amadan

carlos / platform Public

Merge pull request #150 from carlosframework/rails-behind-edge-poc-spec

docs: Rails-behind-the-edge PoC — design record + implementation plan
Paul Campbell pushed by paul@keymail.dev 2ca7d60fd10b437055efe5e1626a66c3b19ad41e parent eb69ae6parent 13d7f57
2 files changed, +1079 −0
  • docs/superpowers/plans/2026-08-11-rails-behind-carlos-edge-poc.md +795 −0
  • docs/superpowers/specs/2026-08-11-rails-behind-carlos-edge-poc-design.md +284 −0
diff --git a/docs/superpowers/plans/2026-08-11-rails-behind-carlos-edge-poc.md b/docs/superpowers/plans/2026-08-11-rails-behind-carlos-edge-poc.md
new file mode 100644
index 000000000..760671678
--- /dev/null
+++ b/docs/superpowers/plans/2026-08-11-rails-behind-carlos-edge-poc.md
@@ -0,0 +1,795 @@
+# Rails behind the CARLOS edge (Classic/Pro staging PoC) Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Serve Tito Classic and Tito Pro staging — full process tree, real staging dependencies — at `classic.platform.tito.io` / `pro.platform.tito.io` through the CARLOS edge, from two tofu-provisioned boxes running the Buildkite amd64 images.
+
+**Architecture:** Two x86 EC2 boxes in the `tito-tito-staging` VPC (native SG access to the staging Aurora/Valkey/Memcached), a new default↔staging VPC peering with `/32` routes for just the two box IPs (the default VPC already routes all of `10.0.0.0/16` to the *production* peering, so only host routes are safe), docker + systemd units running the k8s images, env from SSM Parameter Store, deploys over SSM RunCommand, and two `carlos add -kind service -addr` routes on the IE edge.
+
+**Tech Stack:** OpenTofu (`platform-infrastructure/tito/`), cloud-init, systemd, docker, AWS SSM, the `carlos` CLI on the IE box.
+
+**Spec:** `docs/superpowers/specs/2026-08-11-rails-behind-carlos-edge-poc-design.md` (this repo).
+
+## Global Constraints
+
+- AWS account `665106592142`, region `eu-west-1`, CLI profile `titogo`; tofu state `s3://carlos-tfstate-665106592142/tito/`.
+- **Targeted plans/applies only** in `platform-infrastructure/tito/` — the tree carries drift from open PRs and a full apply would strip live tcp/25 SMTP ingress (recorded in `tito/object-store.tf`'s commit message).
+- No SSH anywhere. All box operation via SSM (`aws ssm send-command` / session-manager).
+- Images are the Buildkite amd64 builds, **unmodified** — no rebuilds.
+- Secrets never in tofu, the repo, or terminal output — SSM Parameter Store `SecureString` only.
+- Never set `-hibernate` on the edge routes; the edge registry is `/data/registry.db`, written by `carlos add` run **as user `carlos`** on the IE box.
+- Staging only is applied; production tofu is written gated off (`rails_poc_prod_enabled = false`) and NOT applied.
+- Both repos use worktree discipline: infra work on a branch in a `platform-infrastructure` worktree; this plan/spec live on branch `rails-behind-edge-poc-spec` in `platform`.
+
+## Facts (preflighted 2026-08-11 — verify only if something disagrees)
+
+| Fact | Value |
+|---|---|
+| Default VPC (edge lives here) | `vpc-02416de3913753989`, `172.31.0.0/16`, main rtb `rtb-08863e575439a6377` |
+| IE edge box | `i-025a0b70f3d93304d`, private IP `172.31.4.137`, SG `carlos-tito-ie` (`module.host_ie.security_group_id`) |
+| Staging VPC | `vpc-03be5678d08549203` (`tito-tito-staging`), `10.0.0.0/16`; subnets `10.0.0.0/23`, `10.0.2.0/23`, `10.0.4.0/23`; igw rtb `rtb-041f57e000fb5797e`, main rtb `rtb-0a01ed1a14d1a52ce` |
+| Production VPC | `vpc-90dffef5`, also `10.0.0.0/16`; **existing peering** `pcx-056b55b69cbf89143`; default VPC routes `10.0.0.0/16 → pcx-056b…` already |
+| Classic staging DB | `tito-classic-mysql-staging-writer.cefg7n9rzaz0.eu-west-1.rds.amazonaws.com` (aurora-**mysql**, 3306) |
+| Pro staging DB | `tito-pro-postgres-staging-writer.cefg7n9rzaz0.eu-west-1.rds.amazonaws.com` (aurora-**postgresql**, 5432) |
+| Staging caches | `tito-classic-memcached-staging` (11211), `tito-classic-redis-staging-001` (valkey, 6379), `tito-pro-memcached-staging`, `tito-pro-redis-staging-001` |
+| Edge env | `PLATFORM_APPS_DOMAIN=platform.tito.io`, `CARLOS_WILDCARD_DOMAINS=go.tito.io` → `classic.`/`pro.platform.tito.io` get per-host HTTP-01 certs automatically once routed |
+| DNS | `*.platform.tito.io` A → IE EIP already live (`tito/dns.tf`) |
+
+**CIDR overlap rule:** staging subnets overlap production's `10.0.0.x–10.0.5.x` subnets exactly. The default VPC may only ever get `/32` routes toward staging, at addresses verified to have **no ENI in the production VPC** (that check is Task 1).
+
+---
+
+### Task 1: Discovery — record the execution facts the tofu needs
+
+Read-only. Everything lands in `tito/rails-poc-facts.md` in the infra worktree so later tasks (and the prod apply later) never re-derive it.
+
+**Files:**
+- Create: `platform-infrastructure/tito/rails-poc-facts.md`
+
+**Interfaces:**
+- Produces: chosen box IPs (`classic_ip`, `pro_ip`), image refs (`classic_image`, `pro_image`), process commands (`WEB_COMMAND`, `WORKER_COMMAND`, `CLOCK_COMMAND` per app), web port + health path per app, staging DB/cache security-group IDs, box subnet ID. Task 2/3/6 consume these verbatim.
+
+- [ ] **Step 1: Create the infra worktree**
+
+```bash
+cd ~/github.com/carlosframework/platform-infrastructure
+git fetch origin
+git worktree add .claude/worktrees/rails-poc -b rails-poc origin/main
+cd .claude/worktrees/rails-poc
+```
+
+- [ ] **Step 2: Find where Buildkite pushes the Classic/Pro images**
+
+```bash
+aws --profile titogo --region eu-west-1 ecr describe-repositories \
+ --query 'repositories[].repositoryUri' --output table | grep -i -E 'classic|pro|tito'
+```
+
+If nothing matches, the images live outside ECR — ask Paul where Buildkite pushes (and how the boxes should authenticate); record the answer. Record the full image URI and the tag currently deployed to k8s staging for each app.
