Frappe Lending quarterly schedules used three days after the first repayment

Xamit Kadirbekov and Daniyal Kadirbekov · GERO · 7 October 2026

AI-assisted numerical investigation and report. Synthetic software examples; no production accounts, customer losses or deployment frequency were measured.

Finding

Frappe Lending v16.6.0 and source commit 605ae2a866aade812a4a147938b8a4f3b7aaa760 used the repayment frequency value 3 as an interest day count after the first quarterly repayment. The first period used its actual calendar length; later 90–92 day intervals used only three days with a 365-day denominator.

A synthetic local Frappe site used a principal of 1,000,000, an annual rate of 12%, eight repayments, posting/disbursement on 1 January 2026, and a first repayment on 1 April 2026. The relevant rows were:

The remaining quarterly rows likewise stored 3, while their calendar gaps were 90–92 days. The site's eight rows totalled 32,776.59 of interest. Applying the same day-based expression to the actual calendar gaps and the balances already produced by the site gives 126,639.60. This comparison isolates the day-count effect; it is not a reconstructed corrected schedule, because fixing one row also changes later balances and payments.

An independent extracted-method replay reproduces the same unit mismatch. On an isolated balance of 1,000,000, the unchanged get_amounts method returns 986.30 for 3 days; the 91-day /365 calculation is 29,917.81.

Cause

get_non_monthly_days() returned 3 for Quarterly. get_days_and_months() then passed that value to get_amounts() as a number of days while retaining the 365-day denominator. The value represented months in one part of the calculation and days in another.

Reproduction boundary

The portable replay executes the unchanged upstream methods from pinned source copies with minimal framework doubles. It does not start a Frappe site or database. A separate saved synthetic site record used Lending 16.6.0, Frappe 16.36.1 and ERPNext 16.37.0 and exercised the disbursement entry point while preserving the quarterly frequency.

The bounded pre-report review covered the relevant source history and public items #2, #100, #217 and #221. No applicable earlier fix was identified in that scope. This is not a worldwide-priority claim and does not exclude private reports.

Upstream correction status

Frappe contributor Nihantra Patel implemented a calendar-day correction in PR #1483. It carries the previous payment date through schedule construction and uses the actual date difference. The regression starts on a month end, checks consecutive calendar gaps and checks a zero final balance.

• PR #1483 merged into develop on 5 October 2026 as

caaecb4482fdc60a035d83a3ef4b390f225b67d3.

• Automatic backport PR #1484

merged into version-16-hotfix as f340425d5bc73d2a0816c522a32fad0ee4171806.

• The latest tagged release checked on 7 October 2026 was still

v16.6.0, published before those merges. A fixed official release was therefore not confirmed at that earlier check. See the dated correction below.

The pull request describes a user report with the same example numbers but does not publicly identify its source. This report does not claim public attribution for the upstream change.

Evidence

The evidence directory contains the pinned source files and licence, the portable replay and its saved result, the synthetic full-site record, the bounded source review, and saved public metadata/diffs for PRs #1483 and #1484. No email contents or private correspondence are included.

This is one day-count defect in a specific schedule path. It is not a finding about Frappe Lending as a whole. No production loan, accounting posting, customer outcome, financial loss, full upstream test suite, was measured in the original evidence package.

Original report: CC BY 4.0. Original GERO replay: MIT. Vendored Frappe Lending source retains GPL-3.0.

Release verification update — 7 October 2026

This update supersedes the earlier statement that a fixed official release had not been confirmed. Frappe Lending v16.6.1, published on 7 October at 05:30:52 UTC, explicitly includes backport PR #1484. The release tag contains the calendar-day correction. Credit for the upstream implementation remains with Nihantra Patel and the Frappe contributors; public attribution of the report to GERO is not claimed.

A bounded comparison of unchanged methods extracted from v16.6.0 and v16.6.1, using minimal framework doubles, checks the same eight quarterly calendar intervals. The old version has seven day-count mismatches; the new release has zero. The isolated constant-balance example is 986.30 for three days versus 29,917.81 for 91 actual days on a balance of 1,000,000 at 12% annually. Decimal arithmetic independently checks the latter value. This is not a corrected full amortization schedule.

The original immutable archive remains unchanged. A separately named release-verification supplement adds the tag source, replay, results and public release/PR metadata. No private correspondence is included. These checks do not establish deployed customer impact or full upstream-suite correctness.

Short explanation on Technology Product. Narration is synthetic; no customer loss is asserted.