Requirements → design (the bridge)
Parent: standmeet
Purpose
The missing middle between product and architecture — how the product surface decomposes into the seven pillars, traced BOTH ways: every promise to its engineering, every pillar back to its justifying promise. Also records which product docs are still authoritative after the June 2026 pivot.
The chains (vision → promise → journeys → design → code → verification)
Each child walks ONE promise all the way down, with concrete identifiers at every level:
- chain-interpretation — a visitor asks; voiced, grounded, role-framed answers. (Fully served since the crawl face landed —
corpus_linkswalks the owner's own links.) - chain-explicit-push — AI conversations → curated corpus via MCP + the vault route.
- chain-selective-access — who may see what, frozen at issue. (The deepest chain: finalized design, heaviest tests.)
- chain-owner-first — the external brain before the audience. (Structurally true; retrieval-thin.)
- chain-machine-readable — when the asker is an agent. (Vision-strong, engineering-thin.)
- chain-hiring — the first application, not the product. (Watch the boundary.)
- chain-sovereignty — your infra, your keys, your trust boundary. (Adversarial verification thinnest; also where the platform migration's implicit justification lives.)
Which product docs are still true
| doc | verdict |
|---|---|
docs/features-and-journeys.md (last touched 2026-09-06, 7037a434e) | LIVE — the product surface of record: personas Owner / Visitor / Cross-actor (numbered journey sets); every feature carries an e2e spec tag; now carries the vault journeys (A12 sync, A33 crawl), microsites, and per-microsite SEO |
docs/design/project/IDEAS.md (May 30) | LIVE backlog — resume=git model (git vocabulary dropped from UI per 2026-05-28 decision), application bundles, the agentic-chat-capabilities idea set; nothing built yet |
docs/product-vision.md (May 16) | HISTORICAL — predates the June 2026 pivot and rests on a direction since killed; do not use. Only the "digital-avatar app = existing code" mapping survives |
docs/growth.md (May 16) | MISFILED — an OpenClaw communication-theory case study, not StandMeet growth; its five wedge principles are transferable but wired to nothing |
The pipeline this repo actually uses
The requirement→design transition mechanism, as practiced: journey/feature in features-and-journeys.md carries an e2e spec tag ([✓ gate-access], [ ] = uncovered; a gap section ranks priorities) → load-bearing decisions get a design doc with pinned markers (P.x / D-x / §x / A.x) → decisions are frozen by red-contract tests before code → roadmap blocks sequence the build. In harness terms: the spec tags are the gate list; red tests freeze the split; the roadmap is the budget schedule.
Journey clusters → pillars
| journey cluster | journeys | pillar(s) | key designs |
|---|---|---|---|
| Owner setup & auth | setup wizard · login · reset | access-control + structure | setup-token claim, owner-keypair-auth |
| Owner curates the corpus | the raw→wiki→output promotion journeys | corpus | three-tier-corpus-promotion, writings-import-export, derived-path-corpus-as-filesystem |
| Codes & invitations | code management · gate entry · quota | access-control | acl-and-quota-granularity, role-snapshot-frozen |
| Visitor chat & public pages | the visitor journey set | agent-core + corpus (render face) | entry-agnostic-agent, ui-cards, corpus-retrieval, unbuffered-sse-passthrough |
| Job loop (outbound) | jobs · resume · applications · the QR loop | the backend/internal/owner/jobs/ domain module (fetch / jobsuc / jobsmcp / jobsadmin / printsess) + connector | the job-loop design doc |
| Connectors admin | the connector admin journeys | connector | connector-plugins, connector-deps |
| Skills & conversations | skills admin · conversations review | capabilities + agent-core | skills-progressive-disclosure |
| API·MCP (machine-readable outlet) | admin API·MCP surface | capabilities | as-mcp-facade, owner-keypair-auth |
| SEO / sitemap / microsites | public surfaces · robots + sitemap · microsites at /p/ (renamed from "custom pages" 2026-09-05, e7fe80e91; SEO is per-microsite since 7037a434e) | corpus + structure | landing/reader, builder |
The orphan audit (both directions)
Promises with no engineering (mostly fine): everything in the pre-pivot product-vision.md stack (HISTORICAL — died with the pivot, not worth enumerating); the IDEAS agentic-chat capabilities (live backlog, unbuilt); growth.md's five wedge principles (unwired — no distribution feature exists).
Engineering with no product journey (the actionable gap, as first found): the corpus-as-vault direction — graph retrieval over wiki_refs, folder collapse, callout/TikZ/widget rendering, CSS-snippet sync — was the roadmap's corpus-as-vault core promise yet had NO journey in features-and-journeys; the platform-ification (MCP externalization, ui://, registry) likewise architecture-driven with no journey; sysroutes/captcha/RSS uncovered [ ] with no journey. Conclusion then: features-and-journeys.md needs a vNext adding the vault journeys (owner: author-in-Obsidian → sync → verify render; visitor: navigate the linked wiki / follow backlinks) — otherwise the roadmap's biggest bet has no product-side acceptance criteria — you get what you can check. Since filled: the doc now carries A12 (vault sync, ten sync-* specs) and A33 (crawl face, retrieval-links); what remains [ ] at 7037a434e is LockedView, microsite rollback, Turnstile captcha, and the login throttle — RSS never shipped (robots + sitemap did, e2e/test/seo-feeds.spec.ts).
Closing thought
The bridge exists as a practice (spec tags → design docs → red tests → blocks); this page makes it inspectable — its first inspection found the flagship promise missing its journey, and the journey has since been written.