+
+- [ ] **Step 3: Record each app's process tree from the k8s staging manifests**
+
+Requires kubectl access to the staging cluster (Paul has it; if the session doesn't, ask Paul to paste the output):
+
+```bash
+kubectl get deploy -n <staging-namespace> -o jsonpath='{range .items[*]}{.metadata.name}{" cmd: "}{.spec.template.spec.containers[0].command}{" "}{.spec.template.spec.containers[0].args}{" port: "}{.spec.template.spec.containers[0].ports}{"\n"}{end}'
+```
+
+Record per app: the web command + container port, worker command(s), clock/cron command (if any), and the HTTP health path the k8s probes use (`.livenessProbe.httpGet.path`).
+
+- [ ] **Step 4: Record the staging DB/cache security groups**
+
+```bash
+aws --profile titogo --region eu-west-1 rds describe-db-clusters \
+ --query 'DBClusters[?contains(DBClusterIdentifier,`staging`)].{id:DBClusterIdentifier,sg:VpcSecurityGroups[].VpcSecurityGroupId}' --output json
+aws --profile titogo --region eu-west-1 elasticache describe-cache-clusters \
+ --query 'CacheClusters[?contains(CacheClusterId,`staging`)].{id:CacheClusterId,sg:SecurityGroups[].SecurityGroupId}' --output json
+```
+
+- [ ] **Step 5: Choose the two box IPs and prove the /32s shadow nothing**
+
+Pick the box subnet: whichever staging subnet is associated with the igw route table `rtb-041f57e000fb5797e` (check `aws ec2 describe-route-tables --route-table-ids rtb-041f57e000fb5797e --query 'RouteTables[].Associations'`). Pick two free IPs high in that subnet's range (e.g. `.200`/`.201` of its last /24), then verify **both** VPCs:
+
+```bash
+for ip in <candidate1> <candidate2>; do
+ aws --profile titogo --region eu-west-1 ec2 describe-network-interfaces \
+ --filters Name=addresses.private-ip-address,Values=$ip \
+ --query 'NetworkInterfaces[].{vpc:VpcId,desc:Description}' --output json
+done
+```
+
+Requirement: no ENI at that address in `vpc-90dffef5` (production — the /32 would shadow it from the default VPC) and none in `vpc-03be5678d08549203` (staging — the IP must be free to assign). Record IPs + the empty query output.
+
+- [ ] **Step 6: Commit the facts file**
+
+```bash
+git add tito/rails-poc-facts.md
+git commit -m "tito: record rails-poc execution facts (IPs, images, commands, SGs)"
+```
+
+---
+
+### Task 2: Networking tofu — peering and host routes
+
+**Files:**
+- Create: `platform-infrastructure/tito/rails-poc-network.tf`
+
+**Interfaces:**
+- Consumes: `classic_ip`, `pro_ip`, box subnet id from Task 1.
+- Produces: `aws_vpc_peering_connection.rails_poc_staging` and `local.rails_poc` locals; Task 3 references both.
+
+- [ ] **Step 1: Write `tito/rails-poc-network.tf`**
+
+```hcl
+# 🤖 Rails-behind-the-edge PoC (see platform repo spec 2026-08-11): the
+# edge (default VPC) reaches two staging Rails boxes over a peering.
+# The default VPC ALREADY routes 10.0.0.0/16 to the PRODUCTION peering
+# (pcx-056b55b69cbf89143) and staging shares that CIDR, so only /32
+# host routes are safe here — each verified ENI-free in the production
+# VPC before choosing (tito/rails-poc-facts.md).
+
+locals {
+ rails_poc = {
+ default_vpc_id = "vpc-02416de3913753989"
+ staging_vpc_id = "vpc-03be5678d08549203"
+ default_rtb_id = "rtb-08863e575439a6377"
+ staging_rtb_id = "<igw rtb from Task 1 step 5>"
+ box_subnet_id = "<subnet from Task 1 step 5>"
+ edge_private_ip = "172.31.4.137"
+ classic_ip = "<from facts>"
+ pro_ip = "<from facts>"
+ }
+}
+
+resource "aws_vpc_peering_connection" "rails_poc_staging" {
+ vpc_id = local.rails_poc.default_vpc_id
+ peer_vpc_id = local.rails_poc.staging_vpc_id
+ auto_accept = true
+ tags = { Name = "carlos-tito-rails-poc-staging" }
+}
+
+resource "aws_route" "rails_poc_to_classic" {
+ route_table_id = local.rails_poc.default_rtb_id
+ destination_cidr_block = "${local.rails_poc.classic_ip}/32"
+ vpc_peering_connection_id = aws_vpc_peering_connection.rails_poc_staging.id
+}
+
+resource "aws_route" "rails_poc_to_pro" {
+ route_table_id = local.rails_poc.default_rtb_id
+ destination_cidr_block = "${local.rails_poc.pro_ip}/32"
+ vpc_peering_connection_id = aws_vpc_peering_connection.rails_poc_staging.id
+}
+
+# Return path: the box subnet's route table learns the default VPC.
+# 172.31.0.0/16 previously fell through to the igw (dropped); no live
+# traffic changes meaning.
+resource "aws_route" "rails_poc_return" {
+ route_table_id = local.rails_poc.staging_rtb_id
+ destination_cidr_block = "172.31.0.0/16"
+ vpc_peering_connection_id = aws_vpc_peering_connection.rails_poc_staging.id
+}
+```
+
+Replace the three `<…>` values from `rails-poc-facts.md` — they are Task 1 outputs, not decisions.
+
+- [ ] **Step 2: Targeted plan, verify it creates exactly 4 resources**
+
+```bash
+tofu -chdir=tito plan \
+ -target=aws_vpc_peering_connection.rails_poc_staging \
+ -target=aws_route.rails_poc_to_classic \
+ -target=aws_route.rails_poc_to_pro \
+ -target=aws_route.rails_poc_return
+```
+
+Expected: `4 to add, 0 to change, 0 to destroy`. Anything else — stop, do not apply, diagnose.
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add tito/rails-poc-network.tf
+git commit -m "tito: rails-poc peering + /32 host routes (staging PoC networking)"
+```
+
+---
+
+### Task 3: Box tofu — SGs, IAM, AMI, instances, cloud-init
+
+**Files:**
+- Create: `platform-infrastructure/tito/rails-poc-boxes.tf`
+- Create: `platform-infrastructure/tito/rails-poc-init.sh.tpl`
+
+**Interfaces:**
+- Consumes: `local.rails_poc` (Task 2), image URIs + ports + commands (Task 1 facts).
+- Produces: `aws_security_group.rails_poc["classic"|"pro"]` (Task 6 references), boxes tagged `carlos-tito-rails-classic-staging` / `carlos-tito-rails-pro-staging` (Tasks 5/8 find them by that tag), output `rails_poc_ips`.
+
+- [ ] **Step 1: Write `tito/rails-poc-boxes.tf`**
+
+```hcl
+# svc-pdf pattern scaled up: SSM-only IAM, SG trust, docker via
+# cloud-init. x86 because the Buildkite images are amd64 (spec §1) —
+# the fleet's shared AMI is arm64, so a separate amd64 data source.
