Large-scale mixed-use community

Community intelligence

In development. One center is live in the production console and deliberately blocked with no data loaded. The capture and evidence tools are built and exercised against a synthetic corpus only; no host uses them today.

A community hears the same things all week — from visitors on tour, from the hosts who show it, from the questions that come back — and almost none of it survives the week. The change is narrow and strict: every observation is captured in one fixed shape, and every theme is graded by evidence, with nothing recommended below three observations from two independent sources.

Where this sits

An initiative inside the homebuilder relationship: one large-scale mixed-use community, with one operating center inside the operating system already running there. The capture and evidence tools are separate builds.

The homebuilder operating system

Dispersed by default

Four sources of the same picture, none of them joined, and none of them reaching what leadership actually reads.

Leasing partnerHeld in their systemReaches nothing below
CRMRecords onlyReaches nothing below
Community channelsNever gatheredReaches nothing below
Tour conversationsUnrecordedReaches nothing below

Nothing above reaches below

The operating picture leadership readsFour sources · none of them connected to it

So the community gets read through anecdote, and through whoever happened to be in the room that week.

Now · live in the production console

One center, deliberately dark

In the client’s production console the community is an operating center at stage “foundation”. Four capabilities sit dark, and each names the dataset it is waiting on, what that data would make possible and who owns it. The only action is to file a request, tracked against that owner until it is answered.

An operating center in the console with nothing loaded yet: a notice saying no production datasets are ingested so no capability here can be run, and four capability cards — tour feedback, leasing signals, partner performance and community sentiment — each naming the dataset it lacks and what having it would make possible. One card’s request has been filed, and reads back the owner it went to. — Four capabilities, each naming the dataset it lacks and what having it would make possible. The request already filed reads back the owner it went to.Client systemApplication shown with synthetic data

A screen that says what it does not have, and who can supply it, is a better thing to hand a leader than a picture assembled from whatever was available.

Next · built, exercised on a synthetic corpus, in no host’s hands

Capture, and the evidence it has to clear

A host speaks for under a minute after a tour. The transcript is made on the phone itself, so the audio never leaves it, and a defined vocabulary reads it into discrete observations — without a model, and with the same answer every time. Nothing counts toward a theme until the host has corrected and confirmed it.

CaptureHost
A spoken or typed debrief on a phone, right after the conversation ends.Two builds, developer-local store
OrganizeExtraction
A defined vocabulary turns the record into discrete observations.Runs in both builds
ExamineEvidence engine
Thresholds decide what may be called a signal and what may only be watched.Implemented · the most tested part
InterpretLeadership view
Every claim opens onto the observations underneath it.Built against a synthetic corpus
ActSales leader
Above the floor a recommendation is raised with its evidence and given an owner; a person decides.Exercised on synthetic data only

A host’s phone just after a tour debrief: the visit summarised in a line, coverage complete, and the structure the rules read out of it — objections ranked by severity, and amenity reactions marked positive or negative. — The debrief a moment after it was given, on the review step: the visit summarised in a line, coverage complete, and the structure read out of it — objections by severity, amenities by reaction.In developmentApplication shown with synthetic data

The leadership view: one community intelligence score out of a hundred, beside the confidence in the evidence behind it, the coverage, how fresh the corpus is and how large; then four figures, each carrying a confidence badge with its sample size, the number of observations behind it and a drill into them; then what changed this week, each delta with its own evidence count. — A weekly read that names its own confidence, and every line on it opens to the records it was counted from.In developmentApplication shown with synthetic data

The severity-ranked blockers, with the leading one opened: parking, raised five times, and the five debriefs it was counted from — each with the host, the floor plan and an excerpt of what the visitor actually said. — A weekly read that names its own confidence, and every line on it opens to the records it was counted from.In developmentApplication shown with synthetic data

What that makes it possible to say

Nothing is promoted on volume alone. The brief carries a guardrail that names how far the corpus will stretch — an early signal, a directional read, or an operating trend — and every figure on the page carries the count of observations behind it and opens to those records. A theme raised five times by one host is not the same as one raised five times by five, and the read says which it is.

What it saysParking was the leading objection across fifteen debriefs from five hosts — and here are the five it was counted from.

What it will not sayThat the wider community shares it, or that anything measured has changed because of it.

Vision · a written specification, not built

Connected marketing intelligence

The specification describes the same evidence rules applied across marketing, leasing and community activity that today sit in separate systems — first for leasing, then for the whole life of a resident in the community, then across communities. Each phase is gated on the one before it being proven on a real corpus. None of it is deployed, and nothing on this page comes from it.

What is built, and what is blocked

The three parts above are the whole account. Read as a list, they stand like this.

The console center — live in the client’s production console, deliberately blocked. No datasets ingested.
Capture and vocabulary — two builds, both saving to a developer-local store. Not yet in a host’s hands.
Evidence and confidence — implemented, and the most thoroughly tested part. Exercised against a synthetic corpus only.
Search across past observations — table, scheduled job and retrieval logic exist. The index is empty.
Connected marketing intelligence — not built. A written specification, and the later phases are unbuilt.

There is no live corpus, no adoption to report, and no host using it today; nothing has been measured. Recording consent and how visitors’ details are held are open questions, to be settled before any host captures a real conversation.