diff --git a/nomos-da/network/core/src/protocols/sampling/behaviour.rs b/nomos-da/network/core/src/protocols/sampling/behaviour.rs
index 10679d1b..53f8184f 100644
--- a/nomos-da/network/core/src/protocols/sampling/behaviour.rs
+++ b/nomos-da/network/core/src/protocols/sampling/behaviour.rs
@@ -73,6 +73,13 @@ impl SamplingError {
             SamplingError::ResponseChannel { peer_id, .. } => peer_id,
         }
     }
+
+    pub fn blob_id(&self) -> Option<&BlobId> {
+        match self {
+            SamplingError::Deserialize { blob_id, .. } => Some(blob_id),
+            _ => None,
+        }
+    }
 }
 
 impl Clone for SamplingError {
diff --git a/nomos-services/data-availability/sampling/Cargo.toml b/nomos-services/data-availability/sampling/Cargo.toml
index ae1c38a0..d917e99f 100644
--- a/nomos-services/data-availability/sampling/Cargo.toml
+++ b/nomos-services/data-availability/sampling/Cargo.toml
@@ -7,6 +7,7 @@ edition = "2021"
 async-trait = "0.1"
 bytes = "1.2"
 futures = "0.3"
+hex = "0.4.3"
 kzgrs-backend = { path = "../../../nomos-da/kzgrs-backend" }
 libp2p-identity = { version = "0.2" }
 nomos-core = { path = "../../../nomos-core" }
diff --git a/nomos-services/data-availability/sampling/src/backend/kzgrs.rs b/nomos-services/data-availability/sampling/src/backend/kzgrs.rs
index 139597f9..2d7bb76e 100644
--- a/nomos-services/data-availability/sampling/src/backend/kzgrs.rs
+++ b/nomos-services/data-availability/sampling/src/backend/kzgrs.rs
@@ -1,2 +1,368 @@
+// std
+use std::collections::{BTreeSet, HashMap, HashSet};
+use std::fmt::Debug;
+use std::time::{Duration, Instant};
 
+// crates
+use hex;
+use rand::distributions::Standard;
+use rand::prelude::*;
+use tokio::time;
+use tokio::time::Interval;
 
