docs: update server spec after US to EU migration

README.md

@@ -13,7 +13,7 @@ Self-hosted backup of your Gmail to your Google Drive. Every email is rendered t
 
 ## Status
 
-Live in production on a Hetzner k3s cluster (~€8/mo). The full pipeline — Gmail History polling → MIME fetch → PDF render (Gotenberg) → Drive upload — runs end-to-end, with KEDA scaling each worker pool 1–5 on queue depth and the reliable-queue pop (BLMOVE into per-worker processing lists, acked after the DB settles, reclaimed by the sweeper when a worker dies).
+Live in production on a Hetzner k3s cluster (~$6.5/mo). The full pipeline — Gmail History polling → MIME fetch → PDF render (Gotenberg) → Drive upload — runs end-to-end, with KEDA scaling each worker pool 1–5 on queue depth and the reliable-queue pop (BLMOVE into per-worker processing lists, acked after the DB settles, reclaimed by the sweeper when a worker dies).
 
 Captured evidence: [a chaos run](docs/chaos-demo-sample.txt) — worker killed mid-job, recovered in 21 s — and [a flood chart](docs/loadtest/flood-run.png) — 1,000-job burst, replicas 1→5, drained in ~50 s.
 
@@ -146,7 +146,7 @@ K8s manifests under `deploy/k8s/`. Validate offline:
 make k8s-validate     # uses kubeconform — `brew install kubeconform`
 ```
 
-Production is k3s on a single Hetzner cpx21 (3 vCPU / 4 GB, ~€8/mo), Traefik ingress on :80. Moving providers is a 20-minute runbook ([OPERATIONS.md §3](docs/OPERATIONS.md)) — it ran DigitalOcean → Hetzner for real. KEDA install + autoscaling: §13; loadtest + chart: §14.
+Production is k3s on a single Hetzner cx23 (2 vCPU / 4 GB, ~$6.5/mo), Traefik ingress on :80. Moving providers is a 20-minute runbook ([OPERATIONS.md §3](docs/OPERATIONS.md)) — it ran DigitalOcean → Hetzner for real. KEDA install + autoscaling: §13; loadtest + chart: §14.
 
 ## Repo layout
 

docs/OPERATIONS.md

@@ -6,7 +6,7 @@ Runbooks for operating DistTaskQueue in production. For *what* each component do
 
 ## 1. System inventory
 
-What runs where. Current home: a Hetzner cpx21 (3 vCPU / 4 GB) running k3s.
+What runs where. Current home: a Hetzner cx23 (2 vCPU / 4 GB, fsn1) running k3s.
 
 | Component | Binary / image | Replicas | Health |
 |---|---|---|---|
@@ -421,7 +421,7 @@ Guidelines based on measured behavior.
 
 | Scenario | Recommended spec |
 |---|---|
-| Personal inbox, ~100 emails/day | 1 vCPU, 2 GB RAM, 30 GB disk minimum. In production: Hetzner cpx21 (3 vCPU / 4 GB, ~€8/mo) with headroom for KEDA scale-out. Runs cold. |
+| Personal inbox, ~100 emails/day | 1 vCPU, 2 GB RAM, 30 GB disk minimum. In production: Hetzner cx23 (2 vCPU / 4 GB, ~$6.5/mo) with headroom for KEDA scale-out. Runs cold. |
 | Personal inbox, ~1,000 emails/day | 2 vCPU, 4 GB RAM. Mostly for the Chromium renders during bursts. |
 | Multi-user, 10 users × 100 emails/day | 2 vCPU, 4 GB RAM + managed Postgres. PVC for workers must be RWX. |
 | Multi-user, 100+ users | Move Postgres + Redis to managed services (RDS + ElastiCache or equivalent). Scale render-worker to 3–5 replicas. Replace `/data` with object storage. |

docs/superpowers/plans/2026-07-09-anki-deck-audit.md

@@ -0,0 +1,226 @@
+# Anki Deck Audit 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:** Produce a current, atomic, import-safe Anki deck for DistTaskQueue and export an identical copy to Downloads.
