diff --git a/dsl/dsl.R b/dsl/dsl.R
index b7400c6..4943e68 100644
--- a/dsl/dsl.R
+++ b/dsl/dsl.R
@@ -4,6 +4,10 @@ library(dsl)
 dsl_csv <-"111725_DSL_frame.csv"
 dsl_df <- read.csv(dsl_csv, header = TRUE) 
 
+dsl_df <- dsl_df |>
+  dplyr::mutate(ttr_days = TTR / 24) |>
+  dplyr::mutate(task_resolution = dsl_score)
+
 base_model <- dsl(
   model = "logit", 
   formula = dsl_score ~ human_EP_prop_adac,
@@ -27,7 +31,8 @@ summary(case_model)
 logit_model <- dsl(
   model = "logit", 
   formula = dsl_score ~ human_EP_prop_adac + human_TSOL_prop_adac + human_RK_prop_adac 
-     + week_index + as.factor(isAuthorWMF) + median_PC4_adac + n_comments_before + as.factor(source),
+     + week_index + as.factor(isAuthorWMF) + median_PC4_adac + median_PC3_adac +  n_comments_before + as.factor(source) + 
+    median_gerrit_reviewers,
   predicted_var = c("human_EP_prop_adac", "human_TSOL_prop_adac", "human_RK_prop_adac"),
   prediction = c("olmo_EP_prop_adac", "olmo_TSOL_prop_adac", "olmo_RK_prop_adac"),
   sample_prob = "sampling_prob", 
@@ -39,15 +44,14 @@ logit_model <- dsl(
 summary(logit_model)
 #anova(dsl_df$olmo_RK_prop, dsl_df$median_gerrit_reviewers)
 #chisq.test(table(dsl_df$isAuthorWMF, dsl_df$author_closer))
-felm_df <- dsl_df |>
-  dplyr::mutate(ttr_days = TTR / 24)
 # https://cscu.cornell.edu/wp-content/uploads/clust.pdf
 # https://statmodeling.stat.columbia.edu/2020/01/10/linear-or-logistic-regression-with-binary-outcomes/
 # https://osf.io/preprints/psyarxiv/4gmbv_v1
 felm_model <- dsl(
   model = "felm", 
   formula = dsl_score ~ human_EP_prop_adac + human_TSOL_prop_adac + human_RK_prop_adac +
-    week_index + median_PC4_adac + n_comments_before + + isAuthorWMF + median_gerrit_reviewers,
+    phase
+  + median_PC4_adac + median_PC3_adac  + n_comments_before + + isAuthorWMF,
   predicted_var = c("human_EP_prop_adac", "human_TSOL_prop_adac", "human_RK_prop_adac"),
   prediction = c("olmo_EP_prop_adac", "olmo_TSOL_prop_adac", "olmo_RK_prop_adac"),
   sample_prob = "sampling_prob",
@@ -59,8 +63,25 @@ felm_model <- dsl(
   data=felm_df
 )
 summary(felm_model)
+#httpsfelm_model#https://github.com/naoki-egami/dsl/blob/537664a54163dda52ee277071fdfd9e8df2572a6/R/estimate_g.R#L39
+
+
+dev_model <- dsl(
+  model = "logit", 
+  formula = task_resolution ~ human_EP_prop_adac + human_TSOL_prop_adac + human_RK_prop_adac 
+    + median_PC4_adac + median_PC3_adac + n_comments_before 
+    + median_gerrit_reviewers + median_gerrit_loc_delta
+    + week_index + as.factor(source) * as.factor(isAuthorWMF),
+  predicted_var = c("human_EP_prop_adac", "human_TSOL_prop_adac", "human_RK_prop_adac"),
+  prediction = c("olmo_EP_prop_adac", "olmo_TSOL_prop_adac", "olmo_RK_prop_adac"),
+  sample_prob = "sampling_prob", 
+  cluster="source",
+  cross_fit = 3,
+  sample_split = 20,
+  data=dsl_df
+)
+summary(dev_model)
 
-#https://github.com/naoki-egami/dsl/blob/537664a54163dda52ee277071fdfd9e8df2572a6/R/estimate_g.R#L39
 
 library(broom)
 library(dplyr)
@@ -79,7 +100,7 @@ tidy.dsl <- function(x, conf.int = FALSE, conf.level = 0.95, exponentiate = FALS
     out <- broom:::exponentiate(out)
   return(out)
 }
-coef_df <- tidy.dsl(felm_model)
+coef_df <- tidy.dsl(dev_model)
 ggplot(coef_df, aes(x = estimate, y = term)) +
   geom_point(size = 1) +
   geom_errorbar(aes(xmin = estimate - 1.96*std.error, xmax = estimate + 1.96 *std.error), height = 0.2) +