Sample portfolio read Sample as of 07/31/26 217 accountsSynthetic data

Demonstration build

Synthetic demonstration This is a fictional portfolio for Meridian Auto Group, a company that does not exist. Every account, name, city, VIN, stock number and dollar amount was generated from a fixed seed by generate.py, published alongside this page. No client data, anonymized or otherwise, is present in this build. The data was tuned to exhibit the same patterns a real subprime auto book exhibits, so the analysis on display can be judged on its merits.

The whole book, read in one place.

Every number on this screen is computed from the sample tape cell by cell, exactly as it would be from a client file, with the collateral decoded from the VIN so it forms part of the read. Click any account anywhere to see the row it came from.

$1.34M

Principal on the book

217 accounts, 175 with live balances

$76K

Past due today

gross; customers also hold $1,974 in credits

$62K

Scheduled collections per month

live accounts only; settled residue excluded

$1.83M

Remaining contract value

payment amount times payments remaining

$194K

Principal at 30+ days

26 accounts, 14.9% of the live book

8% vs 38%

30+ rate by car age, loans 12 months and older

age adjusted, so seasoning is not doing the work

44% vs 13%

30+ rate, semi-monthly vs monthly plans

payment schedule is the second marker

Stores reporting

plus two accounts with no store code filed

Three reads for this week

UnderwritingOlder collateral carries more trouble, and it is not just an age of loan effect. Comparing only loans that have been on the book twelve months or more, the 30+ rate runs 8% on cars 8 to 10 years old against 38% on cars 14 and older. The raw gap looks far wider, but most of that is seasoning rather than steel. The controlled comparison is in The vehicles.

StructureSemi-monthly plans are failing at three times the monthly rate. 9 live accounts on semi-monthly plans run 44% past 30 days against 13% for monthly. Small cohort, strong signal. Detail in Where deals break.

Integrity19 contracts have more payments remaining than the contract ever had. $139K of principal sits on paper that has been extended or rewritten without the term fields keeping up. Sort The book by risk score to see them, or open any store profile for its own count.

Where the principal sits

Principal balance by days past due, as filed in the sheet. The 30 to 59 window is the money a phone call can still save. Past 60 days you are choosing between repo and write-off, not collecting.

Highest risk accounts right now

The eight highest scoring accounts. The full scored register is in The book, sorted by risk score, where it can be filtered and exported.

The nine stores at a glance

Bar length is principal. The percentage is how much of that store's live book is 30 or more days past due. Open Stores to click into any one of them.

Every account, searchable.

All 217 rows from the sample tape, with the VIN and vehicle attached. Search by account number, VIN, vehicle or city, filter by aging bucket or store, sort by what matters. Click a row for the full record and its sheet row number.

Account	Vehicle	VIN	Store	DPD	Past due	Principal	Last paid	Score

The collateral.

On a live engagement every VIN on the tape is decoded against the federal vehicle database. That turns a column of serial numbers into an underwriting instrument: what the book is riding on, how old it is, and which metal is carrying the delinquency.

2014

Median model year, live book

the median car here is 12 years old

27 / 24 / 21

Toyota, Kia, Nissan counts

top three makes on the live book

$265K

On cars 7 years or newer

34 accounts, but a median of 5.7 months on book

$779K

On cars 11 years or older

103 accounts, 19.4% past 30 days

Delinquency climbs with vehicle age, step by step

Share of live accounts at 30 or more days past due, by the age of the car under the loan.

The read, with the obvious objection answeredThe raw gradient overstates the case, so here is the controlled version. Read straight off the tape, delinquency rises steadily as the metal ages, which looks decisive. It is not, because the newest cars also carry the newest loans: a median of 5.7 months on book, and this book barely defaults before month six. Restricting to loans at least twelve months old, the gradient holds and is in fact sharper: 8% on 8 to 10 year cars against 38% on 14 year and older. That is delinquency attributable to collateral age rather than to seasoning, and it is a real underwriting lever. Anyone who quotes the raw number without the seasoning control is reading un-seasoned paper.