+use kzgrs_backend::common::blob::DaBlob;
+
+//
+// internal
+use crate::{backend::SamplingState, DaSamplingServiceBackend};
+use nomos_core::da::BlobId;
+use nomos_da_network_core::SubnetworkId;
+
+#[derive(Clone)]
+pub struct SamplingContext {
+    subnets: HashSet<SubnetworkId>,
+    started: Instant,
+}
+
+#[derive(Debug, Clone)]
+pub struct KzgrsDaSamplerSettings {
+    pub num_samples: u16,
+    pub old_blobs_check_interval: Duration,
+    pub blobs_validity_duration: Duration,
+}
+
+pub struct KzgrsDaSampler<R: Rng> {
+    settings: KzgrsDaSamplerSettings,
+    validated_blobs: BTreeSet<BlobId>,
+    pending_sampling_blobs: HashMap<BlobId, SamplingContext>,
+    rng: R,
+}
+
+impl<R: Rng> KzgrsDaSampler<R> {
+    fn prune_by_time(&mut self) {
+        self.pending_sampling_blobs.retain(|_blob_id, context| {
+            context.started.elapsed() < self.settings.blobs_validity_duration
+        });
+    }
+}
+
+#[async_trait::async_trait]
+impl<R: Rng + Sync + Send> DaSamplingServiceBackend<R> for KzgrsDaSampler<R> {
+    type Settings = KzgrsDaSamplerSettings;
+    type BlobId = BlobId;
+    type Blob = DaBlob;
+
+    fn new(settings: Self::Settings, rng: R) -> Self {
+        let bt: BTreeSet<BlobId> = BTreeSet::new();
+        Self {
+            settings,
+            validated_blobs: bt,
+            pending_sampling_blobs: HashMap::new(),
+            rng,
+        }
+    }
+
+    async fn prune_interval(&self) -> Interval {
+        time::interval(self.settings.old_blobs_check_interval)
+    }
+
+    async fn get_validated_blobs(&self) -> BTreeSet<Self::BlobId> {
+        self.validated_blobs.clone()
+    }
+
+    async fn mark_in_block(&mut self, blobs_ids: &[Self::BlobId]) {
+        for id in blobs_ids {
+            self.pending_sampling_blobs.remove(id);
+            self.validated_blobs.remove(id);
+        }
+    }
+
+    async fn handle_sampling_success(&mut self, blob_id: Self::BlobId, blob: Self::Blob) {
+        if let Some(ctx) = self.pending_sampling_blobs.get_mut(&blob_id) {
+            tracing::info!(
+                "subnet {} for blob id {} has been successfully sampled",
+                blob.column_idx,
+                hex::encode(blob_id)
+            );
+            ctx.subnets.insert(blob.column_idx as SubnetworkId);
+
+            // sampling of this blob_id terminated successfully
+            if ctx.subnets.len() == self.settings.num_samples as usize {
+                self.validated_blobs.insert(blob_id);
+                tracing::info!(
+                    "blob_id {} has been successfully sampled",
+                    hex::encode(blob_id)
+                );
+                // cleanup from pending samplings
+                self.pending_sampling_blobs.remove(&blob_id);
+            }
+        }
+    }
+
+    async fn handle_sampling_error(&mut self, blob_id: Self::BlobId) {
+        // If it fails a single time we consider it failed.
+        // We may want to abstract the sampling policies somewhere else at some point if we
+        // need to get fancier than this
+        self.pending_sampling_blobs.remove(&blob_id);
+        self.validated_blobs.remove(&blob_id);
+    }
+
+    async fn init_sampling(&mut self, blob_id: Self::BlobId) -> SamplingState {
+        if self.pending_sampling_blobs.contains_key(&blob_id) {
+            return SamplingState::Tracking;
+        }
+        if self.validated_blobs.contains(&blob_id) {
+            return SamplingState::Terminated;
+        }
+
+        let subnets: Vec<SubnetworkId> = Standard
+            .sample_iter(&mut self.rng)
+            .take(self.settings.num_samples as usize)
+            .collect();
+        let ctx: SamplingContext = SamplingContext {
+            subnets: HashSet::new(),
+            started: Instant::now(),
+        };
+        self.pending_sampling_blobs.insert(blob_id, ctx);
+        SamplingState::Init(subnets)
+    }
+
+    fn prune(&mut self) {
+        self.prune_by_time()
+    }
+}
+
+#[cfg(test)]
+mod test {
+
+    use std::collections::HashSet;
+    use std::time::{Duration, Instant};
+
+    use rand::prelude::*;
+    use rand::rngs::StdRng;
+
+    use crate::backend::kzgrs::{
+        DaSamplingServiceBackend, KzgrsDaSampler, KzgrsDaSamplerSettings, SamplingContext,
+        SamplingState,
+    };
+    use kzgrs_backend::common::{blob::DaBlob, Column};
+    use nomos_core::da::BlobId;
+
+    fn create_sampler(subnet_num: usize) -> KzgrsDaSampler<StdRng> {
+        let settings = KzgrsDaSamplerSettings {
+            num_samples: subnet_num as u16,
+            old_blobs_check_interval: Duration::from_millis(20),
+            blobs_validity_duration: Duration::from_millis(10),
+        };
+        let rng = StdRng::from_entropy();
+        KzgrsDaSampler::new(settings, rng)
+    }
+
+    #[tokio::test]
+    async fn test_sampler() {
+        // fictitious number of subnets
+        let subnet_num: usize = 42;
+
+        // create a sampler instance
+        let sampler = &mut create_sampler(subnet_num);
+
+        // create some blobs and blob_ids
+        let b1: BlobId = sampler.rng.gen();
+        let b2: BlobId = sampler.rng.gen();
+        let blob = DaBlob {
+            column_idx: 42,
+            column: Column(vec![]),
+            column_commitment: Default::default(),
+            aggregated_column_commitment: Default::default(),
+            aggregated_column_proof: Default::default(),
+            rows_commitments: vec![],
+            rows_proofs: vec![],
+        };
+        let blob2 = blob.clone();
+        let mut blob3 = blob2.clone();
+
+        // at start everything should be empty
+        assert!(sampler.pending_sampling_blobs.is_empty());
+        assert!(sampler.validated_blobs.is_empty());
+        assert!(sampler.get_validated_blobs().await.is_empty());
+
+        // start sampling for b1
+        let SamplingState::Init(subnets_to_sample) = sampler.init_sampling(b1).await else {
+            panic!("unexpected return value")
+        };
+        assert!(subnets_to_sample.len() == subnet_num);
+        assert!(sampler.validated_blobs.is_empty());
+        assert!(sampler.pending_sampling_blobs.len() == 1);
+
+        // start sampling for b2
+        let SamplingState::Init(subnets_to_sample2) = sampler.init_sampling(b2).await else {
+            panic!("unexpected return value")
+        };
+        assert!(subnets_to_sample2.len() == subnet_num);
+        assert!(sampler.validated_blobs.is_empty());
