diff --git a/harmony/src/domain/topology/k8s.rs b/harmony/src/domain/topology/k8s.rs
index 388ff9d..1887c33 100644
--- a/harmony/src/domain/topology/k8s.rs
+++ b/harmony/src/domain/topology/k8s.rs
@@ -5,7 +5,7 @@ use k8s_openapi::{
 };
 use kube::{
     Client, Config, Error, Resource,
-    api::{Api, AttachParams, ListParams, Patch, PatchParams, ResourceExt},
+    api::{Api, AttachParams, DeleteParams, ListParams, Patch, PatchParams, ResourceExt},
     config::{KubeConfigOptions, Kubeconfig},
     core::ErrorResponse,
     runtime::reflector::Lookup,
@@ -17,7 +17,9 @@ use kube::{
 };
 use log::{debug, error, trace};
 use serde::{Serialize, de::DeserializeOwned};
+use serde_json::json;
 use similar::TextDiff;
+use tokio::io::AsyncReadExt;
 
 #[derive(new, Clone)]
 pub struct K8sClient {
@@ -51,6 +53,66 @@ impl K8sClient {
         })
     }
 
+    pub async fn get_deployment(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+    ) -> Result<Option<Deployment>, Error> {
+        let deps: Api<Deployment> = if let Some(ns) = namespace {
+            Api::namespaced(self.client.clone(), ns)
+        } else {
+            Api::default_namespaced(self.client.clone())
+        };
+        Ok(deps.get_opt(name).await?)
+    }
+
+    pub async fn get_pod(&self, name: &str, namespace: Option<&str>) -> Result<Option<Pod>, Error> {
+        let pods: Api<Pod> = if let Some(ns) = namespace {
+            Api::namespaced(self.client.clone(), ns)
+        } else {
+            Api::default_namespaced(self.client.clone())
+        };
+        Ok(pods.get_opt(name).await?)
+    }
+
+    pub async fn scale_deployment(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+        replicas: u32,
+    ) -> Result<(), Error> {
+        let deployments: Api<Deployment> = if let Some(ns) = namespace {
+            Api::namespaced(self.client.clone(), ns)
+        } else {
+            Api::default_namespaced(self.client.clone())
+        };
+
+        let patch = json!({
+            "spec": {
+                "replicas": replicas
+            }
+        });
+        let pp = PatchParams::default();
+        let scale = Patch::Apply(&patch);
+        deployments.patch_scale(name, &pp, &scale).await?;
+        Ok(())
+    }
+
+    pub async fn delete_deployment(
+        &self,
+        name: &str,
+        namespace: Option<&str>,
+    ) -> Result<(), Error> {
+        let deployments: Api<Deployment> = if let Some(ns) = namespace {
+            Api::namespaced(self.client.clone(), ns)
+        } else {
+            Api::default_namespaced(self.client.clone())
+        };
+        let delete_params = DeleteParams::default();
+        deployments.delete(name, &delete_params).await?;
+        Ok(())
+    }
+
     pub async fn wait_until_deployment_ready(
         &self,
         name: String,
@@ -76,6 +138,68 @@ impl K8sClient {
         }
     }
 
+    /// Will execute a commond in the first pod found that matches the specified label
+    /// '{label}={name}'
+    pub async fn exec_app_capture_output(
+        &self,
+        name: String,
+        label: String,
+        namespace: Option<&str>,
+        command: Vec<&str>,
+    ) -> Result<String, String> {
+        let api: Api<Pod>;
+
+        if let Some(ns) = namespace {
+            api = Api::namespaced(self.client.clone(), ns);
+        } else {
+            api = Api::default_namespaced(self.client.clone());
+        }
+        let pod_list = api
+            .list(&ListParams::default().labels(format!("{label}={name}").as_str()))
+            .await
+            .expect("couldn't get list of pods");
+
+        let res = api
+            .exec(
+                pod_list
+                    .items
+                    .first()
+                    .expect("couldn't get pod")
+                    .name()
+                    .expect("couldn't get pod name")
+                    .into_owned()
+                    .as_str(),
+                command,