+
+**Architecture:** Treat the current repository as the factual source and the existing deck as an editable inventory. Audit facts before rewriting notes, then validate the completed tab-separated artifact structurally and against repository evidence before exporting it.
+
+**Tech Stack:** Anki text import, tab-separated UTF-8 text, Go source and tests, Redis, Postgres, Kubernetes, shell validation
+
+---
+
+### Task 1: Capture the deck inventory and authoritative project facts
+
+**Files:**
+- Read: `All_Decks__System_Design__General.txt`
+- Read: `INTERVIEW-BRIEF.md`
+- Read: `README.md`
+- Read: `docs/ARCHITECTURE.md`
+- Read: `docs/OPERATIONS.md`
+- Read: `docs/WAR-STORIES.md`
+- Read: `internal/queue/queue.go`
+- Read: `internal/worker/worker.go`
+- Read: `internal/worker/backoff.go`
+- Read: `internal/sweeper/sweeper.go`
+- Read: `internal/store/store.go`
+- Read: `internal/store/migrations/0001_init.up.sql`
+- Read: `internal/scheduler/scheduler.go`
+- Read: `internal/handler/fetch.go`
+- Read: `internal/handler/render.go`
+- Read: `internal/handler/upload.go`
+- Read: `internal/drive/client.go`
+- Read: `deploy/k8s/01-config.yaml`
+- Read: `deploy/k8s/29-keda-scaledobjects.yaml`
+- Read: `docs/loadtest/flood-run.csv`
+
+- [ ] **Step 1: Record the current import shape and card count**
+
+Run:
+
+```bash
+head -n 5 All_Decks__System_Design__General.txt
+awk 'NR > 5 {count++} END {print count}' All_Decks__System_Design__General.txt
+```
+
+Expected: five Anki headers followed by the current note count.
+
+- [ ] **Step 2: Locate all existing DistTaskQueue prompts and volatile claims**
+
+Run:
+
+```bash
+rg -n 'Task Queue|DistTaskQueue|5K|5,000|5\.8s|2,400|heartbeat|TTL|sweep|attempt|KEDA|Hetzner|Gmail|Drive' All_Decks__System_Design__General.txt
+```
+
+Expected: every project-specific card and every measurement or deployment fact is visible for verification.
+
+- [ ] **Step 3: Verify runtime constants and state transitions from source**
+
+Run:
+
+```bash
+rg -n 'BLMove|LMove|LRem|Heartbeat|MaxAttempts|Compute|Sweep|MarkFailed|ClaimJob|AdvanceJob|MarkDone|time\.(Second|Minute)|Ticker|TTL|next_run_at' internal cmd deploy/k8s
+```
+
+Expected: code evidence for queue ordering, processing lists, heartbeats, retries, sweeps, claims, and stage transitions.
+
+- [ ] **Step 4: Verify measured claims from checked-in artifacts**
+
+Run:
+
+```bash
+rg -n '5000|5,000|jobs/sec|recovery|seconds|replica|1.*5|€8|Hetzner' README.md INTERVIEW-BRIEF.md docs scripts internal/loadtest
+```
+
+Expected: each retained metric has a checked-in source and its measurement context.
+
+### Task 2: Rewrite the deck using minimum-information cards
+
+**Files:**
+- Modify: `All_Decks__System_Design__General.txt`
+
+- [ ] **Step 1: Preserve the import contract**
+
+Keep these lines unchanged at the top of the file:
+
+```text
+#separator:tab
+#html:false
+#notetype column:1
+#deck column:2
+#tags column:5
+```
+
+Each subsequent record must contain exactly these five fields:
+
+```text
+Basic<TAB>deck path<TAB>prompt<TAB>answer<TAB>space-separated tags
+```
+
+- [ ] **Step 2: Correct existing DistTaskQueue cards**
+
+Use `apply_patch` to replace claims that conflict with Task 1 evidence. Keep one recall target per prompt. Split combined cards for timings, retry policy, recovery passes, delivery semantics, queue mechanics, scaling limits, and deployment facts. Add `as-of-2026-07` to volatile-fact tags and include the measurement context in the prompt.
