diff --git a/content/blog/a-backup-is-not-disaster-recovery.md b/content/blog/a-backup-is-not-disaster-recovery.md
new file mode 100644
index 000000000..2e4a7ebb8
--- /dev/null
+++ b/content/blog/a-backup-is-not-disaster-recovery.md
@@ -0,0 +1,491 @@
+---
+title: "A Backup Is Not Disaster Recovery: Three Kubernetes Demos That Prove It"
+seoTitle: "A Backup Is Not Disaster Recovery: Three Kubernetes DR Demos"
+seoDescription: "Three live Kubernetes demos - Velero restores, the GitOps trap, and VolumeGroupSnapshot - that show why backups alone are not disaster recovery."
+datePublished: 2026-08-14T10:00:00.000Z
+slug: a-backup-is-not-disaster-recovery
+authors: ["saiyam-pathak", "saloni-narang"]
+cover: /img/blog/a-backup-is-not-disaster-recovery/lab-setup.svg
+tags: ["kubernetes", "disaster-recovery", "velero", "gitops"]
+---
+At KubeCon + CloudNativeCon Japan 2026, the two of us gave a talk called "Is
+Your Kubernetes Disaster Recovery Actually Ready?" We opened with a poll.
+Raise your hand if you take backups of your clusters. Almost every hand in the
+room went up. Keep it up if you have ever restored a complete stateful
+application into a clean cluster. We looked around the room. Nobody.
+
+That gap between the two questions is this entire post. Everything below ran
+live on stage on one laptop with the Wi-Fi off, and every terminal output you
+see here is a real capture from a rerun of the same lab, not a mockup. The
+full lab is reproducible from this repo:
+https://github.com/saiyam1814/kubecon-japan-dr-demo
+
+## The one sentence version
+
+A backup is a recovery point you have. Disaster recovery is a capability you
+can prove. The three demos walk from the comfortable half of that sentence to
+the uncomfortable half.
+
+## Two numbers rule everything
+
+![RPO and RTO on one timeline](/img/blog/a-backup-is-not-disaster-recovery/rpo-rto.svg)
+
+- **RPO (recovery point objective)** is how much data the business can afford
+  to lose. Hourly backups can support roughly an hour of RPO, but only if
+  every backup completes and the recovery point is usable. Your achieved RPO
+  is the latest recoverable point, not the schedule.
+- **RTO (recovery time objective)** is how long until you serve users again,
+  including detection, decision making, infrastructure recovery, data restore,
+  validation, and traffic cutover.
+
+We kept returning to one point on stage: almost nobody has a real RTO
+number. People have estimates. An estimate is a guess until a stopwatch has
+watched a restore. By the end of this post you will see our stopwatch number.
+
+And what has to come back for recovery to count? Four layers: cluster state,
+application definitions, persistent data, and access plus traffic. Individual
+layers usually recover fine. Recovery fails at the joins between them.
+
+![The four recovery layers](/img/blog/a-backup-is-not-disaster-recovery/layers.svg)
+
+## The lab
+
+![The whole lab in one picture](/img/blog/a-backup-is-not-disaster-recovery/lab-setup.svg)
+
+Let's walk through the lab. Two clusters and two services, all local:
+
+- **Production** runs on [kiac](https://github.com/saiyam1814/kiac), Saiyam's
+  open source project that runs Kubernetes in Apple containers. Every node is its
+  own lightweight VM with its own kernel. This matters in demo 2: losing
+  production means powering off a machine, not stopping a container that
+  pretends to be one.
+- **Recovery** is a kind cluster that exists before anything goes wrong,
+  which is the first rule of DR.
+- **The backup vault** is SeaweedFS, an S3-compatible object store. Both
+  clusters point their Velero at the same bucket. The vault lives outside
+  both clusters on purpose, so losing either cluster can never take the
+  recovery points with it. To be clear, "vault" here means the backup store,
+  not a secret store.
+- **Git** is a local Gitea with the app manifests, watched by Argo CD in the
+  recovery cluster.
+
+The app is a Postgres guestbook with four rows. Row four is Fatima, our
+fictional on-call engineer. Her note says "owns the pager tonight." Her pager
+goes off at 2:07 AM when the region hosting her cluster disappears, and the
+monitoring dashboard times out because monitoring lived in the same cluster.
+Keep her in mind.
+
+## Demo 1: the backup is real, and what Velero actually does
+
+![Velero moves objects and volume bytes to an S3 store](/img/blog/a-backup-is-not-disaster-recovery/velero-flow.svg)
+
+[Velero](https://www.cncf.io/projects/velero/) entered the CNCF Sandbox in
+March 2026. It backs up Kubernetes
+resources and persistent volume data, and it can protect that volume data
+three ways: provider or CSI snapshots, file system backup, or CSI Snapshot
+Data Movement. Our lab uses the last one.
+
+Here is production before anything happens:
+
+```console
+$ kubectl -n guestbook exec statefulset/postgres -- \
+    psql -U postgres -d guestbook -c 'select * from attendees;'
+ id |  name  |            note
+----+--------+-----------------------------
+  1 | Priya  | came for the demos
+  2 | Marcus | still calls it swarm
+  3 | Chen   | has tested a restore. once.
+  4 | Fatima | owns the pager tonight
+(4 rows)
+```
+
+One Velero backup of the namespace already ran. Find it and keep its name:
+
+```bash
+kubectl -n velero get backups.velero.io
+export BACKUP=guestbook-rehearsal-20260727001126   # your newest Completed one
+```
+
+Most people stop checking at the green status. Go one step further and look
+at the part almost nobody checks, the DataUpload:
+
+```console
+$ kubectl -n velero get datauploads -l velero.io/backup-name=$BACKUP \
+    -o custom-columns='NAME:.metadata.name,PHASE:.status.phase,BYTES:.status.progress.bytesDone'
+NAME                                       PHASE       BYTES
+guestbook-rehearsal-20260727001126-q2j9m   Completed   47989888
+```
+
+That is Velero's data mover (Kopia) confirming that 47,989,888 bytes of
+actual Postgres volume data left the cluster and landed in the S3 vault - the
+data itself, not just the YAML. If your backup tool cannot show you this
+number, ask it why.
+
+On stage we then deleted the whole namespace, PVC included, and restored it
+with one command:
+
+```bash
+kubectl delete namespace guestbook --wait
+velero restore create --from-backup "$BACKUP" --wait
+```
+
+The same four rows came back. Useful, real, and about two minutes. But this
+is the happy path, and it hides three things Velero does not do
+automatically:
+
+1. Protecting volume data does not automatically make a database backup
+   application consistent. You configure
+   [backup hooks](https://velero.io/docs/v1.18/backup-hooks/) to flush or
+   quiesce when the app requires it.
+2. Restoring onto different infrastructure may require storage class mappings
+   and other transformations. Velero provides the mechanisms; your team must
+   design and test them.
