Backend System Design

---
name: backend-system-design
description: |-
  Produce a staff-grade backend system-design RFC for the active task: a 12-section design document (problem, requirements, architecture, data model and storage, API contract, caching, scaling and bottlenecks, reliability and SLOs, security, observability, rollout and migration, trade-offs) for a new service, endpoint, or backend feature, persisted as BACKEND_SYSTEM_DESIGN.md. Capability-routed, not stack-specific; composes with slo-define, performance-budget, api-contract-review, and release-plan rather than duplicating them. Use when a backend service or feature needs an architecture-level design before planning and slicing. Do not use to slice an already-designed change (use implementation-plan), to analyze blast radius (use impact-analysis), to review an API contract in isolation (use api-contract-review), to design the frontend surface (use frontend-system-design), or with no active task folder (run task-init first).
metadata:
  category: "discovery-and-scoping"
  primary-cursor-mode: "Agent"
  multi-repo-aware: "false"
  context-layers-consumed: "memory"
  context-layers-produced: "memory"
  tools: "Read, Write, Edit, Bash, Glob, Grep"
  x-wos-profiles: "full"
  provenance: "first-party"
  suggested-model: "claude-sonnet-5-5"
---

Act as a staff backend engineer writing the system-design document for a new service, endpoint, or backend feature, so the architecture is decided and reviewable before any slicing or code.

Goal:
Produce a 12-section backend system-design RFC for the active task, grounded in the task's decisions and constraints, and persist it as `BACKEND_SYSTEM_DESIGN.md`. The command is capability-routed: it designs backend systems on any stack (a monolith route, a service, a queue worker, a serverless function) and is not tied to one language or framework. It is the backend sibling of `frontend-system-design`, and it composes with `slo-define`, `performance-budget`, `api-contract-review`, `migration-safety-steward`, and `release-plan` rather than duplicating them.

Mandatory context bootstrap (before any output):
- Read these sections in `WORKFLOW_OPERATING_SYSTEM.md` first:
  - `## LLM execution contract`
  - `## Editor mode policy` (mode definitions only; the tool mapping table is lazy-loaded in `wos/editor-mode-mappings.md` and needed only for non-Claude-Code tools)
  - `## Global output contract` (including **Adaptive handoff** and **Mode selection rule**)
  - `## Cross-cutting workflow guardrails`
- **Bootstrap tiers:** the light-weight commands (`branch-commit`, `what-next`, `where-we-at`, `slice-closure`, `compact-task-memory`) plus the high-frequency `implement-approved-slice` and `sync-task-state` (v3 wave1 item D) read the four sections above with two subsections of `## Cross-cutting workflow guardrails` skipped: `### External web access (centralized)` and `### Sequencing heuristics (by phase)`. Everything else is read at every tier, including `### Proposal vs approved persistence` and `### Substrate peer ownership (per ADR-0034)`, since all seven write substrate sections and reason about PROPOSED (`state-reconcile` stays on the full tier for cross-artifact judgment). The full tier is measured at 12109 tokens, the four always-read sections combined; the two skipped subsections are 1,034 of those (measured 2026-09-28), so the reduced tier is about 11,074. The leaf-reviewer tier (`verify-against-rubric`, ADR-0226) reads only `## Global output contract`, measured at 5215 tokens, plus its rubric.
- **Session bootstrap reuse (skip-if-unchanged; v3 wave1 item D):** WHEN this conversation already read the bootstrap sections in an earlier turn still VISIBLE in the context window AND `WORKFLOW_OPERATING_SYSTEM.md` has not changed since, the command MAY skip the re-read and cite the earlier one, emitting one Command transcript line: `Bootstrap: reusing turn <N> read, WOS unchanged`. Scoped exception to the context-budget re-fetch rule (`wos/context-budget.md`, "The re-fetch rule"), because these sections are one large, static, byte-identical read repeated every turn; every other tool result still re-fetches. VISIBLE means the section text itself is still present and quotable now, not merely that a record of the earlier read exists: a harness that clears a tool result while the record survives (ADR-0114) has not satisfied VISIBLE, and self-declared memory after a compaction never qualifies. A stateless-per-turn harness is excluded. The transcript line is mandatory; a silent skip is invalid output.
- **Resolving the spec and `wos/<topic>.md`:** canonical repository FIRST. Otherwise use active-adapter docs: `~/.claude/workflow-docs` (`CLAUDE_WORKFLOW_DOCS_DIR`), `~/.cursor/workflow-docs` (`WORKFLOW_DOCS_DIR`; project `.cursor/workflow-docs` on project installs), `~/.codex/workflow-docs` (`CODEX_WORKFLOW_DOCS_DIR`), or `<kimi-home>/workflow-docs` (`KIMI_WORKFLOW_DOCS_DIR`). Nonempty overrides win. Kimi home: `KIMI_CODE_HOME`, else existing `~/.kimi-code`, else existing `~/.kimi`, default `~/.kimi-code`. Use one root for both paths; try other adapters if absent or unknown. Report resolved root, fallback to other adapters, or missing mandatory context in `### Command transcript`. Installed snapshots refresh and prune retired topics only on sync.
- Read additional sections only when relevant to this command's role.
- Align all routing recommendations and next-command suggestions with the current command set.
- **Official next-command names only:** recommendations, including `Run now`, use registered names from `commands/<name>.md` or `commands/<name>/SKILL.md`; installed mode uses the available Fhorja command or skill inventory. No invented aliases. `Run now: none` with `Mode: N/A` ends only if no honest next step exists, defined under `### Official command names (routing integrity)` (ADR-0126).
Required inputs:
- active task folder path
- TASK_STATE.md, SOURCE_OF_TRUTH.md, DECISIONS.md (the design must respect locked decisions)
- IMPACT_ANALYSIS.md and INVARIANTS_AND_NON_GOALS.md when present
- the service, endpoint, or backend feature to design, named in the task or the prompt
- the expected scale signal when known (users, request rate, data volume, read-to-write ratio); mark unknowns rather than inventing them
- relevant external references already captured in `projects/<client>__<project>/REFERENCES.md` (read-only grounding)

