How one question becomes a record the firm owns

Step 1 of 6

The question the firm could not answer

Hollister Reed LLP is a regional accounting and advisory firm of 380 people and forty-one partners, with six offices in North and South Carolina and no chief information officer. Its professional liability insurer sent a renewal questionnaire on 14 August 2026.

Does the firm maintain an inventory of artificial intelligence (AI) tools in use?

In the record: Question 1, asked 14 August 2026, due 30 November 2026, recorded as unanswered when asked.

The insurer also asked whether the firm has a written AI use policy and whether its staff have been trained in AI. On 19 August a manufacturing client wrote to ask whether the firm uses AI on its data. The firm could answer all four questions only with assurances.

The brick thread · 1 of 7

This walkthrough follows one entry in the record, the firm's inventory of AI in use, from that question through its review to the page a board reads and the next question it serves. The inventory is marked with the document swatch wherever it appears below.

Next: intake

Step 2 of 6

Intake: files become proposed bricks

MortarCloud.ai's knowledge-building process starts from what the firm already holds: its documents, its people's testimony and its expense ledger. Each fact becomes a knowledge brick, one plain statement with its source, its evidence category and its date.

The brick thread · 2 of 7

The originating record: one document, dated 14 October 2026

On 14 October 2026 the discovery produced one document, the firm's inventory of AI in use. It lists every use found, each under its own Tail Number with its owner or no owner, and every tool and vendor that supplies AI. The record files this entry in the document class, which means a document shows it, at medium confidence. It holds three pilots: the Copilot pilot, the only sanctioned use; a tax partner's own returning-client briefing, which only the ledger shows; and routine client message answering, which an administrator connected and which only asking found.

• Document: written down by the firm

  The firm's inventory of AI in use, dated 14 October 2026: every use of AI the discovery found, each under its own Tail Number with its owner or no owner, and every tool and vendor that supplies AI. It holds the Copilot pilot, the returning-client briefing and the routine client message answering.

  Source

  Hollister Reed kit, 03-discovery-and-register.md, the register as found, extended by the discovery interviews to the two pilots the kit does not hold

  Confidence

  medium

  Holds from

  14 October 2026

What the inventory holds: the three pilots, each a brick of its own.

• Document

  Copilot pilot, HRL-IT-CPL-0005. The only sanctioned use.

  Confidence

  high

  Holds from

  28 September 2026

• Testimony

  Returning-client briefing, HRL-TAX-BRF-0010. A tax partner's own build; only the ledger shows it.

  Confidence

  medium

  Holds from

  14 October 2026

• Testimony

  Routine client message answering, HRL-FIN-MSG-0011. Connected by an administrator; only asking found it.

  Confidence

  medium

  Holds from

  14 October 2026

Three evidence classes, three example bricks

• Document: written down by the firm

  The Copilot pilot has twenty licensed users, eighteen of them active. It is the firm's only sanctioned use of AI, and it has no written scope or review date.

  Source

  The firm's register as found (Hollister Reed kit, 03-discovery-and-register.md)

  Holds from

  28 September 2026

• Testimony: someone said so

  At least four tax staff upload photographs of clients' prior-year returns to a consumer chatbot on personal accounts, and the images carry taxpayer identification numbers.

  Source

  The firm's register as found (Hollister Reed kit, 03-discovery-and-register.md)

  Holds from

  14 October 2026

• Observed: seen in the ledger

  Orrin Notes, an AI note-taking service, was charged monthly to one employee's company card from 16 November 2024 to 16 September 2026, and no agent in the register uses it.

  Source

  The expense export (intake/expense-export.csv)

  Holds from

  16 November 2024

Today this step is AI-assisted extraction from the files the client provides, followed by review by people; the expense ledger is read by fixed rules rather than by a model. Here the files are fictional.

Next: checking and trust

Step 3 of 6

Checking: what the record can stand behind

Every brick keeps its evidence category, so a reader sees how much rests on what someone said. As found on 14 October 2026, the record held 300 entries: 253 nodes and 47 measured values.

• Observed in logs, ledgers or systems 22%
• Document 53%
• Testimony 25%

The brick thread · 3 of 7

What the brick asserts, and how far it can be trusted

The brick asserts one thing: as of 14 October 2026 the firm holds a dated inventory of AI in use, and that inventory answers question 1. The record states its uncertainty beside it: document class, medium confidence, valid from 14 October 2026 with no end date, and not verified. Verified has one meaning in this record, that a check which could have failed found the entry holding, and no such check has been run on any of the 300 entries. The firm can now say yes to the insurer, with a date on the yes.