+
+- [ ] **Step 3: Add missing high-value project cards**
+
+Add atomic prompts for:
+
+```text
+pipeline stages and why they scale separately
+Postgres source-of-truth invariant
+BLMOVE processing-list protocol and ACK behavior
+conditional UPDATE as the only claim mutex
+four independent sweeper recovery paths
+fail-stop behavior after post-claim settlement failure
+enqueue, claim, folder, and upload idempotency boundaries
+Gmail History cursor and backfill safety net
+filesystem metadata flow and shared PVC constraint
+OAuth token encryption and refresh persistence
+KEDA queue-depth scaling and its limits
+Redis-loss recovery and simultaneous-store-loss boundary
+synthetic demo isolation
+real test strategy and load-test scope
+current war stories and the design lesson from each
+```
+
+Each answer must begin with the shortest sufficient response. Retain a second sentence only when it supplies the causal reasoning expected in an interview.
+
+- [ ] **Step 4: Ankify retained non-project cards**
+
+Use `apply_patch` to split prompts that test multiple independently forgettable facts, shorten wordy answers, add context labels where similar concepts interfere, and remove exact duplicates. Do not alter personal metrics or stories unless repository evidence directly contradicts them.
+
+### Task 3: Validate the Anki artifact
+
+**Files:**
+- Test: `All_Decks__System_Design__General.txt`
+
+- [ ] **Step 1: Verify header and field structure**
+
+Run:
+
+```bash
+awk -F '\t' 'NR <= 5 {next} NF != 5 {print NR ": " NF " fields"; bad=1} END {exit bad}' All_Decks__System_Design__General.txt
+```
+
+Expected: exit 0 with no output.
+
+- [ ] **Step 2: Verify required fields are populated**
+
+Run:
+
+```bash
+awk -F '\t' 'NR > 5 && ($1=="" || $2=="" || $3=="" || $4=="" || $5=="") {print NR; bad=1} END {exit bad}' All_Decks__System_Design__General.txt
+```
+
+Expected: exit 0 with no output.
+
+- [ ] **Step 3: Verify normalized prompt uniqueness**
+
+Run:
+
+```bash
+awk -F '\t' 'NR > 5 {key=tolower($3); gsub(/[[:space:]]+/," ",key); if (seen[key]++) print key}' All_Decks__System_Design__General.txt
+```
+
+Expected: no output.
+
+- [ ] **Step 4: Review formulation hot spots**
+
+Run:
+
+```bash
+awk -F '\t' 'NR > 5 && (length($3) > 140 || length($4) > 300) {print NR "\t" $3}' All_Decks__System_Design__General.txt
+rg -n '\?|;| · .* · | and | vs ' All_Decks__System_Design__General.txt
+```
+
+Expected: every long or apparently multi-target note is manually justified or split.
+
+- [ ] **Step 5: Run repository tests that support retained claims**
+
+Run:
+
+```bash
+go test ./...
+```
+
+Expected: all unit-test packages report `ok` and the command exits 0.
+
+Run when Docker is available:
+
+```bash
+make test-integration
+```
+
+Expected: the Redis/Postgres integration suite exits 0. If Docker is unavailable, report that exact limitation without claiming the integration suite passed.
+
+### Task 4: Export and verify the final file
+
+**Files:**
+- Read: `All_Decks__System_Design__General.txt`
+- Create: `~/Downloads/All_Decks__System_Design__General.txt`
+
+- [ ] **Step 1: Copy the validated artifact to Downloads**
+
+Run:
+
+```bash
+cp All_Decks__System_Design__General.txt ~/Downloads/All_Decks__System_Design__General.txt
+```
+
+Expected: the command exits 0.