The makes, ranked by exposure

Live accounts only, makes with 8 or more loans. The share pill is that make's own 30+ rate.

Make	Accounts	Principal	Avg vehicle age	30+ accounts	30+ share

Model hotspots

Models with 5 or more live loans and at least 2 past 30 days.

Model	Accounts	Principal	30+ accounts	30+ share

How this works on a real bookOn a live engagement every VIN is decoded through the federal NHTSA database, which turns the VIN column into an underwriting instrument. On the real portfolio this demo is modelled on, that decode surfaced a cluster of loans sitting in a known engine-recall model generation, with a third of them already past 30 days. That is the kind of concentration a dealer cannot see and a lender will ask about. In this sample the vehicles are generated alongside synthetic VINs, so the VIN strings here decode to nothing by design.

Where deals break.

For every live account we compute how far through its contract it is: payments made against payments written. Then we look at where the accounts now past 30 days actually sit on that line. The shape of that curve tells you whether you have an underwriting problem, a servicing problem, or both.

60%

Median contract progress at break

the typical broken deal is 25 payments in

20 mo

Median months on book at break

these are not first-payment defaults

8 of 26

Break in the first 30% of term

the classic underwriting failures

10 of 26

Break past 70% of term

deals dying within sight of payoff

Contract progress of the 26 broken deals

Each bar is a decile of contract progress. Height is how many accounts now 30 or more days past due sit in that decile. Accounts with rewritten term fields are excluded from this chart.

The readThis book breaks at both ends, and the two ends need different medicine. The early cluster, 8 deals dead inside the first third, is deal selection: old cars, thin down payments, wrong structure. The late cluster is more unusual: 10 deals are failing past the 70% mark, a median of 49 payments in. Customers who have paid that much rarely walk from the car; late breaks usually mean extensions catching up, fee balances surfacing, or payoff friction. They carry only $15K of principal but they poison the charge-off rate and they are the cheapest saves on the book.

Default by payment schedule

Live accounts, grouped by the plan written at the desk.

Plan	Accounts	Principal	30+ share	60+ share	Avg DPD	Past due

The readThe monthly book behaves like a portfolio; the semi-monthly book behaves like a problem. 44% of semi-monthly accounts are past 30 days against 13% of monthly. The cohorts are small, 9 semi-monthly and 11 bi-weekly live accounts, so treat the exact rates gently. But the direction is strong enough to change behavior at the desk: either these plans are being written for the customers least able to hold a schedule, or the schedule itself is generating the breakage. Either answer changes underwriting.

Delinquency by origination year

Live accounts, grouped by the year of the first payment date on the contract.

Origination year	Accounts	Principal	30+ share

The readDelinquency is flat at 25% across the 2023, 2024 and 2025 vintages. The 2026 paper sits at 12% only because it has not seasoned yet. A flat curve like this means the problem is not one bad year of originations. It is the standing selection standard, which is fixable, and the fix shows up in every vintage that follows it.

Vintage curves, static pools and roll.

These are the three measures a credit shop actually runs a book on. Two of them this tape supports today, and the third needs a second tape. What follows is split honestly on that line, because a roll rate invented from one snapshot is worse than no roll rate at all.

0% to 21%

30+ rate, first 6 months vs months 6 to 12

this book's delinquency is born in one window

0.88

Median payment completion, seasoned accounts

payments made against payments due by now

52%

Seasoned accounts 3 or more payments behind

holding $237K of principal

2025Q1

Worst static pool on the book

43% at 30+, 10.3 payments behind

The seasoning curve

Share of accounts at 30 or more days past due, grouped by how long the loan has been on the book. Months on book comes from the first payment date in column S. This is a vintage curve read across the book: it answers when in a loan's life this portfolio breaks.

