For credit teams
Your credit box, encoded — and testable before you save it.
Your box is the firm. So it is not three fields in a settings tab. It is the four tests a deal is actually run through, in the order the engine runs them, with both numbers on every rung — the one that flags a deal and the one that declines it.
Your own recent deals sit on the rung that caught each. Change a number and watch them move. Nothing takes effect until you save it.
Your last ten deals, resting on the rung that caught each.
Can the rent cover the payment?
DSCRflags below 1.25x · declines below 1.15x
- 240 Halsey St, Brooklyn, NY — declined
- 1815 Coral Way, Miami, FL — flagged for conditions
Are we lending too much against the value?
LTVflags above 70% · declines above 75%
- 700 NE 2nd Ave, Fort Lauderdale, FL — declined
- 88 Kent Ave, Brooklyn, NY — could not be tested
Is the yield on our money enough?
Debt yieldflags below 9.5% · declines below 8%
- 1120 Gerard Ave, Bronx, NY — flagged for conditions
Does it survive a bad year?
Stressdeclines above 2 scenarios under 1.00x
- 3400 Hillsboro Blvd, Deerfield Beach, FL — declined
Through every test
nothing caught them- 415 Ocean Dr, Miami Beach, FL — cleared every test
- 2201 NW 7th St, Miami, FL — cleared every test
- 1470 Flatbush Ave, Brooklyn, NY — cleared every test
- 6120 Biscayne Blvd, Miami, FL — cleared every test
- Declined
- Conditions
- Untested — a file is missing
- Cleared
A fictional firm, a fictional box, ten fictional deals — the same four tests, in the same order, with the same words the workspace uses. Every deal below is placed by the rule the product uses: the verdict first, then the first test that explains it.
Your box is not only governance. It is the filter.
The alternative every lending shop lives with is an inbox. Decks arrive all week, most of them wrong on geography, asset class or leverage, and the only way to find out is to open them. The screening happens at the top of the funnel, by a person, on exactly the deals least worth their afternoon.
Encoding the box is what changes that. The criteria you configure run against every deal that reaches you — submitted on the platform or emailed in — before you open one. A deal is in front of you because it cleared your criteria, not because a broker guessed you might like it.
The criteria that route deals to you
- Loan size
- $2M – $15M
- States
- NY · FL
- Asset classes
- Mixed-Use · Multifamily · Retail · Industrial · Office
- Leverage ceiling
- 75% LTV
- Coverage minimum
- 1.15x
Two layers, and they are not the same fields. These decide which deals reach you. The underwriting policy above decides what happens to one once it does. We keep them apart because a coverage floor that declines a deal and a coverage minimum that stops it being sent are different decisions with different consequences.
Six deals arrived. Your criteria had already read all six.
- 412 Bergen StBrooklyn, NY · PLATFORM · $8.2M · Mixed-Use
clears your criteria — Clears every criterion.
- 1200 Brickell AveMiami, FL · EMAIL · $22.0M · Office
does not clear — Loan amount ($22.0M) exceeds your maximum ($15.0M).
- 3300 NW 12th StMiami, FL · PLATFORM · $4.9M · Industrial
clears your criteria — Clears every criterion.
- 91 Loop RdAustin, TX · EMAIL · $6.4M · Retail
does not clear — State “TX” is not in your geographic coverage (NY, FL).
- 780 Federal HwyBoca Raton, FL · PLATFORM · $7.1M · Retail
worth a look — Passes every filter, scores 61 of 100 — first-time sponsor.
- 55 Wythe AveBrooklyn, NY · EMAIL · $9.6M · Hotel
does not clear — Hotel is not an asset class you take.
Nothing is hidden from you — it is sorted, and every row carries the reason it landed where it did. A fourth state exists for a box you have not configured yet: it says so rather than guessing.
A broker can send you a deal before you have an account
The package arrives as email. Your reply threads back into the deal, where the broker sees it. Nothing asks you to sign up first, and if you never open a workspace you have still received a complete file and answered it from your own inbox. When the file does arrive it arrives assembled — documents read and their figures pulled with the page they came from, property intelligence already attached — rather than a deck and a hope.
The deal file, the underwriting, and the lender's own credit box — in one place, with the math shown.
AI in the loop, never in the chair.
What it reads
It opens the rent roll, the operating statement, the appraisal, the personal financial statement and the bank statements, and pulls the figures out with the page they came from attached. That is the part of underwriting that is transcription, and it is the part nobody should be doing by hand.