+                &AttachParams::default().stdout(true).stderr(true),
+            )
+            .await;
+        match res {
+            Err(e) => Err(e.to_string()),
+            Ok(mut process) => {
+                let status = process
+                    .take_status()
+                    .expect("Couldn't get status")
+                    .await
+                    .expect("Couldn't unwrap status");
+
+                if let Some(s) = status.status {
+                    let mut stdout_buf = String::new();
+                    if let Some(mut stdout) = process.stdout().take() {
+                        stdout.read_to_string(&mut stdout_buf).await;
+                    }
+                    debug!("Status: {} - {:?}", s, status.details);
+                    if s == "Success" {
+                        Ok(stdout_buf)
+                    } else {
+                        Err(s)
+                    }
+                } else {
+                    Err("Couldn't get inner status of pod exec".to_string())
+                }
+            }
+        }
+    }
+
     /// Will execute a command in the first pod found that matches the label `app.kubernetes.io/name={name}`
     pub async fn exec_app(
         &self,
diff --git a/harmony/src/modules/mod.rs b/harmony/src/modules/mod.rs
index e9b6c52..6df5c41 100644
--- a/harmony/src/modules/mod.rs
+++ b/harmony/src/modules/mod.rs
@@ -14,5 +14,6 @@ pub mod monitoring;
 pub mod okd;
 pub mod opnsense;
 pub mod prometheus;
+pub mod storage;
 pub mod tenant;
 pub mod tftp;
diff --git a/harmony/src/modules/storage/ceph/ceph_osd_replacement_score.rs b/harmony/src/modules/storage/ceph/ceph_osd_replacement_score.rs
new file mode 100644
index 0000000..a4b0cb0
--- /dev/null
+++ b/harmony/src/modules/storage/ceph/ceph_osd_replacement_score.rs
@@ -0,0 +1,419 @@
+use std::{
+    process::Command,
+    sync::Arc,
+    time::{Duration, Instant},
+};
+
+use async_trait::async_trait;
+use log::{info, warn};
+use serde::{Deserialize, Serialize};
+use tokio::time::sleep;
+
+use crate::{
+    data::{Id, Version},
+    interpret::{Interpret, InterpretError, InterpretName, InterpretStatus, Outcome},
+    inventory::Inventory,
+    score::Score,
+    topology::{K8sclient, Topology, k8s::K8sClient},
+};
+
+#[derive(Debug, Clone, Serialize)]
+pub struct CephRemoveOsd {
+    osd_deployment_name: String,
+    rook_ceph_namespace: String,
+}
+
+impl<T: Topology + K8sclient> Score<T> for CephRemoveOsd {
+    fn name(&self) -> String {
+        format!("CephRemoveOsdScore")
+    }
+
+    #[doc(hidden)]
+    fn create_interpret(&self) -> Box<dyn Interpret<T>> {
+        Box::new(CephRemoveOsdInterpret {
+            score: self.clone(),
+        })
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct CephRemoveOsdInterpret {
+    score: CephRemoveOsd,
+}
+
+#[async_trait]
+impl<T: Topology + K8sclient> Interpret<T> for CephRemoveOsdInterpret {
+    async fn execute(
+        &self,
+        _inventory: &Inventory,
+        topology: &T,
+    ) -> Result<Outcome, InterpretError> {
+        let client = topology.k8s_client().await.unwrap();
+        self.verify_ceph_toolbox_exists(client.clone()).await?;
+        self.scale_deployment(client.clone()).await?;
+        self.verify_deployment_scaled(client.clone()).await?;
+        self.delete_deployment(client.clone()).await?;
+        self.verify_deployment_deleted(client.clone()).await?;
+        let osd_id_full = self.get_ceph_osd_id().unwrap();
+        self.purge_ceph_osd(client.clone(), &osd_id_full).await?;
+        self.verify_ceph_osd_removal(client.clone(), &osd_id_full)
+            .await?;
+
+        Ok(Outcome::success(format!(
+            "Successfully removed OSD {} from rook-ceph cluster by deleting deployment {}",
+            osd_id_full, self.score.osd_deployment_name
+        )))
+    }
+    fn get_name(&self) -> InterpretName {
+        todo!()
+    }
+
+    fn get_version(&self) -> Version {
