====== Data Horizon ======

Data Horizon controls which part of the loaded market data is used for Generator calculations.

Use this section to:

  * limit the history window by dates
  * limit bars count
  * define when trading starts inside the selected window

This is one of the key sections for building an in-sample and out-of-sample workflow.

<code ini>

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Data Horizon             ;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

max_data_bars  = 100000

; Use these parameters to set the data range between "data_start" and "data_end".
; Set from where the actual trading starts with "trade_start".
; Text: "2023-05-25 00:00 UTC", "2023-05-25 00:00", "2023-05-25" or "25 May 2023 00:00 UTC"
; Negative number: number of days before the current date: -365, -30, -7

use_data_start  = false
data_start      = "01 Jan 2020 00:00 UTC"

use_data_end    = false
data_end        = "30 Jun 2023 00:00 UTC"

use_trade_start = false
trade_start     = "01 Jun 2023 00:00 UTC"

</code>

===== What each option does =====

  * ''max_data_bars'' limits the maximum number of bars loaded after date filters are applied.
  * ''use_data_start'' + ''data_start'' cut bars before the start date.
  * ''use_data_end'' + ''data_end'' cut bars after the end date.
  * ''use_trade_start'' + ''trade_start'' keep data for indicators, but start opening trades later.

===== How ''data_end'' limits data =====

When ''use_data_end = true'', Generator stops the dataset at ''data_end''.
Bars after that timestamp are excluded from calculations in that run.

Practical effect:

  * your strategies are generated or validated only on the capped history window
  * newer bars are not seen in that run

===== OOS workflow with recalculation on full data =====

You can use ''data_end'' to create a clear split between historical training data and newer unseen data.

Typical 2-step process:

  * Step 1 (in-sample build): run Generator with ''use_data_end = true'' to cap history at an older date.
  * Step 2 (recalculation/validation): run Generator again on the full dataset (for example ''use_data_end = false'').

In Step 2, the bars that were excluded by Step 1 become your practical out-of-sample segment.
This lets you check how strategies behave on data that was not included during the capped run.

===== Example profile =====

Build collection only up to 30 Jun 2023:

<code ini>
use_data_end = true
data_end     = "30 Jun 2023 00:00 UTC"
</code>

Then validate the same collection on full history:

<code ini>
use_data_end = false
</code>

===== Notes =====

  * Data Horizon is applied before Out of Sample percentage filters.
  * For additional control of unseen-data validation, see [[forward-testing|Forward testing]].
  * Percent-based splitting is documented in [[out-of-sample|Out of sample]].

===== Related pages =====

  * [[forward-testing|Forward Testing]]
  * [[out-of-sample|Out of sample]]