What decides
Your policy and a deterministic engine. Sizing takes the lowest of the constraint ceilings; the verdict walks your targets and your gates in a fixed order. Same inputs, same answer, every time — and reproducible with a calculator if your committee asks.
What you can check
Every number traces back to the page it came from. The memo is a first draft your analyst reviews, adjusts and owns. Where a figure could not be established, the file says untested rather than guessing — a missing appraisal is a missing file, not a judgement against a borrower.
You can write your own checks
Some of what a credit officer looks for is not a threshold. Write it in plain English — flag any lease with a co-terminous clause inside twenty-four months — and it runs on your deals and comes back as an observation with the documents it rests on. Observations are filed beside the deal for a human to weigh. They do not move a verdict, a status, or an action, and one that cites nothing is marked low confidence by construction rather than by our good intentions.
Test the change against your last ten deals. Then decide.
Tightening a coverage floor is a five-second edit with a quarter of consequences. Nobody can hold ten deals in their head at once and say which ones the new number would have caught, which is why the change usually gets made anyway and litigated later, one deal at a time.
So the edit runs against your own recent deals first — the real ones, not a demo book — and shows you what moves before anything is written. It is a read-only pass through the same engine. Nothing is saved, nothing is notified, no deal changes status.
- Coverage target
- 1.25x1.30x
- Coverage decline floor
- 1.15x1.20x
Three of your last ten move.
Two that cleared now arrive flagged for conditions. One you would have taken to committee with conditions is now an automatic decline. That is the deal you want to see before Tuesday, not after.
The same ten deals under the unsaved change. A ringed marker moved.
Can the rent cover the payment?
DSCRflags below 1.30x · declines below 1.20x
- 240 Halsey St, Brooklyn, NY — declined
- 1815 Coral Way, Miami, FL — declined, moved with this change
- 415 Ocean Dr, Miami Beach, FL — flagged for conditions, moved with this change
- 2201 NW 7th St, Miami, FL — flagged for conditions, moved with this change
Are we lending too much against the value?
LTVflags above 70% · declines above 75%
- 700 NE 2nd Ave, Fort Lauderdale, FL — declined
- 88 Kent Ave, Brooklyn, NY — could not be tested
Is the yield on our money enough?
Debt yieldflags below 9.5% · declines below 8%
- 1120 Gerard Ave, Bronx, NY — flagged for conditions
Does it survive a bad year?
Stressdeclines above 2 scenarios under 1.00x
- 3400 Hillsboro Blvd, Deerfield Beach, FL — declined
Through every test
nothing caught them- 1470 Flatbush Ave, Brooklyn, NY — cleared every test
- 6120 Biscayne Blvd, Miami, FL — cleared every test
- Declined
- Conditions
- Untested — a file is missing
- Cleared
| Deal | DSCR | LTV | Debt yield | In force | Dry run |
|---|---|---|---|---|---|
| 240 Halsey StBrooklyn, NY | 1.09x | 61.0% | 8.48% | Declined · DSCR | Declined · DSCR |
| 1815 Coral WayMiami, FL | 1.18x | 66.2% | 9.70% | Conditions · DSCR | Declined · DSCRmoved |
| 700 NE 2nd AveFort Lauderdale, FL | 1.26x | 78.1% | 10.40% | Declined · LTV | Declined · LTV |
| 88 Kent AveBrooklyn, NY | 1.31x | — | 9.80% | Untested · LTV | Untested · LTV |
| 1120 Gerard AveBronx, NY | 1.33x | 68.0% | 8.90% | Conditions · Debt yield | Conditions · Debt yield |
| 3400 Hillsboro BlvdDeerfield Beach, FL | 1.41x | 64.0% | 10.80% | Declined · Stress | Declined · Stress |
| 415 Ocean DrMiami Beach, FL | 1.27x | 63.5% | 10.00% | Cleared · — | Conditions · DSCRmoved |
| 2201 NW 7th StMiami, FL | 1.28x | 58.9% | 11.20% | Cleared · — | Conditions · DSCRmoved |
| 1470 Flatbush AveBrooklyn, NY | 1.52x | 55.4% | 12.00% | Cleared · — | Cleared · — |
| 6120 Biscayne BlvdMiami, FL | 1.44x | 61.8% | 10.60% | Cleared · — | Cleared · — |
88 Kent Ave has no supported value on file, so the collateral test cannot run. It is left untested under both policies. That is a missing document, and the file says so instead of failing a borrower for it.