Task repository files to create or update:
- `BACKEND_SYSTEM_DESIGN.md` in the active task folder: the 12-section RFC.

Operating rules:
- Do not implement product code. This command produces a design document, not a slice plan and not an implementation.
- **Handoff:** end with the adaptive `### Handoff` block per `WORKFLOW_OPERATING_SYSTEM.md` `## Global output contract` (Mode A compact or Mode B full).
- **Capability-routed, not stack-locked.** Design for the stack the task actually uses (read it from SOURCE_OF_TRUTH.md or DECISIONS.md). Name a specific database, queue, or runtime only when the task already chose it; never assume one. The structure below holds for any backend stack.
- **Scale-honest, not scale-cargo-culted.** Design for the task's real scale, not FAANG scale. A single Postgres instance is the correct answer for most solo and small-team work; reach for sharding, multi-region, or a message bus only when a stated requirement forces it. Do not import distributed-systems machinery a solo builder will never operate. When a number (users, request rate, data volume) is not known, mark it `unknown` and design for the plausible near-term, not an imagined peak.
- **The 12 sections.** Produce every section; mark a section `not applicable` with a one-line reason rather than dropping it:
  1. Problem statement and context: the user or business problem this backend serves, scope boundaries, who calls it.
  2. Requirements: functional and non-functional, split explicitly; core versus nice-to-have; success metrics; the expected scale (users, request rate, data volume, read-to-write ratio) or an explicit `unknown`.
  3. High-level architecture: the services or components and their relationships (entry point, application layer, data layer, background workers), and the synchronous-versus-asynchronous boundaries. Name the trade-off pairs at play (see `wos/architecture-tradeoffs.md`).
  4. Data model and storage: entities, the schema shape, the storage engine choice and why, indexing, and the consistency model. Compose with `db-context-postgres` or `db-context-supabase` when a real schema exists.
  5. API and interface contract: transport (REST, RPC, GraphQL, or an event), payload shape, pagination, error and retry semantics, idempotency, and versioning. Compose with `api-contract-review` for the endpoint-level audit rather than duplicating it.
  6. Caching and data access: the read and write paths, and the cache update strategy when a cache is in play, named from `wos/cache-update-strategies.md` (cache-aside, write-through, write-behind, refresh-ahead) with its failure mode guarded.
  7. Scaling and bottlenecks: the likely bottleneck under the stated scale (the database, a hot path, an external call), whether the design scales vertically or horizontally, and what statelessness or partitioning that requires. Name the one bottleneck you would hit first and how you would relieve it.
  8. Reliability and SLOs: the availability and latency targets, the failure modes, and the resilience posture (timeouts, retries with backoff, circuit breakers, idempotent retries). Compose with `slo-define` for the target artifact rather than restating it.
  9. Security: authentication and authorization model, secret handling, input validation, data protection at rest and in transit, and multi-tenant isolation when relevant. Route deep review to `security-review` or `rls-auth-boundary-auditor`.
  10. Observability and operations: the logs, metrics, and traces that make this service debuggable in production, the key alerts, and the one dashboard you would watch on launch day.
  11. Rollout and migration: deploy strategy, feature flags, backward compatibility, and schema-migration safety. Compose with `release-plan` for the rollout and `migration-safety-steward` for any DDL rather than duplicating them.
  12. Trade-offs and alternatives: the options considered and why the chosen design wins, citing the named axes in `wos/architecture-tradeoffs.md`. This is the section that separates a design from a policy statement; never leave it empty.
- **Ground external contracts.** When the design commits to an external library, SDK, datastore, or protocol, ground it in a captured `REFERENCES.md` entry; when it is not captured, name the gap and route to `capture-references` rather than designing the contract from memory. The captured entry wins over recollection (per `WORKFLOW_OPERATING_SYSTEM.md` `## Evidence priority`).
- **Respect locked decisions.** Read DECISIONS.md; the design must not silently reopen a locked decision. When the design needs a decision that is not yet made, label it `PROPOSED` and route to `decision-interview` instead of asserting it.
- **No invented metrics.** Scale numbers, latency targets, availability SLOs, and cost figures must cite a source (a measured baseline, a published standard, or a user-supplied target) or be marked `PROPOSED-pending-baseline`.
- Treat task-memory write policy per `WORKFLOW_OPERATING_SYSTEM.md`: write the file and mark it `APPLIED`.
- **Self-review before emit.** Before writing the file, check it for placeholders, contradictions, an empty trade-offs section, distributed-systems machinery unjustified by a stated requirement, and any section asserting a number without a source; fix them inline.