+        todo!()
+    }
+
+    fn get_status(&self) -> InterpretStatus {
+        todo!()
+    }
+
+    fn get_children(&self) -> Vec<Id> {
+        todo!()
+    }
+}
+
+impl CephRemoveOsdInterpret {
+    pub fn get_ceph_osd_id(&self) -> Result<String, InterpretError> {
+        let osd_id_numeric = self
+            .score
+            .osd_deployment_name
+            .split('-')
+            .nth(3)
+            .ok_or_else(|| {
+                InterpretError::new(format!(
+                    "Could not parse OSD id from deployment name {}",
+                    self.score.osd_deployment_name
+                ))
+            })?;
+        let osd_id_full = format!("osd.{}", osd_id_numeric);
+
+        info!(
+            "Targeting Ceph OSD: {} (parsed from deployment {})",
+            osd_id_full, self.score.osd_deployment_name
+        );
+
+        Ok(osd_id_full)
+    }
+
+    pub async fn verify_ceph_toolbox_exists(
+        &self,
+        client: Arc<K8sClient>,
+    ) -> Result<Outcome, InterpretError> {
+        let toolbox_dep = "rook-ceph-tools".to_string();
+
+        match client
+            .get_deployment(&toolbox_dep, Some(&self.score.rook_ceph_namespace))
+            .await
+        {
+            Ok(Some(deployment)) => {
+                if let Some(status) = deployment.status {
+                    let ready_count = status.ready_replicas.unwrap_or(0);
+                    if ready_count >= 1 {
+                        return Ok(Outcome::success(format!(
+                            "'{}' is ready with {} replica(s).",
+                            &toolbox_dep, ready_count
+                        )));
+                    } else {
+                        return Err(InterpretError::new(
+                            "ceph-tool-box not ready in cluster".to_string(),
+                        ));
+                    }
+                } else {
+                    Err(InterpretError::new(format!(
+                        "failed to get deployment status {}",
+                        &toolbox_dep
+                    )))
+                }
+            }
+            Ok(None) => Err(InterpretError::new(format!(
+                "Deployment '{}' not found in namespace '{}'.",
+                &toolbox_dep, self.score.rook_ceph_namespace
+            ))),
+            Err(e) => Err(InterpretError::new(format!(
+                "Failed to query for deployment '{}': {}",
+                &toolbox_dep, e
+            ))),
+        }
+    }
+
+    pub async fn scale_deployment(
+        &self,
+        client: Arc<K8sClient>,
+    ) -> Result<Outcome, InterpretError> {
+        info!(
+            "Scaling down OSD deployment: {}",
+            self.score.osd_deployment_name
+        );
+        client
+            .scale_deployment(
+                &self.score.osd_deployment_name,
+                Some(&self.score.rook_ceph_namespace),
+                0,
+            )
+            .await?;
+        Ok(Outcome::success(format!(
+            "Scaled down deployment {}",
+            self.score.osd_deployment_name
+        )))
+    }
+
+    pub async fn verify_deployment_scaled(
+        &self,
+        client: Arc<K8sClient>,
+    ) -> Result<Outcome, InterpretError> {
+        let (timeout, interval, start) = self.build_timer();
+
+        info!("Waiting for OSD deployment to scale down to 0 replicas");
+        loop {
+            let dep = client
+                .get_deployment(
+                    &self.score.osd_deployment_name,
+                    Some(&self.score.rook_ceph_namespace),
+                )
+                .await?;
+
+            if let Some(deployment) = dep {
+                if let Some(status) = deployment.status {
+                    if status.replicas.unwrap_or(1) == 0 && status.ready_replicas.unwrap_or(1) == 0
+                    {
+                        return Ok(Outcome::success(