+
+data "aws_ami" "rails_poc_debian_amd64" {
+ most_recent = true
+ owners = ["136693071363"] # Debian official
+ filter {
+ name = "name"
+ values = ["debian-13-amd64-*"]
+ }
+ filter {
+ name = "architecture"
+ values = ["x86_64"]
+ }
+}
+
+locals {
+ rails_poc_apps = {
+ classic = { ip = local.rails_poc.classic_ip, port = <classic web port>, image = "<classic image uri>" }
+ pro = { ip = local.rails_poc.pro_ip, port = <pro web port>, image = "<pro image uri>" }
+ }
+}
+
+resource "aws_security_group" "rails_poc" {
+ for_each = local.rails_poc_apps
+ name = "carlos-tito-rails-${each.key}-staging"
+ description = "Rails PoC ${each.key}: web port from the IE edge only"
+ vpc_id = local.rails_poc.staging_vpc_id
+ ingress {
+ from_port = each.value.port
+ to_port = each.value.port
+ protocol = "tcp"
+ cidr_blocks = ["${local.rails_poc.edge_private_ip}/32"]
+ }
+ egress {
+ from_port = 0
+ to_port = 0
+ protocol = "-1"
+ cidr_blocks = ["0.0.0.0/0"]
+ }
+}
+
+resource "aws_iam_role" "rails_poc" {
+ for_each = local.rails_poc_apps
+ name = "carlos-tito-rails-${each.key}-staging"
+ assume_role_policy = jsonencode({
+ Version = "2012-10-17"
+ Statement = [{
+ Effect = "Allow"
+ Principal = { Service = "ec2.amazonaws.com" }
+ Action = "sts:AssumeRole"
+ }]
+ })
+}
+
+resource "aws_iam_role_policy_attachment" "rails_poc_ssm" {
+ for_each = local.rails_poc_apps
+ role = aws_iam_role.rails_poc[each.key].name
+ policy_arn = "arn:aws:iam::aws:policy/AmazonSSMManagedInstanceCore"
+}
+
+resource "aws_iam_role_policy" "rails_poc_pull_and_env" {
+ for_each = local.rails_poc_apps
+ name = "pull-and-env"
+ role = aws_iam_role.rails_poc[each.key].id
+ policy = jsonencode({
+ Version = "2012-10-17"
+ Statement = [
+ { Effect = "Allow", Action = ["ecr:GetAuthorizationToken"], Resource = "*" },
+ { Effect = "Allow",
+ Action = ["ecr:BatchGetImage", "ecr:GetDownloadUrlForLayer", "ecr:BatchCheckLayerAvailability"],
+ Resource = "<repo arn(s) for ${each.key} from Task 1>" },
+ { Effect = "Allow", Action = ["ssm:GetParameter"],
+ Resource = "arn:aws:ssm:eu-west-1:665106592142:parameter/tito/rails-poc/${each.key}-staging/*" },
+ ]
+ })
+}
+
+resource "aws_iam_instance_profile" "rails_poc" {
+ for_each = local.rails_poc_apps
+ name = "carlos-tito-rails-${each.key}-staging"
+ role = aws_iam_role.rails_poc[each.key].name
+}
+
+resource "aws_instance" "rails_poc" {
+ for_each = local.rails_poc_apps
+ ami = data.aws_ami.rails_poc_debian_amd64.id
+ instance_type = "t3a.large"
+ subnet_id = local.rails_poc.box_subnet_id
+ private_ip = each.value.ip
+ associate_public_ip_address = true # egress for SSM/ECR/apt; ingress is SG-closed
+ vpc_security_group_ids = [aws_security_group.rails_poc[each.key].id]
+ iam_instance_profile = aws_iam_instance_profile.rails_poc[each.key].name
+ user_data = templatefile("${path.module}/rails-poc-init.sh.tpl", {
+ app = each.key
+ port = each.value.port
+ })
+ user_data_replace_on_change = false
+ root_block_device {
+ volume_size = 40
+ volume_type = "gp3"
+ encrypted = true
+ }
+ tags = { Name = "carlos-tito-rails-${each.key}-staging" }
+ # Same floating-AMI trap as svc-pdf.tf: without this a newer Debian
+ # image turns `tofu apply` into "replace the box".
+ lifecycle {
+ ignore_changes = [ami]
+ }
+}
+
+output "rails_poc_ips" {
+ value = { for k, v in aws_instance.rails_poc : k => v.private_ip }
+}
+```
+
+- [ ] **Step 2: Write `tito/rails-poc-init.sh.tpl`**
+
+```bash
+#!/bin/bash
+# 🤖 Rails PoC box init (${app}): docker + ECR helper + env from SSM +
+# systemd units for the full process tree. Deploys happen via
+# scripts/rails-poc-deploy.sh over SSM — first boot starts nothing
+# until a deploy writes /etc/rails-poc/image.
+set -euo pipefail
+export DEBIAN_FRONTEND=noninteractive
+apt-get update
+apt-get install -y docker.io amazon-ecr-credential-helper awscli jq
+
+mkdir -p /root/.docker /etc/rails-poc
+cat > /root/.docker/config.json <<'EOF'
+{ "credsStore": "ecr-login" }
+EOF
+
+# App env from SSM (SecureString), 0600. Refreshed on every deploy too.
+cat > /usr/local/bin/rails-poc-fetch-env <<'EOF'
+#!/bin/bash
+set -euo pipefail
+aws ssm get-parameter --region eu-west-1 --with-decryption \
+ --name /tito/rails-poc/${app}-staging/env \
+ --query Parameter.Value --output text > /etc/rails-poc/app.env.new
+chmod 600 /etc/rails-poc/app.env.new
+mv /etc/rails-poc/app.env.new /etc/rails-poc/app.env
+EOF
+chmod 755 /usr/local/bin/rails-poc-fetch-env
+
+# One runner for every role; the role's command comes from the env file
+# (WEB_COMMAND / WORKER_COMMAND / CLOCK_COMMAND, recorded from the k8s
+# manifests in rails-poc-facts.md).
+cat > /usr/local/bin/rails-poc-run <<'EOF'
+#!/bin/bash
+set -euo pipefail
+role="$1"
+image="$(cat /etc/rails-poc/image)"
+source /etc/rails-poc/app.env
+var="$(echo "$role" | tr a-z A-Z)_COMMAND"
+publish=""
+[ "$role" = web ] && publish="-p ${port}:${port}"
+exec docker run --rm --name "rails-$role" $publish \
+ --env-file /etc/rails-poc/app.env "$image" $${!var}
+EOF
+chmod 755 /usr/local/bin/rails-poc-run
+
+for role in web worker clock; do
+cat > /etc/systemd/system/rails-$role.service <<EOF
+[Unit]
+Description=Rails PoC ${app} $role
+After=docker.service network-online.target
+Requires=docker.service
+[Service]
+ExecStartPre=-/usr/bin/docker rm -f rails-$role
+ExecStart=/usr/local/bin/rails-poc-run $role
+Restart=always
+RestartSec=5
+[Install]
+WantedBy=multi-user.target
+EOF
+done
+systemctl daemon-reload
+# Deliberately NOT enabled/started: the first deploy does that, once
+# /etc/rails-poc/image exists and the env parameter is populated.
