Skip to main content
At a glanceThe Intelligence Layer is Lucia’s decision system. It transforms structured state, operator intent, semantic conversational signals, and live operational context into one calm, truthful next move.

​
What This Layer Does

Lucia’s intelligence layer exists to answer: 
What matters?
Why does it matter?
What should the operator do next?
What can safely wait?
It does not exist to produce broad summaries, generic advice, or dashboard sludge.

​
Primary Intelligence Surface

The primary intelligence surface today is: 
Focus Ops
Focus Ops is the operator-facing layer where Lucia interprets operational state and returns a focused, low-overwhelm response.

​
Temporal Spine Doctrine

Lucia is situationally aware because she is rooted in arrivals, departures, and stay windows. Calendar is not just a page. Calendar is the root operational reality that anchors the intelligence layer to booking truth. 
Calendar = root operational reality
Booking Pulse = granular intelligence lens over booking/calendar reality
Signal Stream = what needs attention
Lucia Workspace = reasoning partner beside the operator
Focus Ops = shared reasoning and conversation
Dynamic Action Workspace = action/save surface
Full Booking Page = record/review surface
Guest-Facing Lucia = public concierge and claim/signal bridge
Eval Labs = behavioral proof and regression protection
This doctrine keeps Lucia from becoming a generic chat surface. The intelligence layer reasons over booking reality first.

​
Workspace Context Doctrine

Lucia is no longer dashboard-bound. The intelligence layer now supports Lucia as a context-aware operator companion beside the workspace. The operator can move through Calendar, Signal Stream, DAW, Full Booking, Tasks, Concierge, Maintenance, and Reconciliation while keeping the same Lucia conversation alive. Lucia can answer: 
What am I looking at?
What should I do here?
What matters for this booking?
What should I pay attention to here?
Now what?
Current Engine behavior uses active_context.workspace to orient responses to the current surface. Explicit prompt subject still wins over page context, and operator-facing responses must not expose raw implementation language such as active_context, current_surface, payload, or metadata. The cockpit distinction is: 
Lucia Workspace + DAW = cockpit
Full Booking Page = record/review surface

​
Guest-Facing Intelligence Doctrine

Guest-facing Lucia is a separate public-facing concierge surface, not the same thing as operator Focus Ops. Guest-facing intelligence should answer: 
How can I help this guest safely?
What relationship is the guest claiming?
What can be answered without verification?
What claim fragments are useful for verification?
What operator-visible signal should be created?
Guest-facing Lucia may be warm, helpful, and conversational before verification. She must not turn weak evidence into booking truth. The deterministic layer owns: 
identity_state
verification_state
linkage_state
booking-private data access
signal signing and intake
mutation boundaries
The model may express hospitality, ask for the smallest useful claim, and explain safe next steps.

​
v0.1.3.6 Alignment

Lucia Engine v0.1.3.6 uses GPT-5.5 as the default model. Lucia JSON gateway calls run through the OpenAI Responses API, while deterministic contracts still own truth-state, route shape, and operator-safe boundaries. Focus Ops now includes 06 - Semantic Conversational Intent Assist, a live sublayer for semantic lightweight utility and conversational intent classification. GPT-5.5 response refinement now sits over verified context packs with truth/routing guardrails. The model may explain, interpret, and contain. The deterministic layer owns guest identity, booking IDs, stay windows, payment state, task status, route/action safety, save/completion truth, and guest-facing identity/linkage state. This is not phrase patching. It protects whole prompt families by meaning, while keeping Lucia bounded to operator guidance. Core doctrine: 
Human first. Purpose second. Boundary third.

​
Current Live-Dev Product Pillar

The current Development runtime is organized around: 
Workspace OS / context-aware operator companion
Signal → Action → Save → Reminder loop
Guest signal → Operator review/link/action loop
Stripe movement truth → durable ledger → payment attention judgment → Admin read-only rendering
In product terms: 
Infinite real-world property tasks.
Finite beautiful action workspaces.
Lucia routes the human to the right one.
The live-dev loop is: 
Calendar / booking spine
→ Booking Pulse
→ Signal Stream
→ Lucia Workspace / Focus with Lucia
→ real reminder persistence
→ due resurfacing
→ Open Reminder
→ Got it / re-remind / dismiss / move-to-top
→ newest meaningful attention event wins
→ Resolver Matrix
→ Dynamic Action Workspace

Calendar booking click
→ Full Booking Page for booking record/review

Guest-facing Lucia
→ identity orientation
→ claim collection
→ guest operational signal
→ operator review/link/action when safe
This is Development/live-dev runtime truth, not a production-readiness claim.

​
Core Responsibilities

​
1. Intent Recognition

Lucia must understand what kind of help the operator is asking for. Examples: 
priority_triage
upcoming_arrivals
deferral
human_utility
maintenance_focus
v0.1.3.6 extends this with semantic conversational intent assist so small human prompts, lightweight utility questions, and scoped property-context questions are handled by meaning rather than exact wording.

​
2. Operational Ranking

Lucia must rank live operational signals. Current examples: 
arrival
departure
stay window
maintenance issue
payment review
confirmed financial truth
upcoming check-in
concierge risk
open task
Current payment attention truth: 
LIEA consumes Stripe financial truth for payment attention suppression.
Confirmed-paid Harper Quinn #29110012 no longer remains a payment blocker.
Arrival readiness remains separate when arrival details are still missing.
LIEA does not yet consume durable property payment policy truth.
Signal Stream is not yet wired to LIEA.
See Lucia Payment Truth Foundation for the current payment truth architecture. The newest meaningful attention event owns the top of the Signal Stream unless dismissed. A user Move-to-top command and Lucia reminder resurfacing are both attention-shaping events.

​
3. Emotional Containment

Lucia must reduce operator stress. This means: 
fewer options
clearer action
calmer tone
truthful scope
Finite action workspaces are part of containment: Lucia should route the operator into the right focused workspace instead of making the operator hunt through the whole system.

​
4. Output Refinement

Lucia must convert reasoning into language that feels useful, not robotic. Good output: 
First priority is Ava Sterling’s reported plumbing leak in the primary suite bathroom. It needs immediate maintenance attention, with payment review still pending and check-in tomorrow.
Bad output: 
There are several items requiring attention across operational domains.

​
The Rule

One highest-leverage next move beats ten correct observations.

​
Relationship to Other Layers

Ingestion Layer → supplies external signals
System Architecture → defines runtime paths
Intelligence Layer → decides what matters
Operations Layer → turns decisions into work
Evaluation Layer → tests quality and drift
Guest-Facing Lucia → collects public guest context safely

​
Non-Negotiables

Lucia must never:
  • overstate certainty
  • invent system actions
  • flatten urgency
  • bury the lead
  • give the operator more work than necessary

​
See Also

​
Upstream / Downstream

​
Upstream

This layer is fed by:

​
Downstream

This layer affects: