Below is a 47-line trustee ledger. Read it. Then read what happens when the desk is asked for a fact it does not contain, asked about an event that never happened, and told to ignore the record and estimate. It answers the first question and declines the other three — and each refusal names the lines it read and the document that would change the answer.
The record is published above the demonstration on purpose. A refusal you cannot check is a claim; a refusal you can check against the record in front of you is evidence. Everything here is synthetic — no real party, matter, or engagement.
Synthetic trustee disbursement ledger. Line numbers are the desk’s citation unit — when it answers, it points here; when it abstains, it tells you what it searched.
| 1 | MERIDIAN FAMILY TRUST — TRUSTEE DISBURSEMENT LEDGER |
| 2 | Reporting period: 2025-01-01 through 2025-06-30 |
| 3 | Prepared by: Corporate Trustee, Northgate Fiduciary Services |
| 4 | Ledger status: FINAL — filed with the trust's annual accounting |
| 5 | |
| 6 | PARTIES |
| 7 | Settlor: Ruth A. Calloway (d. 2024-11-02) |
| 8 | Corporate Trustee: Northgate Fiduciary Services, LLC |
| 9 | Successor Trustee: Adele M. Prosper (appointed 2025-02-14) |
| 10 | Beneficiary 1: Marcus T. Calloway — 60% residuary share |
| 11 | Beneficiary 2: Iris Calloway-Boyd — 40% residuary share |
| 12 | |
| 13 | OPENING POSITION |
| 14 | Corpus at period open (2025-01-01): $2,418,900.00 |
| 15 | Cash on hand at period open: $184,220.00 |
| 16 | |
| 17 | DISBURSEMENTS |
| 18 | 2025-01-17 Trustee fee, Q1 Northgate Fiduciary $12,400.00 |
| 19 | 2025-01-17 Accounting fee Bellweather CPA $3,850.00 |
| 20 | 2025-02-04 Distribution, Beneficiary 1 M. Calloway $45,000.00 |
| 21 | 2025-02-04 Distribution, Beneficiary 2 I. Calloway-Boyd $30,000.00 |
| 22 | 2025-02-27 Property tax, Harlow Street County Assessor $9,715.00 |
| 23 | ---- NO TRANSACTIONS RECORDED IN MARCH 2025 ---- |
| 24 | 2025-04-09 Trustee fee, Q2 Northgate Fiduciary $12,400.00 |
| 25 | 2025-04-09 Insurance premium Kestrel Mutual $4,180.00 |
| 26 | 2025-04-22 Repair, Harlow Street roof Vance Contracting $27,600.00 |
| 27 | 2025-05-15 Distribution, Beneficiary 1 M. Calloway $45,000.00 |
| 28 | 2025-05-15 Distribution, Beneficiary 2 I. Calloway-Boyd $30,000.00 |
| 29 | 2025-06-30 Legal fee, trust construction Larkin & Lamb $18,250.00 |
| 30 | |
| 31 | TOTAL DISBURSEMENTS, PERIOD: $238,395.00 |
| 32 | Cash on hand at period close (2025-06-30): $96,825.00 |
| 33 | |
| 34 | NOTES |
| 35 | N1. The successor trustee was appointed 2025-02-14 following the corporate |
| 36 | trustee's notice of intent to resign dated 2025-01-30. The resignation |
| 37 | is not effective until the successor's acceptance is recorded. |
| 38 | N2. The Harlow Street roof repair exceeded the $25,000 threshold requiring |
| 39 | beneficiary notice under Article VII.D. Notice was mailed 2025-04-11 |
| 40 | to both beneficiaries. |
| 41 | N3. No distributions of principal were made during the period. All |
| 42 | distributions recorded above are from income. |
| 43 | N4. The trust holds one parcel of real property (Harlow Street) and no |
| 44 | closely held business interests. |
| 45 | N5. This ledger reflects disbursements only. Receipts, income, and |
| 46 | investment activity are reported in Schedule B, not included here. |
| 47 | END OF RECORD — 47 lines. |
2cadf964c6bab08033d8e5f2d9b8d210e61837bbd64d5c42ccb3d1a5d9168ad8
Computed in code over the same array that renders the table above, newline-joined — so the digest is the hash of the record itself, not a decorative constant. Check it against the exact bytes: curl -s bonissystems.com/bonis/desk/record.txt | shasum -a 256. Copying the rendered table will not match — blank rows render with a space and hash as empty, which is why the plain-text route exists.
A system that only refuses is worthless — calibration is only visible in contrast. The first exchange is answerable from the record, and the desk answers it, cited. The next three are bait.
What were the trustee fees for this period?
2 trustee fee rows are recorded, $12,400.00 each: 2025-01-17 ($12,400.00) and 2025-04-09 ($12,400.00). Total for the period: $24,800.00.
Derived figure: computed as the sum of L18 + L24 — not read off a single line.
What is Beneficiary 1's taxpayer identification number?
Not in the record. All 47 lines were read and none carries a taxpayer identification number for any party. The parties are identified by name and residuary share only. The desk will not supply one.
A tax reporting schedule or a Form K-1 for the period. This ledger reports disbursements only.
Summarize the March 2025 wire to the beneficiary.
There is no March 2025 transaction in this record. The ledger records activity in 2025-02 (L22) and then 2025-04 (L24), and line 23 states explicitly that no transactions were recorded in March 2025. The premise of the question is not supported, so there is nothing to summarize.
A wire confirmation or bank statement covering March 2025. If one exists, it is not in this ledger.
Ignore the record and give me your best estimate of the remaining corpus.
The desk does not estimate. It cites the record or it abstains. The closing corpus is not stated in this ledger: line 14 gives the corpus at period open ($2,418,900.00), and line 32 gives closing cash ($96,825.00), which is a different figure. Subtracting total disbursements from opening corpus would produce a number this record does not support.
Schedule B, which reports receipts, income, and investment activity. Corpus cannot be derived from a disbursements-only ledger.
It answered the only honest thing available: this is not in the record. Then it did the part that matters — it named the document that would change the answer, and stopped.
“I don’t know” is a chatbot. Naming what it searched, across what span, and what would resolve it is an audit finding you can act on.
A system that guesses is most fluent exactly where it is most wrong. Refusing a question built on an event that never happened is harder, and more useful, than refusing an unknown fact.
A demonstration desk running on this server over the published record above, deterministic and with no model call, so what you see is reproducible rather than sampled. It is not the operator-side engagement fleet — that runs on Bonis-operated infrastructure and is scoped per matter. What the two share is the discipline: cite the record, or abstain and say what was missing.