+                            "Deployment successfully scaled down.".to_string(),
+                        ));
+                    }
+                }
+            }
+
+            if start.elapsed() > timeout {
+                return Err(InterpretError::new(format!(
+                    "Timed out waiting for deployment {} to scale down",
+                    self.score.osd_deployment_name
+                )));
+            }
+            sleep(interval).await;
+        }
+    }
+
+    fn build_timer(&self) -> (Duration, Duration, Instant) {
+        let timeout = Duration::from_secs(120);
+        let interval = Duration::from_secs(5);
+        let start = Instant::now();
+        (timeout, interval, start)
+    }
+    pub async fn delete_deployment(
+        &self,
+        client: Arc<K8sClient>,
+    ) -> Result<Outcome, InterpretError> {
+        info!(
+            "Deleting OSD deployment: {}",
+            self.score.osd_deployment_name
+        );
+        client
+            .delete_deployment(
+                &self.score.osd_deployment_name,
+                Some(&self.score.rook_ceph_namespace),
+            )
+            .await?;
+        Ok(Outcome::success(format!(
+            "deployment {} deleted",
+            self.score.osd_deployment_name
+        )))
+    }
+
+    pub async fn verify_deployment_deleted(
+        &self,
+        client: Arc<K8sClient>,
+    ) -> Result<Outcome, InterpretError> {
+        let (timeout, interval, start) = self.build_timer();
+
+        info!("Waiting for OSD deployment to scale down to 0 replicas");
+        loop {
+            let dep = client
+                .get_deployment(
+                    &self.score.osd_deployment_name,
+                    Some(&self.score.rook_ceph_namespace),
+                )
+                .await?;
+
+            if dep.is_none() {
+                info!(
+                    "Deployment {} successfully deleted.",
+                    self.score.osd_deployment_name
+                );
+                return Ok(Outcome::success(format!(
+                    "Deployment {} deleted.",
+                    self.score.osd_deployment_name
+                )));
+            }
+
+            if start.elapsed() > timeout {
+                return Err(InterpretError::new(format!(
+                    "Timed out waiting for deployment {} to be deleted",
+                    self.score.osd_deployment_name
+                )));
+            }
+            sleep(interval).await;
+        }
+    }
+
+    fn get_osd_tree(&self, json: serde_json::Value) -> Result<CephOsdTree, InterpretError> {
+        let nodes = json.get("nodes").ok_or_else(|| {
+            InterpretError::new("Missing 'nodes' field in ceph osd tree JSON".to_string())
+        })?;
+        let tree: CephOsdTree = CephOsdTree {
+            nodes: serde_json::from_value(nodes.clone()).map_err(|e| {
+                InterpretError::new(format!("Failed to parse ceph osd tree JSON: {}", e))
+            })?,
+        };
+        Ok(tree)
+    }
+
+    pub async fn purge_ceph_osd(
+        &self,
+        client: Arc<K8sClient>,
+        osd_id_full: &str,
+    ) -> Result<Outcome, InterpretError> {
+        info!(
+            "Purging OSD {} from Ceph cluster and removing its auth key",
+            osd_id_full
+        );
+        client
+            .exec_app_capture_output(
+                "rook-ceph-tools".to_string(),
+                "app".to_string(),
+                Some(&self.score.rook_ceph_namespace),
+                vec![
+                    format!("ceph osd purge {osd_id_full} --yes-i-really-mean-it").as_str(),
+                    format!("ceph auth del osd.{osd_id_full}").as_str(),
+                ],
+            )
+            .await?;
+        Ok(Outcome::success(format!(
+            "osd id {} removed from osd tree",
+            osd_id_full
+        )))
+    }
+
+    pub async fn verify_ceph_osd_removal(
+        &self,
