Follow three source excerpts into a short draft, check the numbers and see what a reviewer would send back to the borrower.
1. The source material
These are the complete excerpts used for this example. The labels and page references refer to the fictional pack, so you can check every claim without uploading documents or creating an account.
A · Borrower proposal — 1 October 2026, page 1
Example Development SPV proposes eight homes. Requested gross facility: £1,500,000. Borrower-estimated gross development value (GDV): £2,400,000. Target sales period: 18 months from first drawdown. The proposal says “planning secured”.
Evidence limit: borrower assertions; no valuation or planning decision is supplied.
B · Cost schedule — 2 October 2026, summary tab
Land £600,000; construction £900,000; professional fees £150,000; contingency £150,000. Total £1,800,000. Finance costs and selling costs excluded.
Evidence limit: an initial budget, not a verified appraisal or quantity surveyor’s report.
C · Borrower email — 5 October 2026
The contractor’s latest estimate adds £120,000 to construction costs. Please use this revision. We have not yet updated the cost schedule. We expect the end values to remain as proposed.
Evidence limit: no revised contractor quotation or independent end-value evidence is attached.
2. A draft with traceable findings
REVIEW DRAFT · NOT APPROVED
Proposal and funding
The borrower requests a £1.5m gross facility for eight homes, against its own £2.4m GDV estimate. The proposed 18-month sales period is unverified. Source A
Budget discrepancy
The £1.8m schedule excludes finance and selling costs. A later email adds £120,000 to construction, implying £1.92m on the same basis. Keep both versions visible and request a reconciled appraisal before treating either as the agreed budget. Source B; Source C
Planning remains unverified
The proposal says planning is secured, but this pack contains no decision notice, approved plans or conditions. Record it as a borrower assertion and an outstanding check. Source A
Calculations the reviewer can reproduce
Check
Working
Meaning and limit
Facility / stated GDV
£1.5m ÷ £2.4m = 62.5%
Based on the borrower’s end-value estimate, not a current property valuation. A
Revised budget
£1.8m + £0.12m = £1.92m
Still excludes finance and selling costs; supporting quotation missing. B, C
Illustrative 10% GDV reduction
£1.5m ÷ £2.16m ≈ 69.4%
An editor-selected sensitivity, not a forecast or an underwriting threshold. Facility held constant. A
3. A reviewer correction
Illustrative wording to reject: “The loan is at 62.5% LTV and planning is confirmed.”
Reviewed wording: “The requested gross facility is 62.5% of the borrower’s stated GDV. This is not a verified current-value LTV. Planning is asserted in the proposal; supporting permission and conditions remain outstanding.”
Why it matters: the arithmetic can be correct while the description overstates what the evidence proves. A citation helps a reviewer check a claim; it does not make the claim true.
4. The borrower and review checklist
Request a reconciled appraisal, revised contractor quotation, finance costs, selling costs and a clear equity contribution.
Obtain appropriate independent valuation evidence and test the proposed exit assumptions.
Inspect the planning decision, approved plans, conditions and delivery constraints.
Verify the borrowing entity, ownership, directors and relevant development experience against source records.
Resolve title, existing security and outstanding debt with the appropriate professional checks; a registered charge alone does not establish the balance.
Complete the identity, sanctions, credit and other checks required by the lender’s process.
This list is illustrative, not a complete underwriting policy. The reviewer decides what further evidence and specialist advice the actual deal needs.
How this fits into Eben
Keep the borrower pack with the project, use the agent to investigate questions and draft findings, then review sources, correct the wording and track unresolved questions. The example above shows the kind of review to perform; it does not promise that every AI draft will identify every issue.