Required output:
1. The service, endpoint, or feature being designed and the stack read from task memory
2. The 12-section design, every section present
3. Exact `BACKEND_SYSTEM_DESIGN.md` content, marked `APPLIED`
4. Any `PROPOSED` decision the design surfaced, with the upstream command to lock it
5. Recommended next command
6. Recommended editor mode
7. Why that is the correct next step

### Reference grounding (design gate)
**Reference grounding (design gate, ADR-0043 D-3).** This command does not edit code, so it does NOT refuse. It marks. Every external contract this document NAMES as a choice, a recommendation, or a thing to build against is either grounded in a captured reference or carried forward as visibly unproven.

1. Detect. Any external library, SDK, API, protocol, or vendor behavior this document names as a decision input: a version to adopt, an endpoint shape to build against, a rate limit to design for, a capability to depend on.
2. Ground or mark. WHEN the contract is present in `projects/<client>__<project>/REFERENCES.md`, read that entry and cite it inline where the decision is stated. WHEN it is absent, keep the recommendation and append `[ungrounded: <contract>]` to that line. Do NOT drop the recommendation, and do NOT silently assert the behavior: an unproven choice a human can see beats a confident one they cannot.
3. Never launder recollection into a design. A version number, a parameter name, or a rate limit recalled from training is outside the grounded set (`wos/active-epistemic-humility.md`). It may appear as a proposal, marked, never as a stated fact.
4. Route, do not block. WHEN two or more contracts are ungrounded, name `capture-references` in the handoff as the next command. This gate exists so `implement-approved-slice` is not the first place the gap is discovered, at which point the plan is already approved and the execution gate refuses mid-slice.
5. Carry the marks forward. Every `[ungrounded: ...]` marker SHALL survive into the persisted artifact. A design doc that resolved its own markers by deleting them has defeated the gate.
### Claim grounding (active epistemic humility)
**Claim grounding (active epistemic humility).** This block governs what you may assert and how you record it. It is keyed to the substrate section you are writing, not to which command is running, and it is INERT on any output that writes none of the claim-bearing sections below. Full contract and rationale: `wos/active-epistemic-humility.md`.

1. When this applies. This block fires ONLY while you are writing a claim-bearing substrate section: `TASK_STATE.md ## Current known facts`, `## Risks to watch`, `## Observations`, `## Active files in scope`, `## Canonical decisions`; `DECISIONS.md ## Locked decisions`; `IMPLEMENTATION_PLAN.md ## Current gaps`, `## Risks and mitigations`; `IMPACT_ANALYSIS.md`; `EXTERNAL_RESEARCH.md`; `REFERENCES.md`; or any section whose content is a statement a later command or a human decision will act on. WHEN your output writes none of these, this block imposes nothing: skip it and proceed. This is the D-13 inert clause; a fully-grounded or claim-free output pays nothing.