+3. A backup phase of Completed means the backup operation completed. It does
+   not prove the application will start, contain the expected data, or serve
+   traffic. Only an end to end recovery test provides that evidence.
+
+**The question the audience asked:** "So Velero restored the namespace. Can
+it restore the whole infrastructure too?"
+
+No, and the boundary is worth memorizing. Velero's scope is selectable, from
+one namespace up to the entire cluster including cluster-scoped resources.
+But Velero restores resources into a cluster that already exists. It does not
+create the cluster, the nodes, the network, the load balancers, or DNS.
+Velero recovers what is inside Kubernetes. Something else must recover
+Kubernetes itself, and that something is infrastructure as code or Cluster
+API. If your DR plan starts with "restore the backup," ask what the backup
+gets restored into.
+
+## Demo 2: everything is green and the database is empty
+
+![The GitOps trap: Argo rebuilds the declarations, the vault holds the data](/img/blog/a-backup-is-not-disaster-recovery/gitops-trap.svg)
+
+This is the demo the room remembers. We recorded the time and powered off the
+production VM:
+
+```console
+$ date
+Sat Aug  8 12:09:33 IST 2026
+$ container stop kiac-drprod-control-plane
+kiac-drprod-control-plane
+
+$ kubectl --context kiac-drprod get nodes --request-timeout=4s
+Unable to connect to the server: net/http: request canceled while waiting
+for connection (Client.Timeout exceeded while awaiting headers)
+```
+
+Production is gone. Monitoring that lived in it is gone. What is left is the
+recovery cluster, which existed before the disaster, with its own Velero
+pointed at the same vault and Argo CD pointed at Git. Let's look at what it
+does not have:
+
+```console
+$ kubectl --context kind-dr-dr get namespaces
+NAME                 STATUS   AGE
+argocd               Active   12d
+default              Active   12d
+kube-node-lease      Active   12d
+kube-public          Active   12d
+kube-system          Active   12d
+ledger               Active   9d
+local-path-storage   Active   12d
+velero               Active   12d
+```
+
+No guestbook. This cluster has never run our app.
+
+Now the 2026 on-call reflex: we have GitOps, the whole app is declared in
+Git, just sync it. So we triggered Argo CD and waited for it to go green:
+
+```bash
+kubectl -n argocd patch application guestbook --type merge \
+  -p '{"operation":{"initiatedBy":{"username":"on-call"},"sync":{"revision":"HEAD"}}}'
+kubectl -n argocd wait application/guestbook \
+  --for=jsonpath='{.status.sync.status}'=Synced --timeout=300s
+kubectl -n guestbook rollout status statefulset/postgres --timeout 300s
+```
+
+```console
+application.argoproj.io/guestbook patched
+application.argoproj.io/guestbook condition met
+partitioned roll out complete: 1 new pods have been updated...
+
+NAME        SYNC STATUS   HEALTH STATUS
+guestbook   Synced        Progressing
+```
+
+Synced. Postgres Running and Ready. Every dashboard green. On stage we said
+this is the moment half the room posts "we are recovered" in Slack. Then we
+queried the data:
+
+```console
+$ kubectl -n guestbook exec statefulset/postgres -- \
+    psql -U postgres -d guestbook -c 'select * from attendees;'
+ERROR:  relation "attendees" does not exist
+LINE 1: select * from attendees;
+                      ^
+command terminated with exit code 1
+```
+
+The database is running and it is empty. Nothing malfunctioned. Git only ever
+contained the declarations, so Kubernetes did exactly what the YAML says:
+create a StatefulSet, create a Service, and provision a brand new, empty
+volume for the PVC. Git has never seen a single row of data. GitOps
+reconstructed the declared state perfectly and restored none of the stored
+state.
+
+**The question the audience asked:** "Why did we need Velero here if GitOps
+already brought the app back?"
+
+Because there were two different things to bring back and each tool carries
+exactly one of them: Git stores intent, and backups store state. The data
+existed in exactly one recoverable place, the Velero backup in the shared
+vault. So the real recovery is both tools in the right order:
+
+```console
+$ kubectl delete namespace guestbook --wait
+namespace "guestbook" deleted
+$ velero restore create --from-backup "$BACKUP" --wait
+Restore completed with status: Completed.
+$ kubectl -n guestbook exec statefulset/postgres -- \
+    psql -U postgres -d guestbook -c 'select * from attendees;'
+ id |  name  |            note
+----+--------+-----------------------------
+  1 | Priya  | came for the demos
+  2 | Marcus | still calls it swarm
+  3 | Chen   | has tested a restore. once.
+  4 | Fatima | owns the pager tonight
+(4 rows)
+$ date
+Sat Aug  8 12:11:28 IST 2026
+```
+
+Fatima is back, on a cluster that never ran the app, and even on different
+infrastructure: the backup was taken on a kiac VM and restored onto kind. A
+disaster may force you onto different infrastructure, so portability is
+something you test, not assume.
+
+Now the stopwatch. On stage, from powering off production to validated data
+in the recovery cluster took four minutes, 16:06 to 16:10 on the on-screen
+clock. The capture above, a rehearsed rerun, took just under two. The line we
+used on stage is the one we want you to keep: the first time we "recovered,"
+when the dashboards went green and the Slack message went out, was not the
+recovery. The second time, when the data came back and we checked it, was.
+And remember this measures only the scripted slice. A production RTO wraps
+detection, decision, traffic cutover, and failback around it.
+
+## Demo 3: two perfect snapshots, one broken application
+
+![Two snapshots from different moments tear the data](/img/blog/a-backup-is-not-disaster-recovery/torn-snapshots.svg)
+
+Real stateful applications span multiple volumes: database data plus WAL,
+Kafka brokers, replica sets. Our stand-in is a tiny ledger app writing
+matched pairs, order n to one PVC and payment n to another, five times a
+second. One invariant: every payment must have its order.
