diff --git a/.github/prompts/review-docs.prompt.md b/.github/prompts/review-docs.prompt.md
index 2a4569e9..c39a478e 100644
--- a/.github/prompts/review-docs.prompt.md
+++ b/.github/prompts/review-docs.prompt.md
@@ -13,6 +13,7 @@ Review the documentation for clarity, completeness, and accuracy.
 - H1 titles under `docs/how-to` should start with "How to".
 - Section headers and index entries should all use sentence case (not title case).
 - Known product names should be capitalized consistently throughout the documentation.
+- Spelling according to US English conventions.
 
 ## Context
 
diff --git a/docs/how-to/configure-and-tune/evaluate-telemetry-volume.md b/docs/how-to/configure-and-tune/evaluate-telemetry-volume.md
index cb657f97..38ffe4e0 100644
--- a/docs/how-to/configure-and-tune/evaluate-telemetry-volume.md
+++ b/docs/how-to/configure-and-tune/evaluate-telemetry-volume.md
@@ -11,6 +11,21 @@ In order to correctly size the VM(s) needed for COS, you need to know how much t
 
 
 ## Metrics rate
+
+### Manual evaluation
+Find out the metrics endpoint manifest for each observed workload. If it is not documented,
+you will need to manually count the number of non-comment lines served on the metrics endpoint,
+for example:
+
+```bash
+curl -sf localhost:8080/metrics | grep -v "^# " | wc -l
+```
+
+This will give you the number of timeseries that will be created for the workload, per unit.
+
+Another option is to deploy a temporary pilot Prometheus charm.
+
+### With charmed Prometheus
 Have your deployment sending all metrics to Prometheus (or Mimir) and inspect the 48hr plot for `count({__name__=~".+"})`.
 The raw data can also be obtained by querying the Prometheus `query` endpoint directly:
 
@@ -36,8 +51,13 @@ load[load generator] ---|db| postgresql
 postgresql ---|metrics-endpoint| prometheus
 ```
 
-
 ## Logs rate
+### Manual evaluation
+The most reliable way to evaluate the logging rate of a workload is with load tests.
+
+Another option is to deploy temporary pilot Loki and Prometheus charms.
+
+### With charmed Loki and Prometheus
 Have your deployment sending all logs to Loki, and inspect the 48hr plot for `loki_distributor_*_received_total`:
 
 ```
diff --git a/docs/how-to/deploy-and-manage/index.md b/docs/how-to/deploy-and-manage/index.md
index 339dbbda..0777d304 100644
--- a/docs/how-to/deploy-and-manage/index.md
+++ b/docs/how-to/deploy-and-manage/index.md
@@ -14,14 +14,10 @@ These guides cover deploying, upgrading, managing, and securing access to COS.
 
 See our [tutorials](/tutorial/index) for guidance on deploying COS.
 
-## Upgrades
-
-Move between COS revisions with confidence.
-
 ```{toctree}
 :maxdepth: 1
 
-Cross-track upgrade instructions <upgrade>
+Install <install>
 ```
 
 ## Secure access
@@ -34,3 +30,14 @@ Protect and expose COS endpoints for production traffic.
 Configure TLS encryption <configure-tls-encryption>
 Configure ingress <configure-granular-ingress>
 ```
+
+## Upgrades
+
+Move between COS revisions with confidence.
+
+```{toctree}
+:maxdepth: 1
+
+Cross-track upgrade instructions <upgrade>
+```
+
diff --git a/docs/how-to/deploy-and-manage/install.md b/docs/how-to/deploy-and-manage/install.md
new file mode 100644
index 00000000..a14a7176
--- /dev/null
+++ b/docs/how-to/deploy-and-manage/install.md
@@ -0,0 +1,83 @@
+---
+myst:
+ html_meta:
+   description: "Install the Canonical Observability Stack: preparation checklist covering sizing, networking, storage, and deployment options."
+---
+
+# How to install COS
+
+## Preparation
+
+Before deploying COS or COS Lite, work through the items below.
+
+### COS flavor
+
+The [flavor of COS](/explanation/overview/what-is-cos) to install depends on your use-case.
+If you want to install on edge devices, then COS Lite is likely the right choice; otherwise
+you should probably go with "full" COS.
+
+```{mermaid}
+graph LR
+
+subgraph env["Monitored environment"]
+opentelemetry-collector
+end
+
+subgraph k8s["K8s cluster"]
+COS
+end
+
+subgraph pc["Public cloud"]
+cos-alerter["COS Alerter"]
+end
+
+subgraph storage["Storage cluster"]
+S3
+end
+
+opentelemetry-collector ---|telemetry| COS
+COS --- S3
+COS --- cos-alerter
+```
+
+### Kubernetes cluster
+
+Deploy COS on a high-availability Kubernetes cluster with at least 3 control plane nodes.
+
+### Sizing
+
+Use the [sizing guide](/reference/system-requirements) to determine the minimum hardware for your deployment.
+If you don't yet know how much telemetry your workloads generate, start with [How to evaluate telemetry volume](/how-to/configure-and-tune/evaluate-telemetry-volume).
+
+Follow the [storage best practices](/reference/storage) to set up a distributed storage backend with a replication factor of 3.
+Do **not** use `hostPath` storage in production.
+
+### Configure networking
+
+Review the [networking best practices](/reference/networking) and ensure:
+
+- A load balancer (for example, MetalLB) is available to give Traefik a stable IP.
+- Egress is open for Charmhub, the Juju OCI registry, and Snapcraft.
+
+### Plan for TLS
+
+Production deployments should use TLS.
+See [How to configure TLS encryption](/how-to/deploy-and-manage/configure-tls-encryption) for the available modes and what you need to prepare (for example, an external certificates provider).
+
+### Authentication and authorization
+Only the Grafana and Traefik charms support auth.
+For exposing Grafana publicly, use two Traefik charms, one for internal connections, and another for external access, which will provide ingress to Grafana.
+
+### Dedicated Juju controller and model
+
+You should bootstrap a dedicated Juju controller and model just for COS.
+
+## Create Terraform plan
+
+```hcl
+module ... { ... }
+```
+
+## Deploy COS Alerter
+
+COS Alerter is a watchdog service for COS you should deploy on a physically different cloud.
\ No newline at end of file