Every platform in this category says configurable.
We are not going to name anyone. Read the claims below, decide for yourself how much of them you have already been sold, and then read what is actually here.
“Configurable lending criteria”
Three scalar thresholds in a settings tab, and a workflow builder somewhere else. The numbers filter deals after underwriting has already happened.
What is actually here
The four tests are the underwriting. Each rung carries two numbers — the target that flags a deal and the floor that declines it — because a box with one number per test cannot express an exception.
“Set your own risk appetite”
One coverage minimum for every deal the firm will ever look at, and a support ticket when it needs to differ.
What is actually here
A matrix you edit yourself: rows by asset class, transaction type, or both, layered over the firm-wide numbers and resolved per deal.
“Document checklists”
A static list a coordinator maintains by hand, and no opinion at all about how old a document may be.
What is actually here
Requirements tied to the document types the extraction layer reads, with a per-type staleness limit — so an eighteen-month-old appraisal is surfaced as stale, not counted as satisfied.
“Powered by AI”
The model reads the file and scores the deal.
What is actually here
The model reads the file and drafts the memo. Your policy and a deterministic engine produce the verdict. An AI observation with no citation is forced to low confidence by construction, and no observation reaches a verdict at all.
What the box actually holds.
Six blocks, all of them yours to edit, all of them resolved per deal before a single number is computed.
Targets
Coverage, leverage and yield, plus the operating haircuts underneath them — vacancy floor, collection loss, management fee, replacement reserves per unit or per square foot, and a minimum expense ratio that stops an implausibly efficient operating statement from carrying a deal.
Decline gates
Four absolutes, held apart from your targets on purpose. The gap between a target and a gate is your committee's exception appetite, and no preset we ship ever fills it in — auto-declining deals nobody authorized is not a starting point.
The override matrix
Rows keyed by asset class, transaction type, or both, layered over your firm-wide numbers. Office at one coverage target and stabilized multifamily at another is two rows, not two systems and not a second workspace.
Document requirements
Which documents a deal must carry, keyed to the types the extraction layer can actually read, per asset class and transaction type — each with a staleness limit, so a rent roll past its age is flagged rather than quietly accepted as current.
The underwriting floor
A rate floor and an amortization ceiling that the engine sizes against regardless of the rate a broker states. A cheap stated rate cannot flatter a thin deal into coverage it does not have.
Stress tolerance
Your rate shock in basis points, your income decline, your added vacancy — and how many income scenarios may fall under 1.00x coverage before the deal is declined rather than conditioned.
Starting from a program preset
If you would rather not start from a blank matrix, there are presets named after the deal program — stabilized multifamily agency-style, bank balance-sheet, core life-company, conduit, transitional bridge, ground-up construction, DSCR rental. Never after an institution: a credit shop's box is its identity, and you look like a bridge lender is not a product. Each card says what it deliberately leaves blank and why, and you dry-run it against your own book before adopting it — a preset is a starting point you test, not a default you inherit.
The objection, before you raise it
I'm not putting my credit box in someone's SaaS.
Then don't put it anywhere you can't audit.
Every save is a version: who, when, and what changed, in the same vocabulary the fields use. Restore any of them. The point is not the undo button — it is being able to answer, eighteen months from now, exactly which numbers were in force the day a particular deal was declined.
- v914 Aug 2026 · 09:12m.reyes@In force
DSCR floor 1.15 → 1.20 · DSCR target 1.25 → 1.30
- v802 Jul 2026 · 16:40a.okafor@
Override added — Office · LTV target 65%
- v719 Jun 2026 · 11:05m.reyes@
Applied preset — Stabilized commercial, bank balance-sheet
Restoring v8 writes a v10 carrying v8's numbers. The history never shortens.
History is append-only
A rollback re-applies the older snapshot as a new version rather than truncating back to it, so the record of what was in force between two saves survives. A record that can be edited is not evidence, and an examiner knows it.
Platform defaults are pinned per version
Every save stores the platform defaults in force at that moment alongside your numbers. If we change a default next quarter, your verdicts under an unchanged policy do not move.
Only the firm can move the firm's box
Any lender seat can read it — they underwrite against it. Only firm owners and admins can write it, the same gate as every other firm setting. An associate working a deal has no business moving a coverage floor.
Presets fork on apply
Applying a program preset copies its values onto your document. There is no runtime lookup and the engine never reads a preset, so us editing one cannot move a verdict of yours. The presets are named after the deal program — never after an institution.