+/usr/local/bin/rails-poc-fetch-env || true
+```
+
+Note the tpl escaping: `${app}` / `${port}` are template vars; `$${!var}` renders as shell `${!var}`. If an app has no clock process (Task 1 facts), that unit simply never gets enabled.
+
+- [ ] **Step 3: Targeted plan**
+
+```bash
+tofu -chdir=tito plan $(printf ' -target=%s' \
+ aws_security_group.rails_poc 'aws_iam_role.rails_poc' \
+ 'aws_iam_role_policy_attachment.rails_poc_ssm' 'aws_iam_role_policy.rails_poc_pull_and_env' \
+ 'aws_iam_instance_profile.rails_poc' 'aws_instance.rails_poc')
+```
+
+Expected: adds only (2 SGs, 2 roles, 2 attachments, 2 policies, 2 profiles, 2 instances = 12), zero change/destroy. Verify the plan shows `t3a.large` and an `debian-13-amd64` AMI.
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add tito/rails-poc-boxes.tf tito/rails-poc-init.sh.tpl
+git commit -m "tito: rails-poc staging boxes (x86, docker+systemd, SSM-only)"
+```
+
+---
+
+### Task 4: SSM env parameters (operator step with Paul)
+
+**Files:** none (parameters live in SSM only, per Global Constraints).
+
+**Interfaces:**
+- Produces: `/tito/rails-poc/classic-staging/env` and `/tito/rails-poc/pro-staging/env` (SecureString, env-file format). Task 3's boxes and Task 5 depend on them existing.
+
+- [ ] **Step 1: Assemble each app's env from the k8s staging config**
+
+The k8s deployment's env (ConfigMap + Secrets) IS the source; dump it with Paul (values must not land in the transcript — pipe to a root-owned temp file, or Paul runs it):
+
+```bash
+kubectl -n <staging-namespace> exec deploy/<web-deploy> -- env | sort > /tmp/classic-staging.env
+```
+
+Then append the three process-command lines from `rails-poc-facts.md`:
+
+```
+WEB_COMMAND=<web command>
+WORKER_COMMAND=<worker command>
+CLOCK_COMMAND=<clock command, omit line if none>
+```
+
+Prune k8s-runtime noise (`KUBERNETES_*`, `*_SERVICE_HOST`, `*_SERVICE_PORT`, `HOSTNAME`, `PATH`, `HOME`). Everything Rails reads stays.
+
+- [ ] **Step 2: Store as SecureString and shred the temp file**
+
+```bash
+aws --profile titogo --region eu-west-1 ssm put-parameter \
+ --name /tito/rails-poc/classic-staging/env --type SecureString \
+ --value file:///tmp/classic-staging.env
+shred -u /tmp/classic-staging.env
+```
+
+Repeat for pro. Verify with `aws ssm get-parameter --name … --query Parameter.Version` (never `--with-decryption` in the transcript).
+
+---
+
+### Task 5: Apply and boot the boxes
+
+**Interfaces:**
+- Consumes: Tasks 2–4 all landed.
+- Produces: two running boxes answering SSM, env fetched; nothing serving yet (first deploy is Task 8).
+
+- [ ] **Step 1: Targeted apply — networking first, then boxes**
+
+```bash
+tofu -chdir=tito apply \
+ -target=aws_vpc_peering_connection.rails_poc_staging \
+ -target=aws_route.rails_poc_to_classic \
+ -target=aws_route.rails_poc_to_pro \
+ -target=aws_route.rails_poc_return
+tofu -chdir=tito apply $(printf ' -target=%s' \
+ aws_security_group.rails_poc 'aws_iam_role.rails_poc' \
+ 'aws_iam_role_policy_attachment.rails_poc_ssm' 'aws_iam_role_policy.rails_poc_pull_and_env' \
+ 'aws_iam_instance_profile.rails_poc' 'aws_instance.rails_poc')
+```
+
+Review each plan before confirming — adds only, and the route apply must show exactly the two `/32`s + return route + peering.
+
+- [ ] **Step 2: Verify both boxes via SSM**
+
+```bash
+for tag in classic pro; do
+ iid=$(aws --profile titogo --region eu-west-1 ec2 describe-instances \
+ --filters Name=tag:Name,Values=carlos-tito-rails-$tag-staging Name=instance-state-name,Values=running \
+ --query 'Reservations[].Instances[].InstanceId' --output text)
+ cid=$(aws --profile titogo --region eu-west-1 ssm send-command --instance-ids "$iid" \
+ --document-name AWS-RunShellScript \
+ --parameters 'commands=["docker --version; ls -l /etc/rails-poc/app.env; systemctl list-unit-files rails-*"]' \
+ --query Command.CommandId --output text)
+ sleep 8
+ aws --profile titogo --region eu-west-1 ssm get-command-invocation \
+ --command-id "$cid" --instance-id "$iid" --query '{s:Status,o:StandardOutputContent}' --output json
+done
+```
+
+Expected: docker installed, `app.env` present mode 0600, three `rails-*.service` unit files. If `app.env` is missing, cloud-init ran before Task 4's parameters existed — rerun `/usr/local/bin/rails-poc-fetch-env` via SSM.
+
+- [ ] **Step 3: Commit the facts-file update (instance IDs)**
+
+```bash
+git add tito/rails-poc-facts.md
+git commit -m "tito: rails-poc boxes applied; record instance ids"
+```
+
+---
+
+### Task 6: Open the staging dependencies to the boxes
+
+**Files:**
+- Create: `platform-infrastructure/tito/rails-poc-deps.tf`
+
+**Interfaces:**
+- Consumes: staging DB/cache SG IDs (Task 1 facts), `aws_security_group.rails_poc` (Task 3).
+
+- [ ] **Step 1: Write `tito/rails-poc-deps.tf`**
+
+```hcl
+# Ingress on the (externally-managed) staging dependency SGs admitting
+# the PoC boxes. aws_security_group_rule against SG ids this tree does
+# not own — removal of the PoC removes exactly these rules and nothing
+# else. Classic is MySQL, Pro is Postgres (rails-poc-facts.md).
+
+locals {
+ rails_poc_dep_rules = {
+ classic_mysql = { sg = "<classic staging DB sg>", port = 3306, app = "classic" }
+ classic_valkey = { sg = "<classic staging redis sg>", port = 6379, app = "classic" }
+ classic_memcached = { sg = "<classic staging memcached sg>", port = 11211, app = "classic" }
+ pro_postgres = { sg = "<pro staging DB sg>", port = 5432, app = "pro" }
+ pro_valkey = { sg = "<pro staging redis sg>", port = 6379, app = "pro" }
+ pro_memcached = { sg = "<pro staging memcached sg>", port = 11211, app = "pro" }
+ }
+}
+
+resource "aws_security_group_rule" "rails_poc_dep" {
+ for_each = local.rails_poc_dep_rules
+ type = "ingress"
+ security_group_id = each.value.sg
+ from_port = each.value.port
+ to_port = each.value.port
+ protocol = "tcp"
+ source_security_group_id = aws_security_group.rails_poc[each.value.app].id
+ description = "rails-poc ${each.value.app} staging box"
+}
+```
+
+(Same-VPC SG references — the boxes live in the staging VPC precisely so this works.)