+
+We snapshotted the orders volume, let the app keep writing for five seconds,
+then snapshotted the payments volume. Each one is a plain CSI VolumeSnapshot,
+and 121152 is the run id the lab script stamps on every object in one run:
+
+```bash
+kubectl -n ledger apply -f - <<SNAP
+apiVersion: snapshot.storage.k8s.io/v1
+kind: VolumeSnapshot
+metadata:
+  name: torn-orders-121152
+spec:
+  volumeSnapshotClassName: csi-hostpath-snapclass
+  source:
+    persistentVolumeClaimName: orders-pvc
+SNAP
+sleep 5   # the app keeps writing matched pairs the whole time
+# then the identical VolumeSnapshot for payments-pvc, five seconds later
+```
+
+```console
+volumesnapshot.snapshot.storage.k8s.io/torn-orders-121152 created
+volumesnapshot.snapshot.storage.k8s.io/torn-orders-121152 condition met
+volumesnapshot.snapshot.storage.k8s.io/torn-payments-121152 created
+volumesnapshot.snapshot.storage.k8s.io/torn-payments-121152 condition met
+```
+
+Both ReadyToUse. Both individually perfect. Then we restored both into new
+PVCs using dataSource, the standard way to restore any CSI snapshot, and ran
+a verifier job that mounts both restored volumes and compares the last
+sequence number on each:
+
+```bash
+kubectl -n ledger apply -f - <<PVC
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: restored-torn-orders
+spec:
+  accessModes: ["ReadWriteOnce"]
+  storageClassName: csi-hostpath-sc
+  resources:
+    requests:
+      storage: 100Mi
+  dataSource:
+    apiGroup: snapshot.storage.k8s.io
+    kind: VolumeSnapshot
+    name: torn-orders-121152
+PVC
+# same PVC for the payments snapshot, then the verifier Job from the repo:
+# manifests/ledger/verify-job.tmpl.yaml
+```
+
+```console
+last order committed   : 108352
+last payment committed : 108377
+
+[FAIL] 25 payments have NO matching order.
+[FAIL] Each snapshot succeeded. The restore is still wrong.
+```
+
+Twenty five payments referencing orders that do not exist. Money moved for
+nothing. No component failed, every operation reported success, and the
+combined recovery point describes a moment in time that never existed. In
+production, that five second gap is simply your backup tool walking a list of
+a hundred PVCs one by one.
+
+The fix is a Kubernetes API that reached GA in 1.36 in April 2026:
+[VolumeGroupSnapshot](https://kubernetes.io/blog/2026/05/08/kubernetes-v1-36-volume-group-snapshot-ga/).
+One object selects the PVCs by label, and the CSI driver receives one request
+for a coordinated, crash consistent recovery point across all of them:
+
+```yaml
+apiVersion: groupsnapshot.storage.k8s.io/v1
+kind: VolumeGroupSnapshot
+metadata:
+  name: ledger-group-snap
+spec:
+  volumeGroupSnapshotClassName: csi-hostpath-groupsnapclass
+  source:
+    selector:
+      matchLabels:
+        group: ledger
+```
+
+![One group snapshot cuts both volumes at the same moment](/img/blog/a-backup-is-not-disaster-recovery/group-snapshot.svg)
+
+The group request produces managed member snapshots, one per volume, from a
+single point in time:
+
+```console
+volumegroupsnapshot.groupsnapshot.storage.k8s.io/group-121152 created
+volumegroupsnapshot.groupsnapshot.storage.k8s.io/group-121152 condition met
+
+NAME               GROUP          READY
+snapshot-63f367... group-121152   true
+snapshot-c3a36e... group-121152   true
+```
+
+The member snapshots are auto-named; find them through the group they belong
+to, then restore them exactly like the individual snapshots above and run the
+same verifier:
+
+```bash
+kubectl -n ledger get volumesnapshot \
+  -o custom-columns='NAME:.metadata.name,GROUP:.status.volumeGroupSnapshotName,READY:.status.readyToUse'
+# restore each member with a dataSource PVC, run the verifier again
+```
+
+```console
+last order committed   : 109169
+last payment committed : 109169
+
+[OK] Every payment has a matching order. Restore is consistent.
+```
+
+Full disclosure, the same one we gave on stage: our lab uses the CSI hostpath
+test driver, which implements the group RPCs but archives member volumes
+sequentially, so the writer is paused during the group snapshot to keep the
+demo deterministic. What the demo shows is the GA API and the restore
+workflow. The same point in time guarantee itself belongs to the storage
+backend of a production driver. Which leads to the practical takeaways:
+
+- Support is driver specific. A driver that supports ordinary
+  VolumeSnapshots proves nothing about group snapshots. Ask your storage
+  vendor whether they implement the CSI group RPCs. As of mid 2026, most of
+  the major cloud drivers we checked do not.
+- Setup is explicit: the CRDs and feature gates on the snapshot controller
+  and CSI sidecar are your job.
+- Crash consistent is not application consistent. The API removes cross
+  volume timing skew. It does not flush or quiesce your database.
+
+## What is still broken
+
+The demos expose gaps that no single tool closes today:
+
+1. **No common cross-cluster failover contract.** Data, workload, cluster,
+   traffic, and identity all have tools, and every row is missing the same
+   thing: a shared contract with the next row. Products answer this inside
+   their own APIs; core Kubernetes does not define the sequence.
+2. **No standard recovery unit for an application.** Core Kubernetes has no
+   maintained Application resource that says which objects, operators, data
+   services, and external dependencies must recover together. Velero uses
+   namespaces and labels, Argo has Applications, Helm has releases, and each
+   draws the boundary differently.
+3. **Backup success is treated as recovery proof.** Some teams test
+   resilience by deleting a pod and watching it return. That tests workload
+   reconciliation, not recovery. A recovery test restores into a clean
+   target, validates data and the user path, and measures the whole thing.
+
+There is community work here: the Cloud Native Business Continuity initiative
+proposal under CNCF TAG Operational Resilience (disclosure: Saiyam is one of
+the TAG chairs). It is an open proposal seeking contributors, aiming at a
+landscape gap analysis, updated backup and DR guidance, and reference
+architectures. If this post resonates, that is where to help:
+https://github.com/cncf/toc/issues/1779
+
+## From a laptop to production
+
+![Two independent failure domains](/img/blog/a-backup-is-not-disaster-recovery/prod-dr.svg)
+
+Every piece of the lab is a stand-in. The kiac cluster is your active region.
+The kind cluster is your recovery region, account, or second provider.
+SeaweedFS is a cross region S3 bucket or offsite object store. Gitea is the
+Git and infrastructure config that must survive the disaster. Powering off
+the VM is the region loss. The principles that transfer as-is: recovery
+points outside the failure domain, a recovery target that exists before the
+disaster, restore order, and application level validation. The exact failure
+modes and timings still need testing on your production platforms.
+
+## Now the questions are for you
+
+We ended the talk with questions instead of answers, and we will end this post
+the same way. Answer these honestly, ideally out loud in your next team
+meeting:
+
+1. Are you treating backup as DR? Backup answers "where is my data." DR
+   answers "how fast am I serving traffic again, and who does what at 2 AM."
+   If your entire DR plan fits inside a Velero schedule, you have a backup
+   plan.
+2. Who owns your Kubernetes DR, end to end, by name? Platform backs up the
+   cluster, the app team assumes the platform handles it, the DBA asks which
+   cluster. If ownership stops at team boundaries, nobody owns recovery.
+3. When did your team last restore a complete stateful application into a
+   clean cluster, validate the data, and time it? Restored, not just backed
+   up.
+4. Can you state RPO and RTO for your most critical application, and has a
+   stopwatch ever confirmed the RTO?
