One index over ~50 specs, designs, and plans — sorted by whether they still govern the work or just record how we got here. It opens with where we actually are today (which the roadmap, written two days before the Flue pivot, no longer shows), lays out the launch path, and links every doc in the tree.
We are on Flue — the whole agent fleet runs as Cloudflare Durable Objects, cells retired. The advisor agent is live and self-driving (born from a real intake, a private Turso brain, hourly tend + nightly scout + dream). The market agent runs a thin slice for one market (sweep → Turso → cards → briefing-on-zero). Everything else is funnel polish, market depth, and — the new long pole — nineteen more market agents.
Reframe, 2026-06-24: we launch in 20 hand-selected markets, each market agent built bespoke. "Born-anywhere" market birth — the biggest item on the old roadmap — is off the launch path. A general "stand one up anywhere" process may fall out of doing 20 by hand; it isn't a gate.
This is the old 12-day-build spine, re-cut for the 20-market reframe. M1 (advisor) is done; "market anywhere" is replaced by "build 20 markets by hand"; the home-agent rungs move past launch. Progress bars are rough, honest first passes.
/visit capture surface + 5 hand-authored surveys liveNear-term build queue first, then the launch-scale work, then the deferred. Each links to the doc that specs it. This is the honest "not done" list — not the vision.
Killed the money slider, per-card meters, the HomeZero score, and the confidence-bar chart; advisor now speaks in prose (read{summary,points}), market numbers in one 5-tile billboard. Shipped + live on www (PR #55 → e79f48c). → cutover plan
Awakening square + live roster on /home; strip the deterministic seed voice so everything perceived as the advisor is agent-authored. → first-run plan
The market agent authors the 5 billboard tiles nightly (on-market, typical asking, median DOM, over-record, honest deals). funnel-v2 + read-led both depend on it. → funnel-v2
Separate the advisor-context questions from the market-context questions in the wizard. → onboarding
Build each on the Slice-2 pattern; provision Turso + Bright Data + deploy per market. The launch long pole. → slice-2 plan
County records ingest, native vectors, comps wiki, metric skill — the substance behind the "$25 deep scan." → market-agent-spec
Clerk subscription + paywall + free-tier meter; landing moves off the one-time Deep Scan framing. → marketing-brief
The advisor's daily push — blocked on a real WhatsApp outbound channel (number, opt-in, Meta template). Armed-but-empty stub in place. → whatsapp-bridge
Codify the happy-path acceptance click-through (land → onboard → reflect → finish → board → zero) in Playwright. → onboarding-smoke-test
Per-house agent born on scan, the visit ingest loop. Explicitly post-launch. → home-twin
Six shelves. Canonical docs drive current and forward work. Active plans are ready to execute. Invariant docs hold the thesis that survived every pivot. Historical docs are kept for lineage — don't build from them. Links are relative; .md files open as raw text.
/home — install hero, WhatsApp card, the 5-tile market billboard, read-led board preview. Folded into the read-led cutover./visit + 5 surveys live; the ingest loop is WS6 (home agent).homezero, dev Convex nautical-rabbit-780; the pre-launch prod checklist.Why the historical shelf is so deep: the design moved fast. Each pivot left docs behind. The full narrative lives in roadmap.html — this is the short version, now with the fourth turn the roadmap predates.
Coverage born from buyers, not geography. The county canvass dies; three births (advisor / market / home).
Specialists killed. The standing team collapses to advisor (person) + market (place) — skills, not sub-agents.
An intent-gated advisor → market → home ladder; cost-to-zero as the spine. (Linear-dogfood since dropped.)
The whole fleet → Cloudflare Durable Objects. Cells retired. The substrate everything now runs on — and where the roadmap stops.
Start at Where we are for the snapshot, The launch path for the plan, and What's left for the queue. Use The doc map to find any spec by what it governs — build only from Canonical and Active plans; read Invariant for the why; treat Historical as lineage. This wiki is the live index — update it when a doc's status changes, not the (frozen) roadmap.