From 79ec0d678df67ce24e7c9c59e051a213dcb066ed Mon Sep 17 00:00:00 2001 From: =?utf8?q?J=C3=A9r=C3=B4me=20Benoit?= Date: Sat, 31 Oct 2026 19:56:52 +0100 Subject: [PATCH] docs: clarify native RL chart training limits --- docs/evaluation.md | 3 +++ 1 file changed, 3 insertions(+) diff --git a/docs/evaluation.md b/docs/evaluation.md index 0b9f353..31b08fe 100644 --- a/docs/evaluation.md +++ b/docs/evaluation.md @@ -50,6 +50,9 @@ when the extracted source loads, but the current server still supplies the FreqAI model and runtime configuration; this does not reproduce a saved policy or freeze the historical environment. +For ReforceXY, this native chart API lacks the pairlist context required to train +a fresh policy. Use normal backtesting for policy training and evaluation. + These checks run in `QuickAdapterRegressorV3` and `ReforceXY`, not in other FreqAI models. Reserve `ft_bt_strategy*` directories containing `_.py` for native archive extraction; do not configure that -- 2.53.0