A proposed fact is not an accepted fact

A brick a model proposes from the firm's files stays outside the record until a named person reviews it, dates the review and gives it its evidence category; an entry a person wrote in directly carries its source and date instead. In a recorded test run, extraction read a memo of 22 September 2026 from Luis Herrera, the director of information technology, and proposed a tool called Ledgerline Assist. No one has accepted it, so the twin does not draw it and no figure here counts it.

An accepted proposal records who accepted it and when. Acceptance puts a proposed brick in the record, and only a check that could have failed makes any brick verified.

The brick thread · 4 of 7

A person reviews it before it counts

A proposed entry stays outside the record until a named assessor reviews it, dates the review, and gives it one of three classes: testimony, document or observed. A review cannot make an entry verified, because a review is a judgment and verification is a second-source check that could have failed. For this brick the review is shown as an illustration: the inventory was entered into the demonstration record directly, and the review exists today only for entries a model proposed from the firm's files, with no screen of its own. One such proposal, a tool called Ledgerline Assist read from a memo of 22 September 2026, still waits for a decision, so no page draws it.

Next: the graph grows

Step 5 of 6

The twin makes it useful

The Trusted Living Twin is the software that reads the graph and presents it to the people who decide. Its board view fits on one printed page.

The trust posture combines twenty-five assessed questions across these five pillars into one level on a five-level scale. Level 1, Absent, means nothing is written down and nobody could answer the question in the room; level 5, Verified, means an independent check in the last twelve months found the practice holding.

What an answer depends on

The insurer's first question is now answered by the firm's inventory of AI in use, dated 14 October 2026. One entry in it, the routine client message answering, calls an assistant that Practica Practice Management, Inc. supplies inside the firm's practice management system, and that assistant reads client correspondence. Nobody approved it, and because it costs nothing extra, no ledger line shows it. The other three questions still read "not yet answered".

The brick thread · 6 of 7

What the board now sees, and what it still cannot see

On the twin's Before and after page, question 1 now reads with four answers: the inventory and the three pilots it holds, each a link into the register. The other three questions still read "not yet answered". The view cannot show that the inventory is kept current, because no later snapshot exists; it cannot show the inventory verified, because no second-source check has been run; and it shows the trust posture as found, 1.2, Absent, with zero of twenty-five scores verified.

The question row as the twin's Before and after page renders it, sponsor lens, the record as found. This row is the generated page's own markup, not simulated.

1. Does the firm maintain an inventory of AI tools in use?

   The firm's professional liability insurer, in its renewal questionnaire on 14 August 2026; an answer is due by 30 November 2026.

   Could we answer it then?Unanswered when it was asked: the firm could answer only with assurances.

   Can we answer it now?The firm's inventory of AI in use, dated 14 October 2026 Evidence · Open in the view; Routine client message answering Evidence · Open in the view; Copilot pilot Evidence · Open in the view; Returning-client briefing Evidence · Open in the view

   Hollister Reed kit, README.md

• It cannot show the inventory kept current: no later snapshot exists in the record.
• It cannot show the inventory verified: no second-source check has been run on any entry.
• It shows the trust posture as found, 1.2, Absent, with 0 of 25 scores verified.

The next decisions, from the ninety-day plan

1. Elaine Marsh, the managing partner, adopts the permitted-use matrix and says so by 1 November 2026.
2. Daniel Okoro, the quality and risk partner, charters the controller and the forum by 3 November.
3. Daniel Okoro puts the three manifests into force by 15 November.
4. The firm answers the insurer with documents by 20 November, a goal that Daniel Okoro owns.
5. Luis Herrera, the director of information technology, closes the pilot's gaps and adds two manifests by 15 December.

Next: the lasting asset

Step 6 of 6

The lasting asset

For a client, the graph is the client's own asset. MortarCloud.ai's knowledge-building process builds it, the client holds it as a versioned graph file, and the Trusted Living Twin, software MortarCloud.ai licenses, reads it inside the client's own environment. MortarCloud.ai does not host it and keeps no client data after delivery. The Trusted Living Agent Framework is the method that governs all of it.

The same graph serves the next question. Hollister Reed's other three questions already sit in the record, and answering any of them adds bricks to that graph rather than starting a new study.

The brick thread · 7 of 7

The same record answers the next question

On 19 August 2026 a manufacturing client asked in writing whether the firm uses AI on its data. The record already holds what an answer needs: the routine client message answering reads client correspondence, the returning-client briefing reads client tax returns, which the record classes as regulated, and the Copilot pilot reads firm and client documents. Those connections are testimony today, at medium confidence. Answering question 4 means adding an answer entry to this record, not beginning a new study.