+5. Do your multi volume applications have one coordinated recovery point, or
+   a pile of individually perfect snapshots from different moments?
+6. If your region disappeared right now, do your Git repos, secrets, and
+   backups survive it, or do they live in the blast radius?
+
+If any answer made you uncomfortable, the fix is one rehearsal, not a new
+tool: pick one stateful app this month, restore it into a clean cluster,
+validate the data and the user path, and write down the time. Turn the guess
+into a number. That is the whole difference between having backups and having
+disaster recovery, and it is the difference between Fatima's night ending in
+panic or ending in a runbook she has already executed.
+
+The lab, the commands, and everything you saw here:
+https://github.com/saiyam1814/kubecon-japan-dr-demo
+
+We would genuinely like to hear what you do today: who owns DR at your
+company, and when did you last run a real restore? Tell us on
+[X](https://x.com/saiyampathak) or
+[LinkedIn](https://linkedin.com/in/saiyampathak).
diff --git a/public/img/blog/a-backup-is-not-disaster-recovery/gitops-trap.svg b/public/img/blog/a-backup-is-not-disaster-recovery/gitops-trap.svg
new file mode 100644
index 000000000..fa17cd7b3
--- /dev/null
+++ b/public/img/blog/a-backup-is-not-disaster-recovery/gitops-trap.svg
@@ -0,0 +1,59 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 940 360" style="font-family:'Chalkboard SE','Comic Sans MS',cursive">
+  <defs>
+    <pattern id="hred" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#FFC9C9" stroke-width="4"/>
+    </pattern>
+    <pattern id="hgreen" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D3F9D8" stroke-width="4"/>
+    </pattern>
+    <pattern id="hyellow" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#FFF3BF" stroke-width="4"/>
+    </pattern>
+  </defs>
+  <g stroke="#1e1e1e" stroke-width="2.4" fill="none" stroke-linecap="round" stroke-linejoin="round">
+
+    <!-- dead prod -->
+    <g transform="rotate(-0.5 130 150)">
+      <path d="M48 84 Q130 78 212 84 Q218 150 212 218 Q130 224 48 218 Q42 150 48 84 Z" fill="url(#hred)"/>
+    </g>
+    <path d="M62 96 Q130 160 200 208" stroke="#E03131" stroke-width="5"/>
+    <path d="M198 98 Q132 158 64 206" stroke="#E03131" stroke-width="5"/>
+
+    <!-- git cylinder top middle -->
+    <path d="M380 40 Q440 32 500 40 L497 74 Q440 82 383 74 Z" fill="url(#hyellow)"/>
+    <path d="M380 40 Q440 50 500 40"/>
+    <!-- arrow git -> recovery -->
+    <path d="M508 64 Q600 84 664 116" stroke="#1971C2" stroke-width="3.2"/>
+    <path d="M646 104 Q656 111 664 116 Q657 122 650 129" stroke="#1971C2" stroke-width="3.2"/>
+
+    <!-- recovery cluster -->
+    <g transform="rotate(0.4 780 170)">
+      <path d="M620 108 Q780 100 916 108 Q924 170 916 234 Q780 242 620 234 Q612 170 620 108 Z" fill="url(#hgreen)"/>
+    </g>
+    <!-- db cylinder inside, empty with ? -->
+    <path d="M726 142 Q772 135 818 142 L816 196 Q772 203 728 196 Z" fill="#fff"/>
+    <path d="M726 142 Q772 151 818 142"/>
+    <!-- green check top right of cluster -->
+    <path d="M856 76 Q866 90 872 100 Q888 68 912 50" stroke="#2F9E44" stroke-width="5"/>
+
+    <!-- vault bottom middle -->
+    <path d="M400 240 Q448 232 496 240 L484 330 Q448 338 414 330 Z" fill="url(#hyellow)"/>
+    <path d="M400 240 Q448 250 496 240"/>
+    <!-- arrow prod -> vault (backup happened before disaster, dashed grey) -->
+    <path d="M220 210 Q300 250 394 268" stroke="#868e96" stroke-width="3" stroke-dasharray="7 6"/>
+    <path d="M377 256 Q386 263 394 268 Q385 271 376 276" stroke="#868e96" stroke-width="3"/>
+    <!-- arrow vault -> recovery (the real restore) -->
+    <path d="M502 276 Q600 262 662 220" stroke="#E8590C" stroke-width="3.4"/>
+    <path d="M646 214 Q654 217 662 220 Q658 228 655 237" stroke="#E8590C" stroke-width="3.4"/>
+  </g>
+
+  <text x="130" y="252" text-anchor="middle" font-size="18" fill="#1e1e1e">prod · powered off</text>
+  <text x="440" y="24" text-anchor="middle" font-size="16" fill="#1e1e1e">Git · the declarations</text>
+  <text x="600" y="66" text-anchor="middle" font-size="14" fill="#1971C2">Argo syncs · green in 30s</text>
+  <text x="772" y="176" text-anchor="middle" font-size="26" fill="#E03131">?</text>
+  <text x="772" y="266" text-anchor="middle" font-size="15" fill="#E03131">running · and EMPTY</text>
+  <text x="780" y="290" text-anchor="middle" font-size="17" fill="#1e1e1e">recovery cluster</text>
+  <text x="448" y="354" text-anchor="middle" font-size="15" fill="#1e1e1e">S3 vault · the only place the data survived</text>
+  <text x="300" y="216" text-anchor="middle" font-size="13" fill="#868e96">backup · before the disaster</text>
+  <text x="600" y="300" text-anchor="middle" font-size="15" fill="#E8590C">Velero restore · the missing step</text>
+</svg>
diff --git a/public/img/blog/a-backup-is-not-disaster-recovery/group-snapshot.svg b/public/img/blog/a-backup-is-not-disaster-recovery/group-snapshot.svg
new file mode 100644
index 000000000..62b5e02a3
--- /dev/null
+++ b/public/img/blog/a-backup-is-not-disaster-recovery/group-snapshot.svg
@@ -0,0 +1,32 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 -14 860 264" style="font-family:'Chalkboard SE','Comic Sans MS',cursive">
+  <defs>
+    <pattern id="hblue" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D0EBFF" stroke-width="4"/>
+    </pattern>
+    <pattern id="hgreen" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D3F9D8" stroke-width="4"/>
