This roadmap separates durable direction, a status snapshot dated 2026-08-12, and the launch blockers on record as of 2026-09-10. It is not the live backlog, a start-date promise, or proof that every listed capability is deployed. See Current System State for the evidence ledger.
Status vocabulary
Durable direction
Launch blockers on record — 2026-09-10
These are the gates the founder or Main named on Linear between a working Development and any production use. Each is dated by its record; a Linear status is not the evidence.
Direction the founder set in September, not a shipped fact: Anthropic’s models are to be Lucia’s predominant voice, flipped task by task behind the bank and the Fieldwork replays with GPT as the recorded fallback; the Fieldwork conversation stays on GPT (ALGO-39, 2026-09-03; ALGO-41 sequencing law, 2026-09-04). The order of work is fixed: Fieldwork and Admin speak one language, then the Algorithm proves it understands intake, then the guest agent and its model mix.
Implemented — status snapshot 2026-08-12
Active development — status snapshot 2026-08-12
- Lucia Admin and Lucia Booking Engine have active
devrelease lines with separately verified Development identities. Their older Staging deployments are not aliases for the current Development commits. - Lucia Fieldwork App and API are production surfaces under active development; each retains its own repository, runtime, persistence, and authentication boundary.
- Eval Labs is an active internal evidence platform with onboarding, role, review, hydration, and rollout hardening still in progress at the surface level.
In Progress. A feature needs its own current, attributable evidence before it can carry that state.
Deferred
The accepted August 12 evidence explicitly left these capabilities unimplemented:Future / release-gated direction
- Guest production hardening: rate-limit, replay, spam, abuse, privacy, consent, operator review/link, Eval Labs, booking, payment, OTA, and property-specific provider gates.
- Richer Calendar context, calendar import/onboarding, and booking-spine proof that does not depend on a synthetic fixture.
- Task, priority, booking-work-queue, concierge, and vendor-dispatch refinement, each with explicit persistence and completion semantics.
- Payment mutations, refunds, charge or mark-paid controls, billing, subscriptions, users, and permission systems under separately owned authorization and evidence contracts.
- Multi-property inventory, property-aware memory, analytics, and mobile/tablet Workspace experiences.
- Any property-specific launch only after fresh source, provider, environment, data, privacy, and operational proof. A Villa Valentin-labeled fixture is never launch evidence or live client data.

