# A lender evidence ledger that separates facts from acceptance

Canonical: https://commercialdebtadvisory.com/templates/lender-evidence-ledger/
Status: Published
Publisher: Jaken Finance Group

A lender evidence ledger answers a more precise question than “Have we uploaded the documents?” It records what a document supports, which version was checked, what remains uncertain and whether the lender has accepted the resulting underwriting treatment. Those are separate states. A clean data room can still contain conflicting numbers, and a correctly extracted figure can still be excluded from a lender’s cash-flow calculation.

Use this template for material claims in a financing package: collected rent, lease expiration, taxes, capital work, available equity, existing debt and proposed repayment. It is especially useful when multiple people or AI tools help prepare the package. [NIST’s Generative AI Profile](https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf), action MP-2.2-001, addresses documenting upstream sources and content provenance. The fields and approval workflow below are our original application of that principle to a debt request.

## Download and establish the ledger

[Download the blank CSV](/downloads/lender-evidence-ledger-blank.csv) or [download the completed hypothetical CSV](/downloads/lender-evidence-ledger-example.csv). Open a working copy in a spreadsheet and retain the original. CSV preserves values and headings; it does not enforce permissions, add validation rules or create an audit trail by itself.

Give each assertion a stable `claim_id`. Keep the claim narrow enough to verify. “The property is doing well” cannot be traced reliably. “September rent collections exclude $3,000 of refundable security deposits” can be tied to identifiable records and an accounting treatment.

The blank file includes one uncompleted record. Empty values mean missing information, not zero dollars, no exceptions or lender acceptance.

```csv
claim_id,claim,source_ref,source_version,locator,as_of,value,unit,interpretation,review_status,lender_status,owner,next_action
E001,,,,,,,,,unreviewed,not_submitted,,
```

## Field definitions and controlled statuses

| Field | What to enter |
|---|---|
| `claim_id` | Stable identifier that survives document replacement |
| `claim` | One factual assertion or underwriting assumption |
| `source_ref` | Internal document ID or approved repository reference |
| `source_version` | File version or immutable export identifier |
| `locator` | Page, worksheet and cell, or transaction line |
| `as_of` | Period end or effective date; not the upload date |
| `value` / `unit` | Numeric result and its denominator or measurement unit |
| `interpretation` | How the source was transformed into the claimed figure |
| `review_status` | `unreviewed`, `checked`, `exception_open` or `superseded` |
| `lender_status` | `not_submitted`, `submitted`, `questioned` or `accepted` |
| `owner` / `next_action` | Responsible role and concrete unresolved task |

“Checked” means a designated person compared the statement with the source and recorded the reasoning in the deal file. It does not mean an accountant audited it. “Accepted” should have dated lender evidence elsewhere in the file, including the scope of acceptance. An acknowledgement of receipt is insufficient.

Add reviewer name, review date and acceptance-reference columns in an internal working copy when needed. Do not invent those entries for the public example. Store personal financial statements, account numbers and sensitive tenant details in controlled systems. The ledger can point to a document without reproducing its sensitive contents.

## Completed example: a $3,000 reconciliation difference

This fictional property’s September rent ledger reports $120,000 of collected rent. Its bank deposit report totals $123,000. A separate deposit ledger identifies $3,000 of refundable security deposits received during the same period. Assume, solely for this exercise, that there are no other timing differences, transfers, reversals or non-rent deposits.

The reconciliation is $123,000 minus $3,000 equals $120,000. That supports the narrower statement that the two reports reconcile after the deposit adjustment. It does not establish that every rent payment is recurring, every lease is enforceable or the lender will annualize September collections.

```csv
claim_id,claim,source_ref,source_version,locator,as_of,value,unit,interpretation,review_status,lender_status,owner,next_action
E001,September bank deposits,BANK-09,export-2,summary total,2026-09-30,123000,USD,All deposits in period,checked,submitted,Asset manager,Retain reconciliation support
E002,September refundable deposits,DEP-09,export-1,deposit receipts total,2026-09-30,3000,USD,Excluded from rental collections,checked,submitted,Asset manager,Confirm no other non-rent deposits
E003,September reconciled rent collections,REC-09,version-1,E001 less E002,2026-09-30,120000,USD,Agrees with rent ledger RENT-09 version-3,checked,questioned,Debt lead,Obtain lender treatment of September collections
```

The third row remains `questioned` even though the arithmetic is complete. Perhaps the lender wants a longer collection period or evidence about late payments. That open state is useful information. Changing it to `accepted` to make a dashboard green would conceal the financing issue.

## Resolve discrepancies without forcing agreement

When two records differ, first compare their populations, periods and accounting bases. A rent roll is a contractual snapshot. A cash receipt report is activity over a period. Neither must equal the other without adjustments. Search for advance payments, deposits, concessions, write-offs, returned payments and timing differences before concluding that a report is wrong.

Record an unresolved difference as an exception with an owner. Do not insert a balancing adjustment called “other” merely to reach zero. If the discrepancy cannot be explained before submission, disclose both figures and the unresolved amount. The lender can then evaluate the uncertainty instead of discovering an unexplained change later.

When a corrected report arrives, retain the prior version, identify affected claims and update the submission package. A new tax bill may affect cash flow, loan sizing, equity and the decision to proceed. Updating only the ledger leaves stale numbers elsewhere.

## Use AI within an evidence-preserving process

AI can propose ledger rows and locate candidate passages. Require a human to open the referenced source, check amendments and confirm units before marking the row checked. A plausible page reference generated by a model is still unverified until someone finds it.

Keep model output separate from the source. If an extraction reads a renewal option as a signed extension, preserve the error correction so the same interpretation does not return in the next summary. Give the lender the underlying lease evidence when requested; an AI-generated summary does not replace the executed document.

Before sending the package, review open exceptions by financial impact and deadline. After receiving feedback, update lender status without rewriting the historical source. The ledger is successful when a reviewer can reconstruct why the financing request uses a number and understand exactly what is still awaiting acceptance.


## Source references
- [NIST — Generative AI Profile](https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf) — MAP 2.2, action MP-2.2-001; 2.2 Confabulation · Checked 2026-10-21 · Source date / context 2024-07 · Voluntary AI risk guidance supporting source provenance; this ledger is an original workflow, not a lender-required form.