+    </pattern>
+  </defs>
+  <g stroke="#1e1e1e" stroke-width="2.4" fill="none" stroke-linecap="round" stroke-linejoin="round">
+    <!-- lasso around both volumes -->
+    <path d="M46 26 Q120 18 196 26 Q206 120 196 216 Q120 226 48 216 Q38 120 46 26 Z" stroke="#E8590C" stroke-width="3" stroke-dasharray="10 8"/>
+    <!-- volumes -->
+    <path d="M74 48 Q120 41 166 48 L164 92 Q120 99 76 92 Z" fill="url(#hblue)"/>
+    <path d="M74 48 Q120 57 166 48"/>
+    <path d="M74 150 Q120 143 166 150 L164 194 Q120 201 76 194 Z" fill="url(#hgreen)"/>
+    <path d="M74 150 Q120 159 166 150"/>
+    <!-- write lanes -->
+    <path d="M212 70 Q440 64 700 69"/>
+    <path d="M212 172 Q440 166 700 171"/>
+    <!-- single cut across both -->
+    <line x1="480" y1="34" x2="480" y2="204" stroke-dasharray="6 6" stroke="#E8590C" stroke-width="3.4"/>
+    <!-- one camera -->
+    <path d="M454 6 Q480 2 506 6 L506 32 Q480 36 454 32 Z" fill="#FFD8A8"/>
+    <path d="M472 6 L476 -2 L490 -2 L494 6"/>
+    <!-- check -->
+    <path d="M726 96 Q740 116 750 130 Q772 84 806 56" stroke="#2F9E44" stroke-width="6"/>
+  </g>
+  <text x="120" y="242" text-anchor="middle" font-size="15" fill="#E8590C">selector: group=ledger</text>
+  <text x="560" y="30" text-anchor="start" font-size="16" fill="#E8590C">one cut · one moment · both volumes</text>
+  <text x="766" y="160" text-anchor="middle" font-size="16" fill="#2F9E44">crash-consistent</text>
+</svg>
diff --git a/public/img/blog/a-backup-is-not-disaster-recovery/lab-setup.svg b/public/img/blog/a-backup-is-not-disaster-recovery/lab-setup.svg
new file mode 100644
index 000000000..bf424cd08
--- /dev/null
+++ b/public/img/blog/a-backup-is-not-disaster-recovery/lab-setup.svg
@@ -0,0 +1,81 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 940 430" style="font-family:'Chalkboard SE','Comic Sans MS',cursive">
+  <defs>
+    <pattern id="hblue" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D0EBFF" stroke-width="4"/>
+    </pattern>
+    <pattern id="hgreen" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D3F9D8" stroke-width="4"/>
+    </pattern>
+    <pattern id="hyellow" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#FFF3BF" stroke-width="4"/>
+    </pattern>
+  </defs>
+  <g stroke="#1e1e1e" stroke-width="2.4" fill="none" stroke-linecap="round" stroke-linejoin="round">
+
+    <!-- Gitea, top center -->
+    <path d="M400 30 Q470 22 540 30 L536 66 Q470 74 404 66 Z" fill="url(#hyellow)"/>
+    <path d="M400 30 Q470 40 540 30"/>
+    <!-- arrow gitea -> DR (argo pulls) -->
+    <path d="M548 58 Q640 76 700 108" stroke="#1971C2" stroke-width="3"/>
+    <path d="M682 96 Q692 103 700 108 Q693 113 686 120" stroke="#1971C2" stroke-width="3"/>
+
+    <!-- PROD: kiac, VM wall drawn as double border -->
+    <g transform="rotate(-0.4 190 220)">
+      <path d="M52 118 Q190 110 328 118 Q336 220 328 322 Q190 330 52 322 Q44 220 52 118 Z" fill="url(#hblue)"/>
+      <path d="M64 130 Q190 123 316 130 Q323 220 316 310 Q190 317 64 310 Q57 220 64 130 Z" stroke-dasharray="3 6"/>
+    </g>
+    <!-- postgres cylinder + pv -->
+    <path d="M120 176 Q166 169 212 176 L210 224 Q166 231 122 224 Z" fill="#fff"/>
+    <path d="M120 176 Q166 185 212 176"/>
+    <!-- velero sail tag -->
+    <path d="M236 168 L236 226 Q262 214 288 226 L288 168 Q262 156 236 168 Z" fill="#fff" transform="rotate(2 262 196)"/>
+    <path d="M252 184 Q262 178 274 184" stroke-width="2"/>
+    <path d="M250 200 Q262 194 276 200" stroke-width="2"/>
+
+    <!-- DR: kind -->
+    <g transform="rotate(0.4 750 220)">
+      <path d="M612 118 Q750 110 888 118 Q896 220 888 322 Q750 330 612 322 Q604 220 612 118 Z" fill="url(#hgreen)"/>
+    </g>
+    <!-- argo octopus-ish head -->
+    <path d="M652 170 Q676 158 698 170 Q706 188 692 198 Q676 206 660 198 Q646 188 652 170 Z" fill="#fff"/>
+    <path d="M660 202 L656 216 M676 205 L676 219 M692 202 L696 216" stroke-width="2"/>
+    <!-- ledger: two small cylinders -->
+    <path d="M740 176 Q766 172 792 176 L790 206 Q766 210 742 206 Z" fill="#fff"/>
+    <path d="M740 176 Q766 182 792 176"/>
+    <path d="M810 176 Q836 172 862 176 L860 206 Q836 210 812 206 Z" fill="#fff"/>
+    <path d="M810 176 Q836 182 862 176"/>
+    <!-- velero sail tag DR -->
+    <path d="M700 242 L700 296 Q724 285 748 296 L748 242 Q724 231 700 242 Z" fill="#fff" transform="rotate(-2 724 268)"/>
+    <path d="M714 258 Q724 252 736 258" stroke-width="2"/>
+    <path d="M712 274 Q724 268 738 274" stroke-width="2"/>
+
+    <!-- vault, bottom center -->
+    <path d="M424 296 Q470 288 516 296 L504 392 Q470 400 436 392 Z" fill="url(#hyellow)"/>
+    <path d="M424 296 Q470 306 516 296"/>
+    <!-- backup arrow prod -> vault (through the VM wall) -->
+    <path d="M334 268 Q380 296 418 320" stroke="#E8590C" stroke-width="3.2"/>
+    <path d="M402 306 Q410 314 418 320 Q408 322 399 327" stroke="#E8590C" stroke-width="3.2"/>
+    <!-- restore arrow vault -> DR (dashed) -->
+    <path d="M522 320 Q560 296 606 268" stroke="#E8590C" stroke-width="3.2" stroke-dasharray="8 7"/>
+    <path d="M588 268 Q598 267 606 268 Q601 276 598 285" stroke="#E8590C" stroke-width="3.2"/>
+  </g>
+
+  <text x="470" y="54" text-anchor="middle" font-size="17" fill="#1e1e1e">Gitea · app YAML in Git</text>
+  <text x="648" y="66" text-anchor="middle" font-size="13" fill="#1971C2">Argo pulls</text>
+
+  <text x="166" y="256" text-anchor="middle" font-size="15" fill="#1e1e1e">Postgres + PV</text>
