Spaced Repetition Email for Busy Professionals

Spaced repetition email is a calendar of retrieval prompts, not a streak of "daily tips." Busy professionals do not fail because they lack Anki decks; they fail because the next review never lands in a surface they already open. This tutorial shows how to map SuperMemo-style intervals onto a small email curriculum, how to write prompts that actually retrieve, and how to run the series on Cadensend — self-hosted, MIT-licensed, currently delivering only to your verified address.

Key Takeaways:

• Spacing without retrieval is just a slower newsletter; every review issue must force recall of a prior objective.
• Use expanding intervals (1–2–4–7–14 days) for procedures; keep daily cadence only while the stack is still being built.
• Ground review issues in the same sources as the original lesson so "answers" cannot drift.
• Cadensend plans modules and delivers exactly once; it is not Anki, not a CRM, and not a bulk sender.
• Cap each review at one question plus a cited reveal, or busy readers will archive the thread.
• Rehearse the schedule in your own inbox before you ever think about other recipients.

Table of Contents:

• What Spaced Repetition Email Means
• Why Inbox Beats Another Deck
• Interval Design for Working Calendars
• Write Retrieval Prompts That Bite
• Plan a Series That Mixes Teach and Review
• Grounding So Reviews Stay True
• Worked Example: Kubernetes Debugging
• Worked Example: Security Policy
• Delivery, Timezones, and Idempotency
• Frequently Asked Questions

What Spaced Repetition Email Means

Spaced repetition email is a scheduled sequence where later issues ask you to reconstruct earlier objectives, with expanding gaps between those asks. It is not "we email you every day so you remember us."

Classic spaced repetition (Leitner, SuperMemo SM-2) tracks per-item difficulty. A full SM-2 scheduler in a marketing tool is the wrong architecture: you would need per-learner state, grades, and a list you are allowed to mail. Cadensend's MVP has none of that — and should not pretend to. What you can do today is author a fixed schedule of teach issues and review issues, send it to yourself, and adjust the plan when you fail a prompt.

That is enough for a staff engineer learning one stack. It is how a daily email course becomes memory instead of content.

Cepeda et al. summarize that expanding or spaced practice beats massed practice for long-term retention; a readable entry point is the spacing effect. Your job is to put the next trial on a day you will still open mail.

Why Inbox Beats Another Deck

The inbox already has a daily ritual; a new SRS app does not. Anki is excellent if you will open it. Most professionals will not add a third flashcard habit on top of Slack and calendar.

Email vs video is covered in email drip course vs video retention. Spacing is why drip can beat binge even when video is "better teaching" in the moment.

Keep cards in the issue, not as a link to a 40-card deck. One prompt. One cited answer below a break, or in the next day's first paragraph if you want a harder delay.

Interval Design for Working Calendars

Build the skill daily, then review on expanding gaps that skip weekends you never read mail. A schedule that assumes Saturday 7am study will desync by week three.

A practical template for one skill (not a university course):

That is not SM-2. It is a curriculum-shaped spacing plan you can lock in Plan Studio. Cadensend validates coverage and prerequisite order so review A cannot require term C before C is taught. Educators who need Bloom-aligned objectives should read grounded email curriculum for educators.

Timezone: store UTC plus IANA zone. A 07:15 Europe/London review should not jump an hour across DST and land in a standup. Cadensend's Deliver pillar keeps local intent.

Do not schedule 14 reviews for 14 facts in one week. Cognitive load (Sweller) still applies. Three live items is plenty.

Write Retrieval Prompts That Bite

Ask for a production decision, not a definition you can grep. "What is a CrashLoopBackOff?" is weak. "A pod restarts every 20s after a config change. Name the first three kubectl commands from our runbook, in order." is a prompt.

Prompt patterns that work in email:

1. Procedure: ordered steps from your doc.
2. Contrast: "Why B-tree here and not hash, given query Q?"
3. Diagnosis: paste a redacted log line from the source PDF, ask the cause.
4. Refusal: "Which of these indexes would we not add, and which section says so?"

Put the cited reveal after the prompt. The citation must be a real chunk: section anchor in the runbook, not "industry best practice." That is the same renderer rule as cite sources in educational emails.

If the model writes the prompt, it must retrieve the chunk first. Otherwise you get unanswerable riddles or answers that contradict the PDF. That is a RAG garbage failure, not a pedagogy failure.

Structured issue output (objective, prompt, reveal, citations) should be a schema, not markdown the model free-styled. See structured LLM output. Cadensend's Write pillar returns a validated content tree and a deterministic HTML/text renderer.

Plan a Series That Mixes Teach and Review

Label each issue teach or review in the plan, lock approved teaches, and never regenerate a review so hard that it cites a rewritten teach. Cadensend lets you lock issues so a planning revision cannot clobber them.

Planning rules:

1. One learning goal. Exclusions listed. Audience level honest ("busy senior, not new grad").
2. Teach issues introduce at most one new procedure.
3. Review issues introduce zero new procedures. They may introduce a new scenario that still maps to old steps.
4. Mixed review only after each item had one successful isolated review.
5. Subject lines name the job: Review: CrashLoopBackOff first three commands not Quick thought.

This is closer to a bounded agentic workflow than to a chatbot: Plan graph → Ground → Write → human gate → Deliver. HinterBuild's AI agent development work uses the same bounded-graph rule. Cadensend is not an autonomous agent that invents a new syllabus overnight.

Onboarding series overlap: a 7-day email onboarding course can put reviews on days 4 and 7 instead of seven teaches in a row.