+
+- [ ] **Step 2: Targeted plan (6 adds only), apply, commit**
+
+```bash
+tofu -chdir=tito plan -target=aws_security_group_rule.rails_poc_dep
+tofu -chdir=tito apply -target=aws_security_group_rule.rails_poc_dep
+git add tito/rails-poc-deps.tf
+git commit -m "tito: rails-poc staging dependency SG ingress"
+```
+
+- [ ] **Step 3: Prove connectivity from each box (SSM)**
+
+```bash
+# on the classic box via send-command:
+timeout 3 bash -c '</dev/tcp/tito-classic-mysql-staging-writer.cefg7n9rzaz0.eu-west-1.rds.amazonaws.com/3306' && echo mysql-ok
+timeout 3 bash -c '</dev/tcp/<classic redis endpoint>/6379' && echo valkey-ok
+timeout 3 bash -c '</dev/tcp/<classic memcached endpoint>/11211' && echo memcached-ok
+# pro box: postgres 5432 + its two caches
+```
+
+Expected: three `*-ok` lines per box. A hang means the SG rule or a subnet route is wrong — fix before proceeding.
+
+---
+
+### Task 7: Edge routes
+
+**Interfaces:**
+- Consumes: box IPs + web ports (facts), boxes booted (Task 5).
+- Produces: `classic.platform.tito.io` / `pro.platform.tito.io` routed (503/refused until Task 8 deploys).
+
+- [ ] **Step 1: Add both routes on the IE box via SSM, as user `carlos`**
+
+```bash
+aws --profile titogo --region eu-west-1 ssm send-command --instance-ids i-025a0b70f3d93304d \
+ --document-name AWS-RunShellScript --parameters 'commands=[
+ "sudo -u carlos /opt/carlos/carlos add -kind service -addr <classic_ip>:<classic port> classic.platform.tito.io",
+ "sudo -u carlos /opt/carlos/carlos add -kind service -addr <pro_ip>:<pro port> pro.platform.tito.io"
+ ]' --query Command.CommandId --output text
+```
+
+Fetch the invocation output; each line must say `added` (or `updated … (existing route)` on a re-run — `carlos add` upserts; that is also how an upstream gets repointed). **No `-hibernate`, ever, on these routes.**
+
+- [ ] **Step 2: Verify TLS + routing end to end**
+
+```bash
+curl -sv https://classic.platform.tito.io/ -o /dev/null 2>&1 | grep -E 'subject:|HTTP/'
+curl -sv https://pro.platform.tito.io/ -o /dev/null 2>&1 | grep -E 'subject:|HTTP/'
+```
+
+Expected: a Let's Encrypt cert for each exact hostname (first hit may take a few seconds while HTTP-01 completes), then `502`/`503` — the edge is up, nothing listens on the box yet. A TLS `internal error` alert means the route didn't land in `/data/registry.db`; re-check Step 1's output.
+
+---
+
+### Task 8: Deploy script + first deploy
+
+**Files:**
+- Create: `platform-infrastructure/scripts/rails-poc-deploy.sh`
+
+**Interfaces:**
+- Consumes: everything above; image tag currently on k8s staging (facts).
+- Produces: `rails-poc-deploy.sh <classic|pro> <image-ref>` — THE deploy path from here on.
+
+- [ ] **Step 1: Write `scripts/rails-poc-deploy.sh`**
+
+```bash
+#!/bin/bash
+# 🤖 Deploy a Rails PoC box over SSM: refresh env, point at the image,
+# restart the tree, wait for health through the edge.
+# Usage: rails-poc-deploy.sh <classic|pro> <image-ref>
+set -euo pipefail
+app="$1"; image="$2"
+region=eu-west-1; profile=titogo
+iid=$(aws --profile $profile --region $region ec2 describe-instances \
+ --filters Name=tag:Name,Values=carlos-tito-rails-$app-staging Name=instance-state-name,Values=running \
+ --query 'Reservations[].Instances[].InstanceId' --output text)
+[ -n "$iid" ] || { echo "no running box for $app" >&2; exit 1; }
+
+cid=$(aws --profile $profile --region $region ssm send-command --instance-ids "$iid" \
+ --document-name AWS-RunShellScript --parameters commands="[
+ \"/usr/local/bin/rails-poc-fetch-env\",
+ \"docker pull $image\",
+ \"echo $image > /etc/rails-poc/image\",
+ \"systemctl enable rails-web rails-worker\",
+ \"systemctl restart rails-web rails-worker\",
+ \"systemctl is-enabled rails-clock >/dev/null 2>&1 && systemctl restart rails-clock || true\"
+ ]" --query Command.CommandId --output text)
+
+for i in $(seq 1 60); do
+ st=$(aws --profile $profile --region $region ssm get-command-invocation \
+ --command-id "$cid" --instance-id "$iid" --query Status --output text)
+ [ "$st" = Success ] && break
+ [ "$st" = Failed ] && { aws --profile $profile --region $region ssm get-command-invocation \
+ --command-id "$cid" --instance-id "$iid" --query StandardErrorContent --output text >&2; exit 1; }
+ sleep 5
+done
+
+url="https://$app.platform.tito.io<health path from facts>"
+for i in $(seq 1 30); do
+ code=$(curl -s -o /dev/null -w '%{http_code}' "$url") && [ "$code" = 200 ] && { echo "$app healthy: $url"; exit 0; }
+ sleep 5
+done
+echo "$app not healthy after deploy ($url last code $code)" >&2; exit 1
+```
+
+`chmod +x scripts/rails-poc-deploy.sh`. (Apps with a clock process get it enabled once, manually, via SSM: `systemctl enable rails-clock` — the script then keeps restarting it.)
+
+- [ ] **Step 2: First deploys, with the tag k8s staging currently runs**
+
+```bash
+./scripts/rails-poc-deploy.sh classic <classic image:tag from facts>
+./scripts/rails-poc-deploy.sh pro <pro image:tag from facts>
+```
+
+Expected: both end with `healthy`. Debug path if not: `journalctl -u rails-web -n 100` via SSM on the box. Likely first-boot issues: a missing env var (fix the SSM parameter, redeploy), or the app binding localhost only (the k8s manifests' web command already binds 0.0.0.0 — reuse it exactly).
+
+- [ ] **Step 3: Verify the full tree and the replica behavior**
+
+```bash
+# via SSM on each box:
+docker ps --format '{{.Names}} {{.Status}}' # rails-web, rails-worker (+ rails-clock)
+```
+
+Then the shared-queue proof: with Paul, enqueue a staging background job through the **k8s** app (or its console) and watch it get picked up on a PoC box: `journalctl -u rails-worker -f` via an SSM session. A job processed on the PoC box while k8s workers also run = the parallel-replica claim, proven.
+
+- [ ] **Step 4: Commit**
+
+```bash
+git add scripts/rails-poc-deploy.sh
+git commit -m "scripts: rails-poc deploy over SSM (pull, restart, edge health)"
+```
+
+---
+
+### Task 9: Production tofu — written, gated off
+
+**Files:**
+- Create: `platform-infrastructure/tito/rails-poc-prod.tf`
+
+- [ ] **Step 1: Write the gated file**
+
+```hcl
+# 🤖 Production twin of the staging PoC — WRITTEN, NOT APPLIED (spec:
+# staging first; flipping this variable is the post-PoC decision).