+        assert!(sampler.pending_sampling_blobs.len() == 2);
+
+        // mark in block for both
+        // collections should be reset
+        sampler.mark_in_block(&[b1, b2]).await;
+        assert!(sampler.pending_sampling_blobs.is_empty());
+        assert!(sampler.validated_blobs.is_empty());
+
+        // because they're reset, we need to restart sampling for the test
+        _ = sampler.init_sampling(b1).await;
+        _ = sampler.init_sampling(b2).await;
+
+        // handle ficticious error for b2
+        // b2 should be gone, b1 still around
+        sampler.handle_sampling_error(b2).await;
+        assert!(sampler.validated_blobs.is_empty());
+        assert!(sampler.pending_sampling_blobs.len() == 1);
+        assert!(sampler.pending_sampling_blobs.contains_key(&b1));
+
+        // handle ficticious sampling success for b1
+        // should still just have one pending blob, no validated blobs yet,
+        // and one subnet added to blob
+        sampler.handle_sampling_success(b1, blob).await;
+        assert!(sampler.validated_blobs.is_empty());
+        assert!(sampler.pending_sampling_blobs.len() == 1);
+        println!(
+            "{}",
+            sampler
+                .pending_sampling_blobs
+                .get(&b1)
+                .unwrap()
+                .subnets
+                .len()
+        );
+        assert!(
+            sampler
+                .pending_sampling_blobs
+                .get(&b1)
+                .unwrap()
+                .subnets
+                .len()
+                == 1
+        );
+
+        // handle_success for always the same subnet
+        // by adding number of subnets time the same subnet
+        // (column_idx did not change)
+        // should not change, still just one sampling blob,
+        // no validated blobs and one subnet
+        for _ in 1..subnet_num {
+            let b = blob2.clone();
+            sampler.handle_sampling_success(b1, b).await;
+        }
+        assert!(sampler.validated_blobs.is_empty());
+        assert!(sampler.pending_sampling_blobs.len() == 1);
+        assert!(
+            sampler
+                .pending_sampling_blobs
+                .get(&b1)
+                .unwrap()
+                .subnets
+                .len()
+                == 1
+        );
+
+        // handle_success for up to subnet size minus one subnet
+        // should still not change anything
+        // but subnets len is now subnet size minus one
+        // we already added subnet 42
+        for i in 1..(subnet_num - 1) {
+            let mut b = blob2.clone();
+            b.column_idx = i as u16;
+            sampler.handle_sampling_success(b1, b).await;
+        }
+        assert!(sampler.validated_blobs.is_empty());
+        assert!(sampler.pending_sampling_blobs.len() == 1);
+        assert!(
+            sampler
+                .pending_sampling_blobs
+                .get(&b1)
+                .unwrap()
+                .subnets
+                .len()
+                == subnet_num - 1
+        );
+
+        // now add the last subnet!
+        // we should have all subnets set,
+        // and the validated blobs should now have that blob
+        // pending blobs should now be empty
+        blob3.column_idx = (subnet_num - 1) as u16;
+        sampler.handle_sampling_success(b1, blob3).await;
+        assert!(sampler.pending_sampling_blobs.is_empty());
+        assert!(sampler.validated_blobs.len() == 1);
+        assert!(sampler.validated_blobs.contains(&b1));
+        // these checks are redundant but better safe than sorry
+        assert!(sampler.get_validated_blobs().await.len() == 1);
+        assert!(sampler.get_validated_blobs().await.contains(&b1));
+
+        // run mark_in_block for the same blob
+        // should return empty for everything
+        sampler.mark_in_block(&[b1]).await;
+        assert!(sampler.validated_blobs.is_empty());
+        assert!(sampler.pending_sampling_blobs.is_empty());
+    }
+
+    #[tokio::test]
+    async fn test_pruning() {
+        let mut sampler = create_sampler(42);
+
+        // create some sampling contexes
+        // first set will go through as in time
+        let ctx1 = SamplingContext {
+            subnets: HashSet::new(),
+            started: Instant::now(),
+        };
+        let ctx2 = ctx1.clone();
+        let ctx3 = ctx1.clone();
+
+        // second set: will fail for expired
+        let ctx11 = SamplingContext {
+            subnets: HashSet::new(),
+            started: Instant::now() - Duration::from_secs(1),
+        };
+        let ctx12 = ctx11.clone();
+        let ctx13 = ctx11.clone();
+
+        // create a couple blob ids
+        let b1: BlobId = sampler.rng.gen();
+        let b2: BlobId = sampler.rng.gen();
+        let b3: BlobId = sampler.rng.gen();
+
+        // insert first blob
+        // pruning should have no effect
+        assert!(sampler.pending_sampling_blobs.is_empty());
+        sampler.pending_sampling_blobs.insert(b1, ctx1);
+        sampler.prune();
+        assert!(sampler.pending_sampling_blobs.len() == 1);
+
+        // insert second blob
+        // pruning should have no effect
+        sampler.pending_sampling_blobs.insert(b2, ctx2);
+        sampler.prune();
+        assert!(sampler.pending_sampling_blobs.len() == 2);
+
+        // insert third blob
+        // pruning should have no effect
+        sampler.pending_sampling_blobs.insert(b3, ctx3);
+        sampler.prune();
+        assert!(sampler.pending_sampling_blobs.len() == 3);
+
+        // insert fake expired blobs
+        // pruning these should now decrease pending blobx every time
+        sampler.pending_sampling_blobs.insert(b1, ctx11);
+        sampler.prune();
+        assert!(sampler.pending_sampling_blobs.len() == 2);
+
+        sampler.pending_sampling_blobs.insert(b2, ctx12);
+        sampler.prune();
+        assert!(sampler.pending_sampling_blobs.len() == 1);
+
+        sampler.pending_sampling_blobs.insert(b3, ctx13);
+        sampler.prune();
+        assert!(sampler.pending_sampling_blobs.is_empty());
+    }
+}
diff --git a/nomos-services/data-availability/sampling/src/backend/mod.rs b/nomos-services/data-availability/sampling/src/backend/mod.rs
index 4d508843..0cc6e758 100644
--- a/nomos-services/data-availability/sampling/src/backend/mod.rs
+++ b/nomos-services/data-availability/sampling/src/backend/mod.rs
@@ -1,21 +1,33 @@
+pub mod kzgrs;
+
 // std
 use std::collections::BTreeSet;
 