+  <text x="264" y="256" text-anchor="middle" font-size="15" fill="#1e1e1e">Velero</text>
+  <text x="190" y="298" text-anchor="middle" font-size="15" fill="#5f5b53">+ CSI snapshots · VGS</text>
+  <text x="190" y="356" text-anchor="middle" font-size="19" fill="#1e1e1e">prod · kiac</text>
+  <text x="190" y="378" text-anchor="middle" font-size="14" fill="#5f5b53">the node is its own VM</text>
+
+  <text x="676" y="232" text-anchor="middle" font-size="15" fill="#1e1e1e">Argo CD</text>
+  <text x="800" y="232" text-anchor="middle" font-size="15" fill="#1e1e1e">ledger · 2 PVCs</text>
+  <text x="724" y="318" text-anchor="middle" font-size="15" fill="#1e1e1e">Velero</text>
+  <text x="750" y="356" text-anchor="middle" font-size="19" fill="#1e1e1e">recovery · kind</text>
+  <text x="750" y="378" text-anchor="middle" font-size="14" fill="#5f5b53">+ CSI snapshots · VGS</text>
+
+  <text x="370" y="308" text-anchor="end" font-size="14" fill="#E8590C">backup</text>
+  <text x="572" y="308" text-anchor="start" font-size="14" fill="#E8590C">restore</text>
+  <text x="470" y="418" text-anchor="middle" font-size="16" fill="#1e1e1e">SeaweedFS · S3 vault outside both clusters</text>
+</svg>
diff --git a/public/img/blog/a-backup-is-not-disaster-recovery/layers.svg b/public/img/blog/a-backup-is-not-disaster-recovery/layers.svg
new file mode 100644
index 000000000..a34c0fcb5
--- /dev/null
+++ b/public/img/blog/a-backup-is-not-disaster-recovery/layers.svg
@@ -0,0 +1,48 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 880 330" style="font-family:'Chalkboard SE','Comic Sans MS',cursive">
+  <defs>
+    <pattern id="hviolet" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#E5DBFF" stroke-width="4"/>
+    </pattern>
+    <pattern id="hblue" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D0EBFF" stroke-width="4"/>
+    </pattern>
+    <pattern id="hgreen" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D3F9D8" stroke-width="4"/>
+    </pattern>
+    <pattern id="hyellow" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#FFF3BF" stroke-width="4"/>
+    </pattern>
+  </defs>
+  <g stroke="#1e1e1e" stroke-width="2.4" fill="none" stroke-linecap="round" stroke-linejoin="round">
+    <g transform="rotate(-0.5 340 69)">
+      <path d="M64 44 Q340 38 616 44 Q621 68 616 94 Q340 100 64 94 Q59 68 64 44 Z" fill="url(#hviolet)"/>
+    </g>
+    <g transform="rotate(0.4 340 139)">
+      <path d="M58 114 Q340 109 622 114 Q626 138 622 164 Q340 169 58 164 Q54 138 58 114 Z" fill="url(#hblue)"/>
+    </g>
+    <g transform="rotate(-0.35 340 209)">
+      <path d="M66 184 Q340 179 614 184 Q619 208 614 234 Q340 240 66 234 Q61 208 66 184 Z" fill="url(#hgreen)"/>
+    </g>
+    <g transform="rotate(0.5 340 279)">
+      <path d="M60 254 Q340 249 620 254 Q625 278 620 304 Q340 310 60 304 Q55 278 60 254 Z" fill="url(#hyellow)"/>
+    </g>
+    <!-- arrows pointing at the joins -->
+    <path d="M760 128 Q700 118 648 104" stroke="#E8590C" stroke-width="3"/>
+    <path d="M664 96 Q656 100 648 104 Q652 111 657 118" stroke="#E8590C" stroke-width="3"/>
+    <path d="M760 176 Q700 176 648 174" stroke="#E8590C" stroke-width="3"/>
+    <path d="M664 166 Q656 170 648 174 Q655 179 663 183" stroke="#E8590C" stroke-width="3"/>
+    <path d="M760 224 Q700 234 648 244" stroke="#E8590C" stroke-width="3"/>
+    <path d="M662 232 Q655 238 648 244 Q655 249 664 253" stroke="#E8590C" stroke-width="3"/>
+  </g>
+  <text x="86" y="78" font-size="21" fill="#1e1e1e">1 · cluster state</text>
+  <text x="300" y="78" font-size="16" fill="#5f5b53">etcd, API objects, the control plane itself</text>
+  <text x="86" y="148" font-size="21" fill="#1e1e1e">2 · app definitions</text>
+  <text x="320" y="148" font-size="16" fill="#5f5b53">YAML, Helm charts, the Git repo</text>
+  <text x="86" y="218" font-size="21" fill="#1e1e1e">3 · persistent data</text>
+  <text x="320" y="218" font-size="16" fill="#5f5b53">PVs, databases, object stores</text>
+  <text x="86" y="288" font-size="21" fill="#1e1e1e">4 · access &amp; traffic</text>
+  <text x="316" y="288" font-size="16" fill="#5f5b53">secrets, certs, DNS, external services</text>
+  <text x="775" y="168" font-size="21" fill="#E8590C">recovery</text>
+  <text x="775" y="192" font-size="21" fill="#E8590C">fails at</text>
+  <text x="775" y="216" font-size="21" fill="#E8590C">the joins</text>
+</svg>
diff --git a/public/img/blog/a-backup-is-not-disaster-recovery/prod-dr.svg b/public/img/blog/a-backup-is-not-disaster-recovery/prod-dr.svg
new file mode 100644
index 000000000..1f2216cff
--- /dev/null
+++ b/public/img/blog/a-backup-is-not-disaster-recovery/prod-dr.svg
@@ -0,0 +1,51 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 940 340" style="font-family:'Chalkboard SE','Comic Sans MS',cursive">
+  <defs>
+    <pattern id="hblue" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D0EBFF" stroke-width="4"/>
+    </pattern>
+    <pattern id="hgreen" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D3F9D8" stroke-width="4"/>
+    </pattern>
+    <pattern id="hyellow" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#FFF3BF" stroke-width="4"/>
+    </pattern>
+  </defs>
+  <g stroke="#1e1e1e" stroke-width="2.4" fill="none" stroke-linecap="round" stroke-linejoin="round">
+    <!-- git/iac/secrets outside both -->
+    <path d="M336 22 Q460 16 586 22 Q590 52 586 84 Q460 90 336 84 Q332 52 336 22 Z" fill="url(#hyellow)"/>
+    <path d="M330 88 Q250 104 210 128" stroke="#1971C2"/>
