zypl.Foundation
Test Now
§ Product · three surfacesOne model · one product's life

Score. Explain. Hand over on your sign-off.

A credit foundation model for the whole life of a product: a scorecard your validator can read before the first outcome, a model that learns every matured loan in context, and Handover once your own outcomes prove it.

Test NowSee the three surfaces
Gini points at 100 matured loans
+2.1Comparator: TabICLv2 with eight members. Ranking quality, not calibration; outcome-stratified contexts. Gini = 2·AUC − 1.
borrower records in pretraining
0Own-brand instances add continued pretraining on the bank's data, inside the bank's environment. Synthetic-only pretraining and bank-local serving are separate properties.
per application on a 16-thread CPU
~50 msAfter a one-time portfolio encoding (about 90 s on CPU, 5 s on GPU, can run offline); 12,000-row context; one server.
model · the whole life of a product
01

Research results from zypl labs, each with its caveat. Every figure with its source →

§ Stage · three surfacesFig. 02

Three surfaces, one model.

Day-ZeroGive Me Some Credit
  • Times 90+ days late0.80

    NumberOfTimes90DaysLate

    Delinquency roll-rate pathway from moderate arrears to severe default

  • Revolving credit utilisation0.65

    RevolvingUtilizationOfUnsecuredLines

  • Times 60–89 days late0.55

    NumberOfTime60-89DaysPastDueNotWorse

  • Times 30–59 days late0.45

    NumberOfTime30-59DaysPastDueNotWorse

6 more weights, each with its reason

ContextIllustrative

Matured loans, as they arrive

  • LN-0412repaidin context
  • LN-0417repaidin context
  • LN-0433defaultedin context
  • LN-0438repaidin context
  • LN-0451repaidin context
  • LN-0469repaidin context

No training run · each loan is read as context

HandoverIllustrative

Shadow period

Your current process
serving
zypl.Foundation
scoring in shadow
Held-out outcomes
sealed until review
Validator
pack reviewed
Risk owner
sign-off pending

Service moves on your sign-off

№ 01Launch day
The Day‑Zero card

A closed-form scorecard from your data dictionary: points, grades and a written reason for every weight.

№ 02Matured loans
In context

Each matured loan joins the model's context. It scores beside your current process, with no training run.

№ 03Your outcomes
Handover

Service moves when your held-out outcomes show the model ranks better, and your risk owner signs.

№ 01Launch day

A scorecard before any repayment.

Send the product's data dictionary and feature statistics. A scorecard comes back with points, grades and reason codes, and a written reason for every weight.

  • A closed-form scorecard: weights, anchors, category tables
  • A written reason and a confidence for every weight
  • Expected default rate and expected ranking quality
7screens · from your book to your validator
Day-ZeroGive Me Some Credit
  • Times 90+ days late0.80

    NumberOfTimes90DaysLate

    Delinquency roll-rate pathway from moderate arrears to severe default

  • Revolving credit utilisation0.65

    RevolvingUtilizationOfUnsecuredLines

  • Times 60–89 days late0.55

    NumberOfTime60-89DaysPastDueNotWorse

  • Times 30–59 days late0.45

    NumberOfTime30-59DaysPastDueNotWorse

6 more weights, each with its reason

№ 02Matured loans

Every matured loan, learned in context.

Each loan that matures becomes part of the model's context, with no training run per product. The same in-context learning reads any labelled table; the figures on this site were measured on lending books.

  • Scores beside your current process
  • Inside your environment, on server CPUs
  • In the demo: your own labelled CSV or the example, ranked on a held-out set
0training runs per product
ContextIllustrative

Matured loans, as they arrive

  • LN-0412repaidin context
  • LN-0417repaidin context
  • LN-0433defaultedin context
  • LN-0438repaidin context
  • LN-0451repaidin context
  • LN-0469repaidin context

No training run · each loan is read as context

№ 03Your outcomes

Service moves on your sign-off.

The model takes over when your own held-out outcomes show it ranks better, and your risk owner signs. Until then it scores in shadow.

  • Your current process keeps serving until then
  • Held-out outcomes, sealed until review
  • Back to your process if the evidence turns
0.651 → 0.699served AUC, Day-Zero alone to Handover

A retrospective replay between Day-Zero and zypl.Foundation — not a live deployment, and not yet with Zehn. Always serving zypl.Foundation reached about 0.711: Handover is a governed, reversible policy, not an accuracy gain over the foundation model. Switches occurred at application index 2,862 to 11,182.

HandoverIllustrative

Shadow period

Your current process
serving
zypl.Foundation
scoring in shadow
Held-out outcomes
sealed until review
Validator
pack reviewed
Risk owner
sign-off pending

Service moves on your sign-off

§ Why declinedFig. 05

A decline, explained.*

* The Day-Zero card's own words on a public lending book: the demo's reference applicants, each against an applicant at the card's medians. zypl.Foundation scores do not ship with reason codes.

Exchange 01Queued

Ans

The card expects 62.0% to default. These three columns raise the risk most.

  • Times 30–59 days lateNumberOfTime30-59DaysPastDueNotWorse3

    Compound risk of near-limit credit lines combined with early arrears indicates imminent default.

  • Times 90+ days lateNumberOfTimes90DaysLate2

    Delinquency roll-rate pathway from moderate arrears to severe default

  • Revolving credit utilisationRevolvingUtilizationOfUnsecuredLines0.985

    Compound risk of near-limit credit lines combined with early arrears indicates imminent default.

Exchange 02Queued

Ans

The card expects 38.0% to default. These three columns raise the risk most.

  • Times 30–59 days lateNumberOfTime30-59DaysPastDueNotWorse1

    Compound risk of near-limit credit lines combined with early arrears indicates imminent default.

  • Revolving credit utilisationRevolvingUtilizationOfUnsecuredLines1.05

    Compound risk of near-limit credit lines combined with early arrears indicates imminent default.

  • Debt-to-income ratioDebtRatio2,850

    Low income combined with heavy debt service burdens sharply reduces shock-absorption capacity.

Exchange 03Queued

Ans

The card expects 0.8% to default. These three columns lower the risk most.

  • Revolving credit utilisationRevolvingUtilizationOfUnsecuredLines0.045

    Compound risk of near-limit credit lines combined with early arrears indicates imminent default.

  • Monthly incomeMonthlyIncome9,500

    Low income combined with heavy debt service burdens sharply reduces shock-absorption capacity.

  • Debt-to-income ratioDebtRatio0.22

    Low income combined with heavy debt service burdens sharply reduces shock-absorption capacity.

§ End of specDemo · public lending books

Score the product you launch next.

Walk the demo on public lending books, or send one product's data dictionary for a Day-Zero Readout within 48 hours.

Test NowRequest a Day‑Zero Readout
Day-Zero Readout
  1. 01Top drivers
  2. 02Expected ranking band
  3. 03Data-readiness gaps
What comes backWithin 48 hours
The card · before any outcomeIn context · every matured loanHandover · on your sign-offYour servers · your outcomes