What we do and don't do with it
Your thresholds, gates, override rows and document rules live on your firm's record. They are not pooled into a benchmark, not shown to another lender, and not published anywhere. What we do measure across firms is marketplace behavior — how often a lender responds to a deal, issues a term sheet, or funds one — and only above a floor of ten deals and five distinct contributing firms, below which the network view says insufficient market data rather than showing you a small-sample number. Your box is not part of that measurement and never has been. Our privacy policy is the binding version of this paragraph; read it before you take our word for it.
One number, and the constraint that produced it.
Sizing takes the lowest of the ceilings your box implies, not the highest. The engine computes each one, names the one that binds, and prints the gap against what the broker asked for. That is the sentence a credit officer needs before anything else: what this deal supports, and which test decided it.
Every figure below reconciles at 6.75% on a thirty-year amortization against an $1,080,000 rebuilt NOI. Check it.
Clears at $11,100,865
— bound by DSCR, at 1.25x coverage. Requested $12,730,000 · $1,629,135 over.
Ceiling under each constraint
- DSCR ceiling$11,100,865binds
- Debt-yield ceiling$11,368,421
- LTV ceiling$14,609,000
NOI $1,080,000 rebuilt on your assumptions · value $20,870,000 · 6.75% · 30-yr am
The file scrub
Eleven deterministic cross-checks over the deal's stated figures and its document extractions — rent-roll arithmetic against stated revenue, NOI against the operating statement, appraised value against the estimate, the tax line against the bill, expense ratio against the asset class, unit mix, liquidity against bank statements, staleness on both sides, and required documents. Each returns pass, warn, fail, or unavailable. A missing input is unavailable, never a fail. The tolerances are yours.
The workbook
A stable-header XLSX of the whole underwriting surface: inputs with provenance, broker figures beside your rebuilt ones, thresholds and gates and exceptions and verdict, and the stress table. It is assembled from the same call the screen renders, so the workbook cannot disagree with it, and the header schema is stable so your macros survive a re-export.
The memo draft
Ten sections, risk-forward, drafted on your assumptions rather than the broker's, in about thirty seconds. Your analyst reviews it, adjusts it and owns the conclusion. It is a first draft, not a credit decision, and it says so on the page.
The questions that decide this.
Does the AI decide anything?
No. AI reads documents, extracts figures with the page they came from attached, and drafts the credit memo. The verdict comes from a deterministic engine running your thresholds — the same inputs produce the same answer every time, and the arithmetic is reproducible on paper. Lender-authored AI checks, written in plain English, return observations with citations that sit next to the deal; an observation with no citation is marked low confidence by construction, and none of them can move a verdict, a status, or an action.
Can you show our credit box to another lender?
No. Your thresholds, gates, override rows and document rules live on your firm's record. They are not pooled into a benchmark, not shown to another lender, and not published. What we do measure across firms is marketplace behavior — how often a lender responds to a deal, issues a term sheet, or funds — and only above a floor of ten deals and five distinct contributing firms, below which we show 'insufficient market data' rather than a small-sample number. Your box is not part of that measurement.
What happens to our policy if we leave?
Every deal's full underwriting surface exports to Excel today, and every policy version is retained with who saved it, when, and what changed. A one-click export of the policy document itself is not built yet. We would rather tell you that on this page than in month four — ask on a call and we will tell you exactly what we can hand you and when.
We already have an origination system. Where does this sit?
In front of it. Relendi is where a deal arrives, gets its documents read, gets scrubbed against itself, gets sized against your box and gets a memo drafted — before it becomes a file in your system of record. It is not a core replacement and we will not pretend it is.
How do we start without committing the firm?
Open a workspace, encode your real thresholds, and run the dry-run against your own recent deals. Nothing is saved until you save it, and nothing about a saved policy changes a live deal you have already decided. If the ladder places your deals somewhere you would not have, that is the most useful conversation we can have.
What you won't find on this page.
No borrowed logos. No lender names we haven't earned the right to print. No originated counter, no accuracy percentage, no case study with the numbers rounded in our favour. Every figure here is either arithmetic you can redo against the sample it sits on, or a description of code we will walk you through line by line on a call. We are an early platform running real deals in New York and Florida — and you are exactly the reader who would catch us.
Encode your box. Then try to break it.
Open a workspace, put your real thresholds in, and dry-run them against your own recent deals before you save anything. If the ladder places a deal somewhere you would not have, that is the conversation we want.
We'll scope it to your institution. Already on Relendi? Sign in