+    <path d="M228 116 Q218 123 210 128 Q216 134 224 141" stroke="#1971C2"/>
+    <path d="M592 88 Q672 104 712 128" stroke="#1971C2"/>
+    <path d="M694 116 Q704 123 712 128 Q706 134 698 141" stroke="#1971C2"/>
+    <!-- Site A -->
+    <path d="M76 134 Q192 128 308 134 Q313 200 308 268 Q192 274 76 268 Q71 200 76 134 Z" fill="url(#hblue)"/>
+    <!-- Site B -->
+    <path d="M612 134 Q728 128 844 134 Q849 200 844 268 Q728 274 612 268 Q607 200 612 134 Z" fill="url(#hgreen)"/>
+    <!-- db replication -->
+    <path d="M322 172 Q460 156 598 172" stroke="#E8590C" stroke-width="3.2"/>
+    <path d="M580 160 Q590 167 598 172 Q591 178 584 185" stroke="#E8590C" stroke-width="3.2"/>
+    <!-- backup vault outside A -->
+    <path d="M410 244 Q458 238 506 244 L494 316 Q458 324 424 316 Z" fill="url(#hyellow)"/>
+    <path d="M410 244 Q458 254 506 244"/>
+    <path d="M312 258 Q356 268 402 274" stroke="#E8590C" stroke-width="3"/>
+    <path d="M386 262 Q394 269 402 274 Q393 278 384 283" stroke="#E8590C" stroke-width="3"/>
+    <path d="M514 274 Q560 268 604 258" stroke="#E8590C" stroke-width="3" stroke-dasharray="8 7"/>
+    <path d="M588 250 Q596 254 604 258 Q598 264 593 271" stroke="#E8590C" stroke-width="3"/>
+    <!-- users -->
+    <path d="M898 158 Q908 154 912 162 Q916 172 906 176 Q896 178 894 168 Q893 160 898 158 Z"/>
+    <line x1="903" y1="178" x2="903" y2="206"/>
+    <line x1="903" y1="186" x2="888" y2="198"/><line x1="903" y1="186" x2="918" y2="198"/>
+    <line x1="903" y1="206" x2="890" y2="226"/><line x1="903" y1="206" x2="916" y2="226"/>
+    <path d="M884 200 Q862 196 850 190" stroke="#2F9E44" stroke-width="3" stroke-dasharray="7 6"/>
+  </g>
+  <text x="460" y="48" text-anchor="middle" font-size="19" fill="#1e1e1e">Git · IaC · secrets</text>
+  <text x="460" y="72" text-anchor="middle" font-size="14" fill="#5f5b53">outside both sites</text>
+  <text x="192" y="298" text-anchor="middle" font-size="18" fill="#1e1e1e">Site A · active</text>
+  <text x="728" y="298" text-anchor="middle" font-size="18" fill="#1e1e1e">Site B · warm</text>
+  <text x="460" y="146" text-anchor="middle" font-size="15" fill="#E8590C">database replication</text>
+  <text x="368" y="302" text-anchor="end" font-size="14" fill="#E8590C">backup</text>
+  <text x="548" y="302" text-anchor="start" font-size="14" fill="#E8590C">restore</text>
+  <text x="458" y="336" text-anchor="middle" font-size="15" fill="#1e1e1e">recovery vault · outside Site A</text>
+  <text x="884" y="242" text-anchor="middle" font-size="13" fill="#2F9E44">traffic, after</text>
+  <text x="884" y="260" text-anchor="middle" font-size="13" fill="#2F9E44">validation</text>
+</svg>
diff --git a/public/img/blog/a-backup-is-not-disaster-recovery/rpo-rto.svg b/public/img/blog/a-backup-is-not-disaster-recovery/rpo-rto.svg
new file mode 100644
index 000000000..aa131d6fe
--- /dev/null
+++ b/public/img/blog/a-backup-is-not-disaster-recovery/rpo-rto.svg
@@ -0,0 +1,26 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 900 280" style="font-family:'Chalkboard SE','Comic Sans MS',cursive">
+  <g stroke="#1e1e1e" stroke-width="2.4" fill="none" stroke-linecap="round" stroke-linejoin="round">
+    <!-- timeline -->
+    <path d="M55 162 Q280 156 480 161 Q680 165 845 159"/>
+    <path d="M826 150 Q836 155 845 159 Q834 163 825 169"/>
+    <!-- camera: last good recovery point -->
+    <path d="M196 122 Q220 118 244 122 L244 150 Q220 154 196 150 Z" fill="#A5D8FF"/>
+    <path d="M212 122 L216 112 L228 112 L232 122"/>
+    <path d="M214 136 Q220 130 227 135 Q230 142 222 145 Q214 144 214 136 Z" fill="#fff"/>
+    <line x1="220" y1="152" x2="220" y2="162"/>
+    <!-- explosion at disaster -->
+    <path d="M470 118 L479 136 L500 130 L488 147 L506 158 L484 160 L487 180 L470 166 L453 180 L456 160 L434 158 L452 147 L440 130 L461 136 Z" fill="#FFC9C9" stroke="#E03131"/>
+    <!-- flag: serving again -->
+    <line x1="760" y1="112" x2="760" y2="160"/>
+    <path d="M760 112 Q784 118 806 112 L804 134 Q782 140 760 134 Z" fill="#B2F2BB" stroke="#2F9E44"/>
+    <!-- RPO bracket -->
+    <path d="M220 108 L220 92 Q345 84 470 92 L470 108" stroke="#E8590C"/>
+    <!-- RTO bracket -->
+    <path d="M470 214 L470 230 Q615 238 760 230 L760 214" stroke="#E8590C"/>
+  </g>
+  <text x="220" y="196" text-anchor="middle" font-size="17" fill="#1e1e1e">last good recovery point</text>
+  <text x="470" y="216" text-anchor="middle" font-size="17" fill="#E03131">02:07 · disaster</text>
+  <text x="760" y="196" text-anchor="middle" font-size="17" fill="#1e1e1e">serving users again</text>
+  <text x="345" y="72" text-anchor="middle" font-size="20" fill="#E8590C">RPO · the data you lose</text>
+  <text x="615" y="262" text-anchor="middle" font-size="20" fill="#E8590C">RTO · the time you are down</text>
+</svg>
diff --git a/public/img/blog/a-backup-is-not-disaster-recovery/torn-snapshots.svg b/public/img/blog/a-backup-is-not-disaster-recovery/torn-snapshots.svg
new file mode 100644
index 000000000..444b28278
--- /dev/null
+++ b/public/img/blog/a-backup-is-not-disaster-recovery/torn-snapshots.svg
@@ -0,0 +1,40 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 880 310" style="font-family:'Chalkboard SE','Comic Sans MS',cursive">
+  <defs>
+    <pattern id="hblue" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D0EBFF" stroke-width="4"/>
+    </pattern>
+    <pattern id="hgreen" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D3F9D8" stroke-width="4"/>
+    </pattern>