 // crates
+use rand::Rng;
+use tokio::time::Interval;
 //
 // internal
 use nomos_da_network_core::SubnetworkId;
 
+pub enum SamplingState {
+    Init(Vec<SubnetworkId>),
+    Tracking,
+    Terminated,
+}
+
 #[async_trait::async_trait]
-pub trait DaSamplingServiceBackend {
+pub trait DaSamplingServiceBackend<R: Rng> {
     type Settings;
     type BlobId;
     type Blob;
 
-    fn new(settings: Self::Settings) -> Self;
+    fn new(settings: Self::Settings, rng: R) -> Self;
     async fn get_validated_blobs(&self) -> BTreeSet<Self::BlobId>;
-    async fn mark_in_block(&mut self, blobs_id: &[Self::BlobId]);
+    async fn mark_in_block(&mut self, blobs_ids: &[Self::BlobId]);
     async fn handle_sampling_success(&mut self, blob_id: Self::BlobId, blob: Self::Blob);
     async fn handle_sampling_error(&mut self, blob_id: Self::BlobId);
-    async fn init_sampling(&mut self, blob_id: Self::BlobId) -> Vec<SubnetworkId>;
+    async fn init_sampling(&mut self, blob_id: Self::BlobId) -> SamplingState;
+    async fn prune_interval(&self) -> Interval;
+    fn prune(&mut self);
 }
diff --git a/nomos-services/data-availability/sampling/src/lib.rs b/nomos-services/data-availability/sampling/src/lib.rs
index 67bf5b8d..9eea10d9 100644
--- a/nomos-services/data-availability/sampling/src/lib.rs
+++ b/nomos-services/data-availability/sampling/src/lib.rs
@@ -6,10 +6,11 @@ use std::collections::BTreeSet;
 use std::fmt::Debug;
 
 // crates
+use rand::prelude::*;
 use tokio_stream::StreamExt;
 use tracing::{error, span, Instrument, Level};
 // internal
-use backend::DaSamplingServiceBackend;
+use backend::{DaSamplingServiceBackend, SamplingState};
 use kzgrs_backend::common::blob::DaBlob;
 use network::NetworkAdapter;
 use nomos_core::da::BlobId;
@@ -46,9 +47,10 @@ pub struct DaSamplingServiceSettings<BackendSettings, NetworkSettings> {
 
 impl<B: 'static> RelayMessage for DaSamplingServiceMsg<B> {}
 
-pub struct DaSamplingService<Backend, N, S>
+pub struct DaSamplingService<Backend, N, S, R>
 where
-    Backend: DaSamplingServiceBackend + Send,
+    R: SeedableRng + RngCore,
+    Backend: DaSamplingServiceBackend<R> + Send,
     Backend::Settings: Clone,
     Backend::Blob: Debug + 'static,
     Backend::BlobId: Debug + 'static,
@@ -60,9 +62,10 @@ where
     sampler: Backend,
 }
 