+
+- [ ] **Step 2: Verify the source and export are identical**
+
+Run:
+
+```bash
+cmp All_Decks__System_Design__General.txt ~/Downloads/All_Decks__System_Design__General.txt
+shasum -a 256 All_Decks__System_Design__General.txt ~/Downloads/All_Decks__System_Design__General.txt
+```
+
+Expected: `cmp` exits 0 and both SHA-256 values match.
+
+- [ ] **Step 3: Report the completed audit**
+
+Report the final note count, notes added/removed, corrected volatile facts, structural-validation output, test output, export path, and any unverifiable personal claims intentionally preserved.

docs/superpowers/specs/2026-07-09-anki-deck-audit-design.md

@@ -0,0 +1,70 @@
+# Anki Deck Audit Design
+
+## Goal
+
+Update `All_Decks__System_Design__General.txt` in place so its DistTaskQueue
+material matches the current repository and every retained note follows effective
+spaced-repetition formulation. Copy the verified import file to
+`~/Downloads/All_Decks__System_Design__General.txt`.
+
+## Sources of truth
+
+Use current implementation and tests as the highest-priority source, followed by
+current operational artifacts and project documentation. Use `INTERVIEW-BRIEF.md`
+and the existing deck as inventories, not unquestioned authorities. Resolve any
+conflict in favor of executable behavior and measured evidence.
+
+Use the supplied SuperMemo twenty-rules article as the formulation standard:
+minimum information, one recall target per prompt, context cues, avoidance of
+sets and interference, useful redundancy, sources, date stamps for volatile
+facts, and prioritization.
+
+## Scope
+
+- Preserve the Anki text-import headers, Basic note type, deck path, tab-separated
+  fields, and tags field.
+- Audit every existing note for atomicity, prompt clarity, answer length,
+  duplication, interference, and import safety.
+- Verify every DistTaskQueue claim against the current repository.
+- Add missing high-value DistTaskQueue notes covering architecture, data flow,
+  invariants, recovery, retry behavior, idempotency, scaling, security,
+  operations, testing, limitations, and demonstrated incidents.
+- Retain useful non-project interview and distributed-systems notes, improving
+  their formulation without attempting to invent evidence for personal claims
+  that the repository cannot verify.
+- Remove exact duplicates and notes made obsolete by corrected replacements.
+- Do not change application code, project behavior, or unrelated files.
+
+## Card formulation
+
+Each prompt will include the shortest context label needed to prevent
+interference. Each answer will test one principal fact. Multi-part answers will
+be split when their parts can be recalled independently. Explanatory material
+will be kept only when it helps reconstruct the reasoning and will follow the
+direct answer.
+
+Stable concepts may appear from more than one useful retrieval direction.
+Volatile measurements and deployment facts will carry an `as of 2026-07` cue or
+tag. Source tags will identify repository evidence without requiring source
+paths to be memorized.
+
+## Verification
+
+The final audit will check:
+
+1. Every non-header row has exactly five tab-separated fields.
+2. Note type, deck, prompts, answers, and tags are non-empty.
+3. Prompts are unique after normalization.
+4. No accidental multiline records or malformed quoting remain.
+5. DistTaskQueue numbers and behavior match current code, tests, configuration,
+   documentation, or checked-in measurement artifacts.
+6. Relevant Go tests and any existing deck-validation tooling pass.
+7. The workspace file and Downloads export are byte-identical.
+
+## Deliverables
+
+- Updated `All_Decks__System_Design__General.txt` in the repository root.
+- Identical export at `~/Downloads/All_Decks__System_Design__General.txt`.
+- A concise completion report listing card counts, important factual changes,
+  validation results, and any claims left intentionally unchanged because the
+  repository cannot verify them.