+# Networking is SIMPLER than staging: the default VPC already routes
+# 10.0.0.0/16 to the existing production peering pcx-056b55b69cbf89143,
+# so prod boxes need NO new peering and NO new routes — only boxes,
+# SG rules on the prod dependency SGs, and /tito/rails-poc/<app>-prod/env
+# parameters. Fill the <…> values from a prod re-run of Task 1
+# (rails-poc-facts.md gets a prod section) before ever enabling.
+
+variable "rails_poc_prod_enabled" {
+ type = bool
+ default = false
+}
+
+resource "aws_instance" "rails_poc_prod" {
+ for_each = var.rails_poc_prod_enabled ? {
+ classic = { ip = "<prod classic ip — ENI-free check vs BOTH VPCs>", port = <port>, subnet = "<prod subnet>" }
+ pro = { ip = "<prod pro ip>", port = <port>, subnet = "<prod subnet>" }
+ } : {}
+ ami = data.aws_ami.rails_poc_debian_amd64.id
+ instance_type = "t3a.large"
+ subnet_id = each.value.subnet
+ private_ip = each.value.ip
+ vpc_security_group_ids = [] # prod SGs defined alongside when enabling
+ user_data = templatefile("${path.module}/rails-poc-init.sh.tpl", {
+ app = each.key
+ port = each.value.port
+ })
+ root_block_device {
+ volume_size = 40
+ volume_type = "gp3"
+ encrypted = true
+ }
+ tags = { Name = "carlos-tito-rails-${each.key}-prod" }
+ lifecycle { ignore_changes = [ami] }
+}
+```
+
+- [ ] **Step 2: Prove it is inert, commit**
+
+```bash
+tofu -chdir=tito plan -target=aws_instance.rails_poc_prod
+# Expected: "No changes." — the empty for_each makes it a no-op.
+git add tito/rails-poc-prod.tf
+git commit -m "tito: rails-poc production twin, gated off (post-PoC decision)"
+```
+
+---
+
+### Task 10: Runbook, PRs, and the evaluation record
+
+**Files:**
+- Modify: `platform-infrastructure/tito/README.md` (append a `## Rails PoC` section)
+- Modify (platform repo, branch `rails-behind-edge-poc-spec`): the spec's §7 — check off what was actually met
+
+- [ ] **Step 1: Append the runbook to `tito/README.md`**
+
+```markdown
+## Rails PoC (classic/pro staging behind the edge)
+
+Spec + plan: platform repo `docs/superpowers/{specs,plans}/2026-08-11-rails-behind-carlos-edge-poc*`.
+Facts (IPs, images, SGs): `tito/rails-poc-facts.md`.
+
+- Deploy: `scripts/rails-poc-deploy.sh <classic|pro> <image-ref>` (SSM; no SSH).
+- Routes live in the IE box's `/data/registry.db` (`carlos add` as user
+ `carlos`; re-run to repoint; NEVER `-hibernate` on these).
+- Logs: `journalctl -u rails-{web,worker,clock}` via SSM on the box.
+- Teardown: remove the routes on the IE box, then targeted destroys of
+ `aws_instance.rails_poc`, `aws_security_group_rule.rails_poc_dep`,
+ the two `/32` `aws_route.rails_poc_to_*`, `aws_route.rails_poc_return`,
+ `aws_vpc_peering_connection.rails_poc_staging`; delete the
+ `/tito/rails-poc/*` SSM parameters.
+- The default VPC's `10.0.0.0/16 → pcx-056b55b69cbf89143` (production)
+ route is why staging only ever gets `/32` routes — do not "simplify".
+- Fallback if a shared cache is unreachable (spec §"Decisions"): an
+ on-box memcached is one unit away —
+ `docker run -d --restart=always -p 127.0.0.1:11211:11211 memcached:1.6`
+ plus pointing the app's cache env var at `127.0.0.1:11211`. Documented
+ fallback only; the default is the shared staging Elasticache.
+```
+
+- [ ] **Step 2: During the parallel-run week, record the §7 gap list against reality**
+
+In the spec (platform repo), annotate each §7 item with what actually happened (e.g. "config delivery: hand-rolled SSM fetch worked but rotation is manual — a real `appconfig` gap"). This annotated list is the deliverable of the evaluation.
+
+- [ ] **Step 3: Open the PRs**
+
+```bash
+# infra:
+cd ~/github.com/carlosframework/platform-infrastructure/.claude/worktrees/rails-poc
+git push -u origin rails-poc
+gh pr create --title "tito: Rails-behind-the-edge staging PoC (classic/pro)" \
+ --body "Boxes, peering, dep SG rules, deploy script per the platform repo spec 2026-08-11-rails-behind-carlos-edge-poc-design.md. Targeted applies only — see plan.
+
+🤖 Generated with [Claude Code](https://claude.com/claude-code)"
+# spec+plan:
+cd ~/github.com/carlosframework/platform/.claude/worktrees/rails-behind-edge-poc-spec
+git push -u origin rails-behind-edge-poc-spec
+gh pr create --title "docs: Rails-behind-the-edge PoC — design record + implementation plan" \
+ --body "Design + plan for serving Tito Classic/Pro staging through the CARLOS edge (spec approved by Paul 2026-08-11).
+
+🤖 Generated with [Claude Code](https://claude.com/claude-code)"
+```
+
+---
+
+## Execution notes
+
+- Tasks 1→8 are strictly ordered; 9 and 10 can follow any time after 3 and 8 respectively.
+- Applies (Tasks 5, 6) and anything touching the IE box (Task 7) or k8s (Tasks 1, 4, 8) are operator-visible actions — do them deliberately, one at a time, reading every plan before confirming.
+- If any targeted plan proposes a change or destroy outside the `rails_poc*` names, STOP — that is the known tree drift; never apply through it.
diff --git a/docs/superpowers/specs/2026-08-11-rails-behind-carlos-edge-poc-design.md b/docs/superpowers/specs/2026-08-11-rails-behind-carlos-edge-poc-design.md
new file mode 100644
index 000000000..04eacc6e6
--- /dev/null
+++ b/docs/superpowers/specs/2026-08-11-rails-behind-carlos-edge-poc-design.md
@@ -0,0 +1,284 @@
+# Rails behind the CARLOS edge: the Classic/Pro proof of concept
+
+**Date:** 2026-08-11
+**Status:** Approved design, staging slice first
+**Prompted by:** James running miren (miren.dev, a container deployment
+tool) behind the carlos router. Question under test: can the CARLOS tito
+deployment front the two big Rails apps — Tito Classic and Tito Pro — well
+enough that migrating them off Kubernetes onto a much simpler all-CARLOS
+deployment is a real option rather than a slogan.