The readAlmost nothing goes wrong in the first six months, and then the floor gives way. Accounts under six months old run 0% past 30 days. Between six and twelve months that jumps to 21%, with 60+ delinquency appearing for the first time. After twelve months the rate plateaus in the low twenties and stays there. Two things follow from that shape. First, a clean first-payment record tells you almost nothing, so early confidence in a deal is misplaced. Second, the intervention window is months five through nine, before the cliff, not after it. Every dollar of collections effort spent on a brand new account is spent in the wrong place.

Static pools by origination quarter

Each quarter of originations treated as its own pool and measured where it stands today. Pools with fewer than three seasoned accounts are not shown. Payment completion is payments actually made divided by payments that should have been made by now.

Pool	Accounts	Principal	Median age	30+ share	Payment completion	Payments behind

The readOne pool does not belong on this list. The 2025Q1 pool performs worse than every neighbor on both measures: 43% past 30 days, and a payment completion of 0.61 against a median of 0.88 across the other pools. Its accounts are a median of 17 months old and 10.3 payments behind, several times the book. Something happened at the desk in that quarter, and whatever it was did not repeat. That is worth an hour with whoever was writing deals then, because the answer is either a person, a lender program, or a vehicle source, and all three are fixable.

Payment completion, and why it matters more than days past due

Days past due only sees the most recent miss. Payment completion sees the whole life of the loan.

Why this measureAn account can sit at zero days past due and still be six payments short of where its contract says it should be, because a partial catch-up resets the aging clock without repairing the balance. On this book the median seasoned account has made 88% of the payments it owed by now, and 52% of seasoned accounts are three or more payments behind, holding $237K of principal. This is the closest thing to a cumulative loss curve a single tape can produce, and it is the metric that will tie directly to charge-offs once the history arrives.

Roll rate

What it takes, what this tape gives, and what the next one unlocks.

Stated plainlyA roll rate cannot be computed from one tape, and this app will not pretend otherwise. A roll rate measures how many accounts sitting in one bucket this month appear in the next bucket the following month. It needs two observations of the same accounts. What a single snapshot gives you is the standing shape of the book below, which is the opening column of the matrix. Send the August tape and the second column fills in, and from then on every month adds a true measured roll.

The matrix, with this tape in place

Bucket	Accounts, 07/27	Principal, 07/27	Accounts, next tape	Measured roll

Column one is measured. Columns three and four populate on receipt of the next month end tape.

What each additional tape adds

Tapes in hand	What becomes measurable
1 (today)	Seasoning curve, static pools, payment completion, the standing bucket shape, every finding in this app.
2	True roll rates bucket to bucket, first cure rates, and whether the 30 to 59 window is filling or draining.
3 to 6	Roll trend rather than a single reading, seasonally aware forecasting, and a loss forecast built on this book's own migration instead of an industry severity assumption.
12	Full vintage loss curves tracked over time, cohort against cohort, which is the language a lender or a portfolio buyer prices in.

Nine stores, two different companies.

Store 6 runs 0% of its live accounts past 30 days on the newest metal in the group, while the weakest store runs several times that. Same paper, same market, same month. The spread between stores is process, and process can be fixed. Click any store below for its full profile: health score, underwriting mix, aging and every account it holds. Two accounts carry no store code at all, worth one question to the DMS.

Pick a store

Bar length is principal, color is stress. Every rate here is computed on live accounts only, so settled paper does not flatter a store. Click any store to open its full profile.

The return.

This math uses the book's own numbers, not industry averages. The average account in the 30 to 59 window here carries $7,340 of principal. Once an account crosses 60 days it is more likely lost than saved, and the net loss after repo typically runs about half the balance. The read exists to catch accounts while a phone call still works, and to fix the selection standard so fewer need catching.