2. The unit is the load-bearing claim. A load-bearing claim is one a downstream command or a human decision consumes. A passing aside is not load-bearing; a statement someone will act on is. Apply the rest of this block per load-bearing claim, not per sentence.

3. Ground it or abstain. Before you assert a load-bearing claim, trace it to the enumerable grounded set: a captured `REFERENCES.md` entry, a file read in this session, command output actually seen, or a passing deterministic gate. A claim supported only by model memory is OUTSIDE the grounded set, including when you are right, because that support is not observable. WHEN a load-bearing claim falls outside the set, do NOT assert it: either investigate until it is grounded, or abstain per rule 6.

4. Status records provenance, never confidence. WHERE you attach an epistemic status to a claim, the status names WHERE THE CLAIM CAME FROM: a `REFERENCES.md` entry title, a file path plus line, or the gate output it came from. It SHALL NOT express a degree of certainty. Do NOT add a confidence field, a numeric threshold, or a self-assessment prompt anywhere; a self-reported confidence signal is not a usable control signal (`wos/active-epistemic-humility.md` Part 1.3). A status whose referent slot is empty is read as UNKNOWN, not as a weak yes.

5. Persisted claims carry the status; chat-only claims carry it when they route. Every load-bearing claim you write into a task-memory artifact carries its provenance referent, and that referent travels with the claim so a later command reads it too; do not drop it at the write boundary. A load-bearing claim that appears only in a chat-turn output carries a status only when it crosses the grounding boundary and triggers a route (an abstention, an escalation).

6. Abstain as a routed continuation, never a bare refusal. WHEN you abstain, name the specific investigation that would settle the question AND route to the command that runs it (`capture-references`, `code-locate`, `incident-triage`, or the fitting one). A withholding that stalls the work is invalid output. Abstention is distinct from `NO_OP`: `NO_OP` means there is no work to do; abstention means there is work and the grounding to do it is missing.

7. An unfired gate is not evidence. The absence of a fired check does not mean grounding existed. Do not read silence here as a pass.
### Standard output layout (required)
Produce the command output using this structure (English only):

### Artifact changes
Follow `## Global output contract` in `WORKFLOW_OPERATING_SYSTEM.md` for `APPLIED` / `PROPOSED` / `SKIP` rules. Every listed file carries one of those three tokens, in Lean output too; a prose verb like "written" is not a label.

### Command transcript
Brief audit trail (max 4 lines; max 3 in no-op runs with `NO_OP_TRACE`).

### Handoff
Use the adaptive ending format of `WORKFLOW_OPERATING_SYSTEM.md` `## Global output contract`. Every Handoff is one fenced `text` block with all four lines, `Run now:`, `Mode:`, `Work complexity:` and `Reason:`, on a stop and on a refusal too; the terminal form is `Run now: none` with `Mode: N/A`. A new `Mode:`, a model or a fresh session is never a stop, an offered choice is one, and `Reason:` names a role, never a model.
### Definition of done (command output)
- All 12 sections are present; a dropped section is invalid output unless marked `not applicable` with a reason.
- The trade-offs and alternatives section is non-empty and names the rejected options, citing the relevant axes in `wos/architecture-tradeoffs.md`.
- No scale, latency, or SLO number is asserted without a cited source or a `PROPOSED-pending-baseline` mark.
- The design is scale-honest: no sharding, multi-region, or message-bus machinery appears without a stated requirement forcing it.
- The design respects DECISIONS.md; any decision it needs but does not have is marked `PROPOSED` and routed to `decision-interview`, not asserted.
- The artifact is written and marked `APPLIED` in every mode (ADR-0199).
- Output ends with a complete `### Handoff` block per the adaptive format in `WORKFLOW_OPERATING_SYSTEM.md` `## Global output contract`.
- Before declaring this output done, confirm it satisfies the shared **Definition of done (command outputs)** and **Gate conditions** in WORKFLOW_OPERATING_SYSTEM.md.

Quality bar:
A reviewer can read the document and understand the backend architecture, the data model, the contracts, the failure modes, and why this design beats the alternatives, with no number asserted on no evidence, no locked decision silently reopened, and no scale machinery the task does not need.