+
+## Context
+
+Tito Classic and Tito Pro run today on Kubernetes, with Postgres, Redis and
+Memcached dependencies, in the **same AWS account and region as the CARLOS
+tito boxes** (665106592142, eu-west-1). Buildkite already builds
+Docker images for the k8s deployment; those images are amd64, and the PoC
+runs them as-is on x86 boxes (Paul, 2026-08-11: these boxes are
+disposable — they'll be turned off after the evaluation, so rebuilding for
+arm64 buys nothing here). arm64 only matters if Rails-on-CARLOS graduates
+to a first-class deployment on the Graviton fleet, and that is recorded as
+roadmap fallout (§7), not PoC work.
+
+CARLOS today has no container runtime and its release format is a single
+static binary — a Rails app cannot be a CARLOS *release*. But the edge
+already fronts processes it didn't build: a route can proxy to an arbitrary
+TCP `host:port` (`-kind service -addr`, `internal/registry/registry.go`),
+and the svc-pdf box (`platform-infrastructure/tito/svc-pdf.tf`) is the live
+precedent for "a Docker-shaped dependency on its own tofu-provisioned box,
+trusted via security group". The PoC composes exactly those two precedents
+and deliberately nothing new in the platform.
+
+Two facts remove whole workstreams: `*.platform.tito.io` already resolves
+to the IE box's EIP (`tito/dns.tf`, wildcard A record), and the edge mints
+per-host HTTP-01 certs automatically the moment a route exists. DNS and TLS
+are zero work.
+
+## Decisions taken during design
+
+- **Runtime on the boxes: bare box — docker + systemd** (svc-pdf pattern
+ scaled up). Miren was considered (closest to James's PoC) and rejected as
+ a second control plane pointing away from "simpler all-CARLOS"; Kamal was
+ rejected because it requires SSH (the fleet is deliberately SSM-only) and
+ inserts its own proxy.
+- **Reuse the k8s images unmodified** — the PoC runs exactly what
+ Buildkite builds (amd64); no separate build pipeline, no arm64 rebuilds,
+ no source checkouts on boxes.
+- **Staging first.** Boxes wired to STAGING Postgres/Redis/Memcached now;
+ production boxes are written in tofu in the same PR but applied only
+ after the staging PoC convinces us.
+- **Full process tree** — web + workers + cron, so the boxes are true
+ parallel replicas of the k8s deployment sharing its queues and cache.
+ Workers pulling real staging jobs from the shared Redis is the point,
+ not a hazard.
+- **Reuse existing staging dependencies** rather than on-box replacements;
+ an on-box memcached container is a documented fallback only.
+
+## Design
+
+### 1. Boxes
+
+Two new EC2 instances in `platform-infrastructure/tito/`, one `.tf` +
+cloud-init file each, copying the svc-pdf shape:
+
+- `rails-classic-staging` and `rails-pro-staging`
+- x86 (amd64), because that is what the Buildkite images are — `t3a.large`
+ to start (t4g is Graviton-only, so the burstable x86 class is t3/t3a),
+ sized honestly after the first week; an amd64 Debian AMI data source
+ (the fleet's shared AMI is arm64), with
+ `lifecycle { ignore_changes = [ami] }` (the floating-AMI replace trap is
+ recorded in svc-pdf.tf)
+- Own security group each; ingress **only** the app port from the IE box's
+ SG (`module.host_ie.security_group_id`) — same trust boundary as svc-pdf
+- IAM: `AmazonSSMManagedInstanceCore` + ECR pull on the Classic/Pro
+ repositories + `ssm:GetParameter` on `/tito/rails-poc/<app>-staging/*`.
+ Nothing else — no bucket grants.
+- No SSH. All operation via SSM, like the rest of the fleet.
+
+`rails-classic-prod.tf` / `rails-pro-prod.tf` are written in the same PR,
+`count = 0` (or commented apply instructions) until the staging PoC passes.
+
+### 2. On each box
+
+Cloud-init installs docker + the ECR credential helper, materializes the
+app env from SSM Parameter Store (`/tito/rails-poc/<app>-staging/env`,
+written to `/etc/rails-poc/<app>.env`, 0600), and writes systemd units —
+one per process type, each a `docker run` on the pinned image tag:
+
+- `rails-web.service` — the Rails/puma web process, port published on the
+ box's private IP
+- `rails-worker.service` — background workers (one unit per worker class
+ if the k8s deployment splits them)
+- `rails-cron.service` — clock/scheduler process, if the app has one
+- all `Restart=always`, logs to journald
+
+The env file carries DATABASE_URL / REDIS_URL / MEMCACHED / SECRET_KEY_BASE
+etc., populated once by hand from the k8s staging secrets (values never in
+tofu, never in the repo).
+
+### 3. Images
+
+The Buildkite-built amd64 images are used unmodified, pulled from the same
+registry the k8s deployment pulls from. The boxes track the same tags k8s
+staging deploys, so the two deployments stay comparable during the
+parallel run. No rebuilds of any kind for the PoC.
+
+### 4. Dependencies
+
+Staging Postgres/Redis/Memcached are reached by adding ingress rules on
+their security groups admitting the new boxes' SGs. **Preflight before
+apply:** confirm the staging RDS/Elasticache share a VPC with the default
+VPC these boxes land in (same account/region is confirmed; the svc-pdf
+pattern uses the default VPC). If EKS keeps its data stores in its own
+VPC, place the PoC boxes in that VPC instead — a variable on the box
+resources, not a redesign — or peer.
+
+### 5. Edge routing
+
+On the IE box (via SSM, as user `carlos`):
+
+```
+carlos add -kind service -addr <classic-private-ip>:<port> classic.platform.tito.io
+carlos add -kind service -addr <pro-private-ip>:<port> pro.platform.tito.io
+```
+
+Gotchas carried from exploration:
+
+- Confirm which registry the edge reads before adding routes — the tito
+ box carries two (`/data/registry.db` platform, `/data/titogo/registry.db`
+ titogo's own).
+- Never set `-hibernate` on these routes: hibernation drives the exec/unit
+ backends and their `--socket/--db/healthz` contract, which these boxes
+ don't speak.
+- Repointing an upstream = re-run `carlos add` (it upserts); there is no
+ `carlos route -addr` setter.
+- Addr-upstream routes are "externally managed by design"
+ (`internal/host/socket_advice.go`) — no config delivery, no version
+ stamping, no log shipping, no managed restarts. That is expected and is
+ part of what the PoC measures (see §7).
+
+### 6. Deploys
+
+A small script per app, run over SSM RunCommand from a developer laptop
+(or CI later): pull the new image tag → restart the units in order →
+poll the health endpoint → report. No SSH, no new tooling. The tag
+deployed is whatever CI pushed for k8s staging, so the two deployments
+stay comparable during the parallel-run phase.
+
+### Recorded deviations (execution)
+
+- **Box SG ingress uses the IE edge's private IP `/32`
+ (`172.31.4.137/32`) rather than an SG reference** — the peering is
+ cross-VPC, and a security group cannot be referenced across the peering
+ boundary the way §2 assumed. Breaks silently if the IE box is ever
+ replaced at a new private IP; re-point the ingress rule then.
+- **The web port is published on all interfaces, not the private IP
+ only** — §2 said "the box's private IP"; what's running binds
+ `0.0.0.0`. Exposure is nil in practice (the SG admits only the edge),
+ but the defense-in-depth intent of §2 was dropped.