If the weekly read saves 2 account(s) per month from crossing 60 days

0 per month5 per month

Average balance protected per save$7,340

Loss avoided per save, at 50% severity$3,670

Past due in the 1 to 59 window today$18,034

One-time harvest, collecting 25% of it in 60 days$4,509

LendRisk weekly read$499 / month

$92,589

The work

The book restated the way capital reads it: dead paper out, credits reconciled, aging that ties, and a monthly memo in the language borrowing bases are written in.

On this book

Removing the 4 dead accounts alone restates the book from $1.34M to $1.30M clean, a correction you want to present, not have discovered.

You get

Borrowing-base preparation, sale-ready tapes, and standing monthly reporting that survives a field exam.

How engagements runThe standing read is $499 a month per portfolio: every tape re-scored, the register refreshed, findings in writing. Underwriting standards, dealer scoreboards and lender-ready work are scoped flat, per project. First month on any new book includes everything this demo shows.

To go further, three files unlock the rest

Payment ledger or transaction history. Settles whether the twelve quiet accounts are truly dark or a data artifact. On a comparable book we showed 94% of one month's charge-offs were flaggable three weeks early using the ledger alone.
Charge-off history, 12 to 24 months. Calibrates loss severity to this book instead of an industry default, and prices the write-off queue properly.
Origination terms per account. Amount financed, cash down and rate connect what happens at the desk to what happens in collections, store by store.

Method & glossary.

What every number means, which column it came from, and where judgment was applied. Three levels are used: Computed comes straight from the sheet, Decoded comes from the VIN through the federal vehicle database, Estimate means an assumption was applied and the assumption is stated.

The risk score Computed

What it is

Each live account gets points for what the tape shows. The register is the sum, worst first.

The weights

60+ days past due +6
30 to 59 days +4
15 to 29 days +2
1 to 14 days +1
No payment for 4x the plan gap +5
No payment for 2x the plan gap +3
Pickup note and already 30+ +3
Pickup note and 1 to 29 +2
Principal $8,000 or more +2, $4,000 or more +1
Two or more payments behind +2

The tiers

8 or more is Severe. 5 to 7 is Elevated. 3 to 4 is Watch. Settled residue is excluded.

Store health score Computed

What it is

Each store starts at 100 and loses points for what its own live accounts show. Every deduction is itemized inside the store profile, so the score can be argued with rather than trusted blindly.

The deductions

30+ share x 0.8 capped at 35
60+ share x 1.0 capped at 25
share of vehicles 11 yr and older x 0.12 capped at 12
broken pickup note share x 0.25 capped at 10
quiet account share x 0.4 capped at 10
rewritten contract share x 0.2 capped at 8

The grades

80 and up is Strong, 65 to 79 Steady, 45 to 64 Strained, under 45 Critical. Stores with fewer than 8 live accounts are marked thin file: the rates move too much on one account to lean on.

The known weakness

The score does not adjust for how old a store's loans are, and this book almost never defaults before month six. A store writing mostly fresh paper will score well on delinquency it has not had time to earn. Stores in that position carry a young book warning, and their profile shows median loan age and the 30+ rate among seasoned loans only, which is the fair comparison. Any store writing mostly fresh paper is the case to watch: it can grade Strong while very few of its loans have reached six months.

Live accounts Computed

Why it matters

Every store and vehicle rate in this app is computed on live accounts only, excluding the 42 settled residue rows. A store with a long tail of paid-off paper would otherwise look healthier than the book it actually carries.

Vehicle decode Decoded

Source

On a client engagement each VIN in column M is decoded through the public NHTSA vPIC database: make, model and model year. Only the VINs are sent, never customer information. In this synthetic sample the VIN strings are constructed to fail the federal check digit, so they match no real vehicle, and the make, model and year were generated alongside them.

Vehicle age

2026 minus model year

Payment completion Computed

Formula

payments made / payments due by now

Payments made is column L minus column U. Payments due by now is months on book converted to the account's own payment frequency, plus the first payment. A result of 1.00 means the account is exactly where its contract says it should be; 0.80 means it has made four payments out of every five it owed.

