ABF loan tape anomaly hunter

SAMPLE DATA · fictional. Every name and number on this page is synthetic, generated by a script in the site source. No real borrower, lender, fund or deal.

Before an ABF analyst computes delinquency or roll rates, someone has to check that the tape is clean. This demo runs ten deterministic checks over a 2,000-row synthetic consumer-instalment tape. The generator planted 33 breaks of known types. The hunter, a separate script, found 33 of them with 0 false positives. Companion to .

2,000SAMPLE loans on the tape

33rows flagged

33/33planted breaks caught

0false positives

Flags by rule

Rule	Rows
Days past due disagree with delinquency status	4
60+ days late to current in one month (check for re-aging)	4
Current balance above original amount	4
Zero balance but still delinquent	3
Maturity on or before origination	3
Originated after the tape's as-of date	3
Invalid state code	3
APR outside 0–36%	3
FICO outside 300–850	3
Duplicate loan ID	3

Every flag

Loan	Rule	Detail
S-00052	Zero balance but still delinquent	zero balance but 45 dpd
S-00137	Days past due disagree with delinquency status	dpd 97 implies 90+, tape says current
S-00151	Maturity on or before origination	maturity 2025-09-27 on/before origination 2026-04-15
S-00178	Originated after the tape's as-of date	originated 2026-10-12 after as-of 2026-09-30
S-00204	Originated after the tape's as-of date	originated 2026-10-12 after as-of 2026-09-30
S-00228	Zero balance but still delinquent	zero balance but 45 dpd
S-00260	Invalid state code	state code ZZ
S-00452	60+ days late to current in one month (check for re-aging)	90+ → current in one month; check for re-age/modification
S-00475	Current balance above original amount	balance 26,338 > original 22,320
S-00600	60+ days late to current in one month (check for re-aging)	90+ → current in one month; check for re-age/modification
S-00618	Maturity on or before origination	maturity 2025-09-16 on/before origination 2026-04-04
S-00655	Current balance above original amount	balance 24,674 > original 20,910
S-00692	Current balance above original amount	balance 35,459 > original 30,050
S-00707	Zero balance but still delinquent	zero balance but 45 dpd
S-00712	APR outside 0–36%	APR 45.9% outside (0, 36%]
S-00724	Maturity on or before origination	maturity 2025-05-31 on/before origination 2025-12-17
S-00746	Days past due disagree with delinquency status	dpd 97 implies 90+, tape says current
S-00805	FICO outside 300–850	FICO 912 outside 300–850
S-00943	Days past due disagree with delinquency status	dpd 97 implies 90+, tape says current
S-01170	Current balance above original amount	balance 16,945 > original 14,360
S-01291	FICO outside 300–850	FICO 912 outside 300–850
S-01302	APR outside 0–36%	APR 45.9% outside (0, 36%]
S-01397	APR outside 0–36%	APR 45.9% outside (0, 36%]
S-01405	Originated after the tape's as-of date	originated 2026-10-12 after as-of 2026-09-30
S-01509	60+ days late to current in one month (check for re-aging)	90+ → current in one month; check for re-age/modification
S-01513	Invalid state code	state code ZZ
S-01581	Invalid state code	state code ZZ
S-01795	FICO outside 300–850	FICO 912 outside 300–850
S-01916	Days past due disagree with delinquency status	dpd 97 implies 90+, tape says current
S-01994	60+ days late to current in one month (check for re-aging)	90+ → current in one month; check for re-age/modification
S-00160	Duplicate loan ID	loan_id appears 2 times
S-00389	Duplicate loan ID	loan_id appears 2 times
S-00509	Duplicate loan ID	loan_id appears 2 times

Then the roll-rate matrix (clean rows only)

Loan counts, prior-month status (rows) to current status (columns), after removing flagged rows. A synthetic tape has no real performance, so read this as the format, not a finding.

Prior \ Now	current	1-29	30-59	60-89	90+
current	1721	21	0	0	0
1-29	0	76	18	18	0
30-59	0	0	49	13	0
60-89	0	0	0	10	21
90+	0	0	0	0	17

Limits

Real tapes break in ways a ten-rule list will not catch: servicer field changes, modification flags, charge-off timing. These rules are a floor.
A clean result on synthetic data is easy, because the generator and the rules were written together. On a real tape, expect false positives. A person decides what counts as a break.
Never upload a real borrower-level tape to a consumer AI tool.

Scripts: data/abf/gen_tape.py (seed 20261016) and data/abf/hunt.py. Download the SAMPLE tape (CSV) or the results. Not investment advice.