+
+### 7. What the PoC deliberately surfaces (the roadmap fallout)
+
+CARLOS gives these routes routing + TLS and nothing else. Every gap met is
+a named candidate roadmap item, and this list is the actual deliverable of
+the migration evaluation:
+
+1. **No image/bundle release kind** — a CARLOS release is one static
+ binary; ExecStart is pinned to `/opt/carlos/live/%i`
+ (`internal/caps/render.go`). Running Rails as a first-class app needs a
+ bundle/image release kind or an equivalent.
+ - observed: confirmed — the gap is real, and the PoC boxes hand-roll
+ docker + systemd unit files in its place, not a CARLOS release at all.
+2. **No attached-database primitive** — the platform design spec already
+ sketches `attached` engines (Postgres provisioned and bound by the
+ platform, `docs/superpowers/specs/2026-08-01-carlos-platform-design.md`
+ §storage); the objstore primitive
+ (`2026-08-10-object-storage-primitive-design.md`) is the implementation
+ template.
+ - observed: no such primitive existed to reach for — dependency access
+ was wired by hand (security-group ingress rules on the shared staging
+ DB/cache SGs, plus connection env delivered separately); it worked,
+ but nothing about it would survive a second app onboarding itself.
+3. **No config/env delivery to addr-upstream routes** — the PoC hand-rolls
+ SSM Parameter Store; a platform answer would ride the existing
+ `appconfig` rails.
+ - observed: confirmed, and it cost a real mistake — the hand-rolled
+ prune rule over-pruned a live key (`PROMETHEUS_EXPORTER_PORT`) from
+ both apps' env on the first pass; caught by review, not by any
+ safeguard the ad hoc delivery had.
+4. **No log shipping for external processes** — journald-only on the PoC
+ boxes; CARLOS logs stage 2 (unit tenants) is the relevant seam.
+ - observed: confirmed — journald-only, no shipping. Adequate for the
+ PoC's own diagnosis needs; not evidence either way for real ops load.
+5. **No health/restart/metrics for off-box upstreams** — `-unit` stamping
+ only works for processes on the edge box itself.
+ - observed: confirmed — the deploy script itself carries the whole
+ health gate (on-box HTTP probe with per-app path/header, then a
+ report-only edge check); CARLOS supplies none of it for this route
+ shape.
+6. **Architecture** — the fleet is Graviton; the Rails images are amd64.
+ A first-class Rails-on-CARLOS deployment means arm64 image builds
+ (Buildkite multi-arch). Deferred deliberately: the PoC boxes are
+ disposable, so they run x86 to match the images rather than the other
+ way round.
+ - observed: the x86 boxes ran the unmodified Buildkite amd64 images
+ with no rebuild and no compatibility issue — the deferral cost
+ nothing for the PoC itself; the arm64 gap for a real deployment
+ stands as stated.
+7. **High availability** (added 2026-08-12, Paul's question). The PoC has
+ deliberately little: one box per app, one AZ, edge proxying to one IP;
+ `Restart=always` covers process death and nothing covers box death.
+ The layers split cleanly into two owners, and **nothing Tito-side
+ waits on anything CARLOS-side**:
+ - **Tito concerns** (this migration's own work, plain AWS, no platform
+ changes): EC2 auto-recovery on the boxes; N boxes per app behind an
+ internal NLB with the edge's `-addr` pointed at the NLB (multi-AZ,
+ health-checked failover, rolling zero-downtime deploys — and
+ networking-trivial in prod, where the existing peering already routes
+ the whole VPC); Classic's supercronic → sidekiq-ent so cron stops
+ being a single-box constraint (Pro's already is, with Redis-backed
+ single-fire); the hostname/host-authorization config.
+ - **CARLOS concerns** (roadmap candidates, priced by whether
+ Rails-on-CARLOS graduates from migration tactic to product):
+ multi-upstream health-checked `addr` routes at the edge — the
+ platform-native replacement for the NLB rung (filed as platform
+ #152); and the edge's own HA posture (standby/EIP failover), which
+ is a whole-deployment question every CARLOS tenant shares — the
+ Rails apps widen the blast radius of a risk `go.tito.io` already
+ carries, they don't create it (filed as platform #153).
+ - Data HA is out of scope by construction: Postgres/MySQL/Valkey/
+ Memcached stay in RDS/Elasticache with their own multi-AZ story —
+ the apps' state was never on k8s, so the migration never touches the
+ layer where losing things actually hurts.
+
+**2026-08-11:** staging is live on both apps (see `tito/README.md`'s Rails
+PoC section in platform-infrastructure for current status). The edge
+host-authorization gap (classic 403, pro 404) is still open and is what
+blocks this route shape from serving a real page today — the PoC proved
+the plumbing, not a finished path to production traffic.
+
+**2026-08-11, later — two verification items closed on-box:**
+
+- **Shared-queue replica behavior (Verification item 4): proven for Pro,
+ organically.** The pro box's sidekiq processed 35 real staging jobs in
+ one observed hour (`ScheduledTasks::ExpireUnfulfilledAllocationsWorker`,
+ real JIDs in journald) — enqueued into the shared staging Redis by
+ Pro's own sidekiq-enterprise cron and won by the box worker in
+ competition with the k8s workers. Classic's staging queues were quiet
+ over the same window, so its half still wants a deliberate enqueue.
+- **Real-page serving: both apps serve correct full responses from the
+ boxes** when addressed by a hostname they already know (curl on-box
+ with `Host:` + `X-Forwarded-Proto: https`, exactly what the edge
+ forwards): `staging.pro.tito.io` → 200 with the real Tito Pro page;
+ `staging.tito.io` → 301 to canonical `staging.ti.to`;
+ `dashboard.staging.tito.io` → 302 into its SSO login flow. The 403/404
+ on the edge hostnames is therefore ONLY hostname configuration
+ (Classic's host allowlist; a Pro tenant/alias for the hostname), not a
+ serving gap — the remaining decision is which hostnames the parallel
+ deployment should answer to, which is a migration-design question, not
+ an infrastructure one.
+
+## Out of scope
+
+- Any platform code change (release kinds, primitives) — the PoC is
+ precedent-composition only.
+- Cutting any real traffic over from k8s; DNS for the apps' real customer
+ domains is untouched.
+- Production apply (written, not applied).
+- Any image rebuilds, including arm64/multi-arch (recorded as roadmap
+ fallout in §7 instead).
+
+## Verification
+
+1. `tofu plan` **targeted** to the new resources — the tito tree carries
+ known drift from open PRs and a full apply would strip live SMTP
+ ingress (recorded in `tito/object-store.tf`'s commit message).
+2. Boxes reachable via SSM; units running; env files 0600.
+3. `https://classic.platform.tito.io` and `https://pro.platform.tito.io`
+ serve the real staging apps with valid certs.
+4. A staging background job enqueued through the k8s deployment is
+ observed processed by a PoC-box worker (proves shared-queue replica
+ behavior).
+5. One deploy exercised end-to-end over SSM (new tag → restart → healthy).
+6. A week running alongside k8s untouched; then the production-apply
+ decision.