Why it beats days past due

A partial catch-up resets the aging clock without repairing the balance, so an account can read current while sitting six payments short. Payment completion sees the whole life of the loan rather than the most recent miss.

Who is excluded

The 19 rewritten contracts, any account with no first payment date, and loans under six months old where the ratio is too noisy to read.

Seasoning curve Computed

What it is

The 30+ rate grouped by months on book, taken from the first payment date in column S. It shows when in a loan's life this portfolio breaks. It is a vintage curve read across the book at one moment, which is valid for locating the failure window but is not the same as tracking one cohort through time.

Static pool Computed

What it is

Originations grouped by calendar quarter of first payment, each pool measured where it stands today. Pools with fewer than three seasoned accounts are suppressed rather than shown at a misleading precision.

The limit

Each pool is observed once. Comparing pools of different ages mixes cohort quality with seasoning, which is why the payment completion column matters: it is already age adjusted.

Roll rate Not yet computable

Why it is absent

A roll rate measures how many accounts in one bucket this month appear in the next bucket the following month. That requires two observations of the same accounts. One tape gives the standing bucket shape, which is the opening column of the matrix and nothing more. Any roll rate derived from a single snapshot assumes a book with flat originations and flat balances, which this book is not. The matrix is shown with the second column empty on purpose.

Contract progress Computed

Formula

(payments written - payments remaining) / payments written

Columns L and U. How far through its contract an account is. Only computed where the term fields are consistent; the 19 rewritten contracts are excluded.

Rewritten contract Computed

Definition

Payments remaining in column U exceeds the number of payments in column L, which no normally amortizing loan can do. It means the note was extended, rewritten, or the fields were not updated at conversion.

Days past due Computed

Source

Column N, as filed by the DMS. We do not recompute it, we test it against the last pay date. Where the two disagree the account scores as quiet, which lifts it up the register in The book and shows in its store profile.

Buckets

Current, 1 to 29, 30 to 59, 60 to 89, 90 to 179, 180 and up.

Past due amount Computed

Source

Column O. Shown gross at $75,841. 42 accounts carry negative amounts, meaning the customer holds a credit; those net the figure down to $73,867.

Scheduled collections per month Computed

Formula

payment amount x plan frequency

Monthly counts once, semi-monthly twice, bi-weekly 2.17 times. It is what the book should produce in a month if every account pays as agreed.

Quiet account Computed

Definition

Reported under 30 days past due in column N while the last pay date in column Q is more than twice the plan gap old. On a monthly plan that means no recorded payment for 62 days or more while the aging still reads current.

Loss severity, 50% Estimate

Assumption

Net loss on a charged-off account is assumed at half the balance after repo and recovery. This is a stated default, not derived from this book. Charge-off history replaces it with your real number.

The columns we used

Map

B Stock number
J Payment amount
K Pay schedule
L Number of payments
M Vehicle VIN
N Days past due
O Current due
P Pickup note amount
Q Last pay date
R Principal balance
S First payment date
T Location code
U Payments remaining

Caveats, stated plainly

Single snapshot

This is one day's tape. Trends, roll rates and vintage curves sharpen with monthly snapshots or the payment ledger.

Small cohorts

The semi-monthly and bi-weekly plan groups are 9 and 11 live accounts. The direction is strong; hold the exact rates loosely until more months arrive.

Data quirks

On real tapes, clusters of identical last pay dates often point to a system conversion rather than genuine payment behaviour. The review flags them rather than assuming either explanation.

Recalls

Recall exposure is reported at the model generation level and labeled inference. Identifying a generation is not the same as confirming the recall status of a specific VIN.

Names

Every name, stock number, city and VIN in this sample is invented. On a client engagement the app carries first name and last initial only; full identities never leave the client’s own sheet.