+    <pattern id="hredx" width="8" height="8" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="8" stroke="#FFA8A8" stroke-width="3.4"/>
+    </pattern>
+  </defs>
+  <g stroke="#1e1e1e" stroke-width="2.4" fill="none" stroke-linecap="round" stroke-linejoin="round">
+    <!-- orders volume + write lane -->
+    <path d="M64 70 Q114 62 164 70 L162 120 Q114 128 66 120 Z" fill="url(#hblue)"/>
+    <path d="M64 70 Q114 80 164 70"/>
+    <path d="M180 96 Q480 90 810 95"/>
+    <line x1="240" y1="88" x2="240" y2="104"/><line x1="300" y1="88" x2="300" y2="104"/><line x1="360" y1="88" x2="360" y2="104"/><line x1="470" y1="88" x2="470" y2="104"/><line x1="530" y1="88" x2="530" y2="104"/><line x1="590" y1="88" x2="590" y2="104"/><line x1="650" y1="88" x2="650" y2="104"/><line x1="710" y1="88" x2="710" y2="104"/>
+    <!-- payments volume + write lane -->
+    <path d="M64 190 Q114 182 164 190 L162 240 Q114 248 66 240 Z" fill="url(#hgreen)"/>
+    <path d="M64 190 Q114 200 164 190"/>
+    <path d="M180 216 Q480 210 810 215"/>
+    <line x1="240" y1="208" x2="240" y2="224"/><line x1="300" y1="208" x2="300" y2="224"/><line x1="360" y1="208" x2="360" y2="224"/><line x1="470" y1="208" x2="470" y2="224"/><line x1="530" y1="208" x2="530" y2="224"/><line x1="590" y1="208" x2="590" y2="224"/><line x1="650" y1="208" x2="650" y2="224"/><line x1="710" y1="208" x2="710" y2="224"/>
+    <!-- camera snap on orders at t=0 -->
+    <path d="M382 30 Q406 26 430 30 L430 58 Q406 62 382 58 Z" fill="#A5D8FF"/>
+    <path d="M398 30 L402 20 L414 20 L418 30"/>
+    <line x1="406" y1="64" x2="406" y2="112" stroke-dasharray="6 6" stroke="#1971C2" stroke-width="3"/>
+    <!-- camera snap on payments at t=5 -->
+    <path d="M596 262 Q620 258 644 262 L644 290 Q620 294 596 290 Z" fill="#B2F2BB"/>
+    <path d="M612 262 L616 252 L628 252 L632 262"/>
+    <line x1="620" y1="196" x2="620" y2="256" stroke-dasharray="6 6" stroke="#2F9E44" stroke-width="3"/>
+    <!-- the skew window on the payments lane -->
+    <path d="M408 200 L618 200 L618 232 L408 232 Z" fill="url(#hredx)" stroke="#E03131" stroke-dasharray="5 5"/>
+  </g>
+  <text x="114" y="52" text-anchor="middle" font-size="17" fill="#1e1e1e">orders PV</text>
+  <text x="114" y="270" text-anchor="middle" font-size="17" fill="#1e1e1e">payments PV</text>
+  <text x="330" y="44" text-anchor="end" font-size="16" fill="#1971C2">snapshot A · t = 0</text>
+  <text x="700" y="280" text-anchor="start" font-size="16" fill="#2F9E44">snapshot B · t = 5s</text>
+  <text x="513" y="180" text-anchor="middle" font-size="17" fill="#E03131">5 seconds of payments with no matching order</text>
+</svg>
diff --git a/public/img/blog/a-backup-is-not-disaster-recovery/velero-flow.svg b/public/img/blog/a-backup-is-not-disaster-recovery/velero-flow.svg
new file mode 100644
index 000000000..b9481c380
--- /dev/null
+++ b/public/img/blog/a-backup-is-not-disaster-recovery/velero-flow.svg
@@ -0,0 +1,38 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 880 250" style="font-family:'Chalkboard SE','Comic Sans MS',cursive">
+  <defs>
+    <pattern id="hblue" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D0EBFF" stroke-width="4"/>
+    </pattern>
+    <pattern id="hyellow" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#FFF3BF" stroke-width="4"/>
+    </pattern>
+    <pattern id="hgreen" width="9" height="9" patternUnits="userSpaceOnUse" patternTransform="rotate(-45)">
+      <line x1="0" y1="0" x2="0" y2="9" stroke="#D3F9D8" stroke-width="4"/>
+    </pattern>
+  </defs>
+  <g stroke="#1e1e1e" stroke-width="2.4" fill="none" stroke-linecap="round" stroke-linejoin="round">
+    <!-- source cluster -->
+    <path d="M62 62 Q160 56 258 62 Q263 125 258 190 Q160 196 62 190 Q57 125 62 62 Z" fill="url(#hblue)"/>
+    <rect x="86" y="84" width="34" height="30" rx="5" fill="#fff" transform="rotate(-1.6 103 99)"/>
+    <rect x="132" y="82" width="34" height="30" rx="5" fill="#fff" transform="rotate(1.4 149 97)"/>
+    <path d="M96 136 Q130 128 164 136 L162 166 Q130 174 98 166 Z" fill="#fff"/>
+    <path d="M96 136 Q130 144 164 136"/>
+    <!-- backup arrow -->
+    <path d="M272 122 Q322 112 372 122" stroke="#E8590C" stroke-width="3.2"/>
+    <path d="M354 108 Q364 116 372 122 Q361 126 351 132" stroke="#E8590C" stroke-width="3.2"/>
+    <!-- bucket -->
+    <path d="M394 84 Q444 78 494 84 L480 192 Q444 200 408 192 Z" fill="url(#hyellow)"/>
+    <path d="M394 84 Q444 94 494 84"/>
+    <!-- restore arrow (dashed) -->
+    <path d="M510 122 Q560 112 610 122" stroke="#E8590C" stroke-width="3.2" stroke-dasharray="8 7"/>
+    <path d="M592 108 Q602 116 610 122 Q599 126 589 132" stroke="#E8590C" stroke-width="3.2"/>
+    <!-- target cluster (dashed = does not exist yet) -->
+    <path d="M628 62 Q726 56 824 62 Q829 125 824 190 Q726 196 628 190 Q623 125 628 62 Z" fill="url(#hgreen)" stroke-dasharray="9 7"/>
+    <path d="M700 116 Q712 132 720 142 Q740 104 766 84" stroke="#2F9E44" stroke-width="5"/>
+  </g>
+  <text x="160" y="218" text-anchor="middle" font-size="18" fill="#1e1e1e">source cluster</text>
+  <text x="322" y="94" text-anchor="middle" font-size="16" fill="#E8590C">objects + PV bytes</text>
+  <text x="444" y="224" text-anchor="middle" font-size="18" fill="#1e1e1e">any S3 vault</text>
+  <text x="560" y="94" text-anchor="middle" font-size="16" fill="#E8590C">restore</text>
+  <text x="726" y="218" text-anchor="middle" font-size="18" fill="#1e1e1e">a different cluster</text>
+</svg>