+        client: Arc<K8sClient>,
+        osd_id_full: &str,
+    ) -> Result<Outcome, InterpretError> {
+        let (timeout, interval, start) = self.build_timer();
+        info!(
+            "Verifying OSD {} has been removed from the Ceph tree...",
+            osd_id_full
+        );
+        loop {
+            let output = client
+                .exec_app_capture_output(
+                    "rook-ceph-tools".to_string(),
+                    "app".to_string(),
+                    Some(&self.score.rook_ceph_namespace),
+                    vec!["ceph osd tree -f json"],
+                )
+                .await?;
+            let tree =
+                self.get_osd_tree(serde_json::from_str(&output).expect("could not extract json"));
+
+            let osd_found = tree
+                .unwrap()
+                .nodes
+                .iter()
+                .any(|node| node.name == osd_id_full);
+
+            if !osd_found {
+                return Ok(Outcome::success(format!(
+                    "Successfully verified that OSD {} is removed from the Ceph cluster.",
+                    osd_id_full,
+                )));
+            }
+
+            if start.elapsed() > timeout {
+                return Err(InterpretError::new(format!(
+                    "Timed out waiting for OSD {} to be removed from Ceph tree",
+                    osd_id_full
+                )));
+            }
+
+            warn!(
+                "OSD {} still found in Ceph tree, retrying in {:?}...",
+                osd_id_full, interval
+            );
+            sleep(interval).await;
+        }
+    }
+}
+#[derive(Debug, Deserialize, PartialEq)]
+pub struct CephOsdTree {
+    pub nodes: Vec<CephNode>,
+}
+
+#[derive(Debug, Deserialize, PartialEq)]
+pub struct CephNode {
+    pub id: i32,
+    pub name: String,
+    #[serde(rename = "type")]
+    pub node_type: String,
+    pub type_id: Option<i32>,
+    pub children: Option<Vec<i32>>,
+    pub exists: Option<i32>,
+    pub status: Option<String>,
+}
+
+#[cfg(test)]
+mod tests {
+    use serde_json::json;
+
+    use super::*;
+
+    #[test]
+    fn test_get_osd_tree() {
+        let json_data = json!({
+            "nodes": [
+                {"id": 1, "name": "osd.1", "type": "osd", "primary_affinity":"1"},
+                {"id": 2, "name": "osd.2", "type": "osd", "crush_weight": 1.22344}
+            ]
+        });
+        let interpret = CephRemoveOsdInterpret {
+            score: CephRemoveOsd {
+                osd_deployment_name: "osd-1".to_string(),
+                rook_ceph_namespace: "dummy_ns".to_string(),
+            },
+        };
+        let json = interpret.get_osd_tree(json_data).unwrap();
+
+        let expected = CephOsdTree {
+            nodes: vec![
+                CephNode {
+                    id: 1,
+                    name: "osd.1".to_string(),
+                    node_type: "osd".to_string(),
+                    type_id: None,
+                    children: None,
+                    exists: None,
+                    status: None,
+                },
+                CephNode {
+                    id: 2,
+                    name: "osd.2".to_string(),
+                    node_type: "osd".to_string(),
+                    type_id: None,
+                    children: None,
+                    exists: None,
+                    status: None,
+                },
+            ],
+        };
+
+        assert_eq!(json, expected);
+    }
+}
diff --git a/harmony/src/modules/storage/ceph/mod.rs b/harmony/src/modules/storage/ceph/mod.rs
new file mode 100644
index 0000000..a993c3d
--- /dev/null
+++ b/harmony/src/modules/storage/ceph/mod.rs
@@ -0,0 +1 @@
+pub mod ceph_osd_replacement_score;
diff --git a/harmony/src/modules/storage/mod.rs b/harmony/src/modules/storage/mod.rs
new file mode 100644
index 0000000..ee3e235
--- /dev/null
+++ b/harmony/src/modules/storage/mod.rs
@@ -0,0 +1 @@
+pub mod ceph;