Grounding So Reviews Stay True

Reviews must retrieve from the same series-scoped corpus as the original lesson. If teach A cited runbook §4.2, review A's reveal must land on §4.2 or an explicit errata you added as a new source.

Ungrounded reviews are how "spaced repetition" becomes spaced hallucination. Read LLM hallucination causes and fixes. Scope queries to workspace and series before the vector store, then diversify by source — Cadensend's Ground pillar, and the same hygiene as production RAG systems.

Chunk quality still dominates. Overlapping policy paragraphs in one chunk make the reveal cite the wrong control. Use chunking strategies. Embeddings are not magic.

Ingested PDFs can contain prompt-injection payloads. Retrieved content is data, never instruction. See prompt injection and Ocherfort if you self-host models and tools. Cadensend's writer has no arbitrary network and no code execution.

If your notes are PDFs, the ingest path is turn PDF notes into an email course. OCR is a later Cadensend roadmap item, not something to assume in the MVP.

Worked Example: Kubernetes Debugging

Example 1. Goal: "From a CrashLoopBackOff, reach a documented next action in under five minutes." Sources: your platform runbook PDF and Kubernetes docs on pod lifecycle. Cadence: 07:20 America/Chicago. Audience: application engineers, not SREs.

Issue 8's reveal must quote the runbook's order, not a blog the model remembers. If retrieval misses, the writer should refuse, not improvise.

Run this to yourself with Cadensend. There is no hosted signup and no student list in the MVP. After you survive day 15, you may export the locked issues into an internal wiki. That export is outside the engine.

Worked Example: Security Policy

Example 2. Goal: "Classify a data request as allowed, legal-review, or deny using policy §2." Sources: the policy PDF only. Exclusions: other companies' policies, "general GDPR takes."

Teach days 1–3 walk the three buckets with tables. Review days 5, 9, and 16 give a new request vignette that still maps to §2. Citations are the entire point; this is teach with sources, not vibes.

If you are comparing tools, a marketing suite can drip "security tips." It will not enforce series-scoped retrieval or structural citations. See self-hosted email course vs ConvertKit. ConvertKit can mail lists you already operate; Cadensend currently cannot and does not claim to.

PII scrubbing belongs in vignettes. Do not paste real ticket text into a source you will embed.

Delivery, Timezones, and Idempotency

A review that arrives twice is worse than one that arrives late: you cannot tell whether you remembered or reread. Cadensend writes the delivery record before the provider call. Keys: workspace, issue, recipient, issue version. Retries replay. Jobs are claimed with FOR UPDATE SKIP LOCKED — see SKIP LOCKED and idempotency. The same discipline shows up in data pipelines.

Approve every issue. Human-in-the-loop gates are how you catch a review that leaked a new (wrong) command. Output guardrails are the machine layer, not a replacement.

Cadensend is not a newsletter blaster. If you need list semantics, that is a different product and a later roadmap (consent, suppression, unsubscribe). Do not configure the MVP as if Update 2 already shipped.

About HinterBuild: we design grounded generation and durable delivery. Contact if you need this architecture in your stack. A newsletter-shaped habit is still a newsletter vs course decision; spacing does not turn a roundup into a curriculum.

Frequently Asked Questions

What is spaced repetition email?

It is a planned email sequence that re-asks earlier objectives after expanding gaps, so you retrieve instead of reread. It is not a daily motivation drip. Cadensend can schedule those issues to your verified inbox with exactly-once delivery; it does not run a per-learner SM-2 engine in the current MVP.

Can I do real SuperMemo SM-2 over email today with Cadensend?

Not as a hosted, multi-learner scheduler. The MVP sends only to you. You can approximate SM-2 by editing your plan when you fail a review (shorten the next gap). Per-reader branching is later roadmap, not current product.

How many review questions per email?

One. Two if they share a citation. Busy professionals treat a quiz packet as homework and archive it. Put extra items in a later review issue.

Should reviews be daily?

Only while you are still laying down the stack (teach days). Once items exist, expanding intervals beat daily rereads. Daily rereads feel like diligence and produce the illusion of memory.

Do I need an LLM?

No. Hand-written review issues with honest citations work. An LLM helps draft from a large PDF set if — and only if — retrieval is scoped and citations are rendered from chunk IDs. Otherwise you are spacing errors.

How is this different from a newsletter?

A newsletter can include a "from last month" blurb. Spaced repetition email has a locked objective, a retrieval prompt, and a cited reveal. See newsletter vs email course.

What if I miss a week of mail?

The schedule should still send exactly once per issue, not "catch up" with four reviews in a day. Massed catch-up undoes spacing. Read the backlog as a stack, one issue per sitting, or regenerate a new mixed review from the same locked teaches.

Is Cadensend a flashcard app?

No. It is a curriculum engine: Plan, Ground, Write, Deliver. Self-hosted, MIT, source on GitHub, no hosted signup, not a CRM or bulk sender.

Conclusion

• Spaced repetition email puts retrieval on a calendar you already open.
• Expand intervals after the teach stack exists; do not daily-reread forever.
• One prompt per issue, cited reveal, series-scoped sources.
• Lock teaches so reviews cannot drift.
• Cadensend rehearses this to your verified address, exactly once — not to a list.

Start with Cadensend, clone the MIT repo, and contact HinterBuild if you need production RAG or send architecture.

Resources:

• Spaced repetition (overview)
• Spacing effect
• Cognitive load theory
• Kubernetes pod lifecycle
• Cadensend on GitHub