidence Deleting

34
00:02:00,017 --> 00:02:02,817
it improves the appearance of the report not the

35
00:02:02,830 --> 00:02:06,241
reliability of the program The trial history is part

36
00:02:06,254 --> 00:02:07,061
of the evidence

37
00:02:08,278 --> 00:02:10,922
Forward testing in the MetaTrader tester can mean a

38
00:02:10,935 --> 00:02:15,440
later historical segment A demo-forward run collects new observations

39
00:02:15,453 --> 00:02:18,395
from now onward These are different and answer different

40
00:02:18,408 --> 00:02:23,903
questions Document data costs and order behavior Out-of-sample success

41
00:02:24,020 --> 00:02:28,343
does not validate reconnection controls or guarantee future returns

42
00:02:28,773 --> 00:02:32,810
Validation requires multiple kinds of evidence If you compare

43
00:02:32,823 --> 00:02:36,494
several forward results and select the best disclose that

44
00:02:36,507 --> 00:02:39,997
additional selection rather than presenting it as an untouched

45
00:02:40,010 --> 00:02:41,104
final evaluation

46
00:02:42,271 --> 00:02:45,644
Hundreds of trials give historical noise more chances to

47
00:02:45,657 --> 00:02:50,331
look impressive Record trial counts manual changes and criteria

48
00:02:50,748 --> 00:02:53,716
A neighborhood of settings with similar behavior may be

49
00:02:53,730 --> 00:02:57,115
more informative than one exceptional point but is not

50
00:02:57,128 --> 00:03:01,711
proof Controlled changes and ablation tests reveal which decisions

51
00:03:01,724 --> 00:03:05,709
drive results The goal is to reduce unknowns not

52
00:03:05,722 --> 00:03:09,667
manufacture a guarantee Fewer parameters alone are not enough

53
00:03:09,680 --> 00:03:13,157
either selection procedure and data use remain important

54
00:03:14,300 --> 00:03:17,438
Solve the table with maximum drawdown four followed by

55
00:03:17,451 --> 00:03:21,097
highest profit the answer is B Add an evaluation

56
00:03:21,110 --> 00:03:25,329
column and retain negative results The worksheet records boundary

57
00:03:25,524 --> 00:03:30,108
criterion version and data reuse Explain whether a failure

58
00:03:30,120 --> 00:03:33,988
calls for further evidence a specific revision or stopping

59
00:03:34,417 --> 00:03:38,141
Next we examine demo safety duplicate orders and lost

60
00:03:38,154 --> 00:03:42,282
communication This assignment teaches a review process it does

61
00:03:42,295 --> 00:03:45,772
not predict that any selected configuration will make money