-impl<Backend, N, S> DaSamplingService<Backend, N, S>
+impl<Backend, N, S, R> DaSamplingService<Backend, N, S, R>
 where
-    Backend: DaSamplingServiceBackend<BlobId = BlobId, Blob = DaBlob> + Send + 'static,
+    R: SeedableRng + RngCore,
+    Backend: DaSamplingServiceBackend<R, BlobId = BlobId, Blob = DaBlob> + Send + 'static,
     Backend::Settings: Clone,
     N: NetworkAdapter + Send + 'static,
     N::Settings: Clone,
@@ -89,12 +92,16 @@ where
     ) {
         match msg {
             DaSamplingServiceMsg::TriggerSampling { blob_id } => {
-                let sampling_subnets = sampler.init_sampling(blob_id).await;
-                if let Err(e) = network_adapter
-                    .start_sampling(blob_id, &sampling_subnets)
-                    .await
+                if let SamplingState::Init(sampling_subnets) = sampler.init_sampling(blob_id).await
                 {
-                    error!("Error sampling for BlobId: {blob_id:?}: {e}");
+                    if let Err(e) = network_adapter
+                        .start_sampling(blob_id, &sampling_subnets)
+                        .await
+                    {
+                        // we can short circuit the failure from beginning
+                        sampler.handle_sampling_error(blob_id).await;
+                        error!("Error sampling for BlobId: {blob_id:?}: {e}");
+                    }
                 }
             }
             DaSamplingServiceMsg::GetValidatedBlobs { reply_channel } => {
@@ -115,15 +122,20 @@ where
                 sampler.handle_sampling_success(blob_id, *blob).await;
             }
             SamplingEvent::SamplingError { error } => {
+                if let Some(blob_id) = error.blob_id() {
+                    sampler.handle_sampling_error(*blob_id).await;
+                    return;
+                }
                 error!("Error while sampling: {error}");
             }
         }
     }
 }
 
-impl<Backend, N, S> ServiceData for DaSamplingService<Backend, N, S>
+impl<Backend, N, S, R> ServiceData for DaSamplingService<Backend, N, S, R>
 where
-    Backend: DaSamplingServiceBackend + Send,
+    R: SeedableRng + RngCore,
+    Backend: DaSamplingServiceBackend<R> + Send,
     Backend::Settings: Clone,
     Backend::Blob: Debug + 'static,
     Backend::BlobId: Debug + 'static,
@@ -138,9 +150,10 @@ where
 }
 
 #[async_trait::async_trait]
-impl<Backend, N, S> ServiceCore for DaSamplingService<Backend, N, S>
+impl<Backend, N, S, R> ServiceCore for DaSamplingService<Backend, N, S, R>
 where
-    Backend: DaSamplingServiceBackend<BlobId = BlobId, Blob = DaBlob> + Send + Sync + 'static,
+    R: SeedableRng + RngCore,
+    Backend: DaSamplingServiceBackend<R, BlobId = BlobId, Blob = DaBlob> + Send + Sync + 'static,
     Backend::Settings: Clone + Send + Sync + 'static,
     N: NetworkAdapter + Send + Sync + 'static,
     N::Settings: Clone + Send + Sync + 'static,
@@ -151,11 +164,12 @@ where
         } = service_state.settings_reader.get_updated_settings();
 
         let network_relay = service_state.overwatch_handle.relay();
+        let rng = R::from_entropy();
 
         Ok(Self {
             network_relay,
             service_state,
-            sampler: Backend::new(sampling_settings),
+            sampler: Backend::new(sampling_settings, rng),
         })
     }
 
@@ -175,6 +189,7 @@ where
         let mut network_adapter = N::new(network_relay).await;
 
         let mut sampling_message_stream = network_adapter.listen_to_sampling_messages().await?;
+        let mut next_prune_tick = sampler.prune_interval().await;
 
         let mut lifecycle_stream = service_state.lifecycle_handle.message_stream();
         async {
@@ -191,6 +206,11 @@ where
                             break;
                         }
                     }
+                    // cleanup not on time samples
+                    _ = next_prune_tick.tick() => {
+                        sampler.prune();
+                    }
+
                 }
             }
         }