Skip to main content

Trust should be inspectable, which means telling you where it stops.

This page covers both what the platform enforces and how it is engineered to enforce it. It also draws the line at the point most vendors leave vague: what an authorized person at your property controls, and what stays with your professionals.

Two questions, one page

What the system enforces, and who is answerable for the rest.

01

The property is the boundary

Authorization carries the property and the role with every operation, rather than being checked once at a login and trusted afterwards. A director at one association cannot reach another’s records by changing an identifier.

02

Control accounts are protected in the database

Receivables, payables, prepaid and deposit accounts cannot be adjusted by a direct write. Postings go through defined paths, and the protection is enforced in the database rather than in a screen that a future feature could bypass.

03

The audit history is not optional

Who did what, when, and to which record — written as part of the operation, not as a log a subsequent change could forget to call.

04

The AI answers from what you approved

The concierge retrieves only from documents an authorized person marked approved, for the audience they set. It cites what it used and abstains when the record does not cover the question.

05

Legal questions are stopped before the model sees them

Ten categories of question never reach the language model at all. What a provision means, which document controls, whether something is legal, what a board should do, anything about one person’s individual rights, drafting an amendment or a demand: a question of that shape is matched and refused by ordinary code, before any model is called. There is exactly one exception, and it is written down: the assistant may read back a compliance date that is already on your calendar and explain the rule behind it, because that is reading your record rather than deciding anything. The refusal is a fixed sentence we wrote, not a sentence the model composes, so there is no wording that talks it into answering and no jailbreak to find. It does not live in the prompt either, which means no settings change and no custom instruction can switch it off. The tuning is deliberately one-sided: a question wrongly refused costs you a support ticket, and a legal question wrongly answered costs a great deal more, so it refuses more than it strictly has to. When it refuses, it offers what it can actually do instead, which is to quote the exact text of records you are authorized to see, with the citation and the date, and assemble that into a packet you can send to your own lawyer.

06

What we do not do

We are a software and services business, not a law firm, an accounting firm or an engineer. We will show you the deadline, the record and the evidence. Whether your board acted correctly is a question for your own professionals, and the platform is built to hand them something readable.

Intended outcome

A board that can explain to owners how their records are protected, and where the platform’s responsibility ends.

Decision boundary

Park & Palm Connect provides software and operational services. Legal conclusions, privilege decisions, required disclosures, and final compliance determinations remain with the client and its qualified counsel.

Ask us the hard version

Bring your most skeptical director. The questions that are awkward to answer are the ones worth answering before you buy.

Request a trust and security review