Status: accepted

Accepted — merged with intent to implement.

RFC 0119: Compare NOI by absolute difference when it is zero or negative (CC-05, CC-06)

Resolution

The UWMD maintainers accepted D1's $500 absolute comparison and D2's model NOI <= 0, including zero applicability on 2026-10-22. These decisions resolve the CC-05/CC-06 comparison gap reported in #491.

Summary

Two cross-checks compare restated NOI with noi_model.net_operating_income by percentage difference:

• CC-05 compares debt_structure.underwritten_noi (the NOI used for DSCR), with a 1% warning threshold.
• CC-06 compares DCF Year 1 NOI, with a 2% warning threshold.

A percentage of a zero or negative NOI is undefined or inverted, so both checks skip when the model NOI is ≤ 0. They also report the skip as field_absent, although the field is present.

The effect: land, lease-up and pre-stabilization records, whose NOI is often negative, have no NOI consistency check at all. In bug bash pass 2, a land record's NOI tampered from −272,000 to −299,200 passed both validate and verify. This RFC compares by absolute difference whenever the model NOI is zero or negative.

Change

1. CC-05 and CC-06, model NOI ≤ 0. When noi_model.net_operating_income is stated and ≤ 0, the check is evaluated, not skipped. It compares the absolute difference between the two stated NOI figures with a fixed tolerance: $500 (accepted, D1). A larger difference is the check's existing warning. The message states both figures and the dollar difference instead of a percentage.
2. Model NOI > 0. No change: the 1% (CC-05) and 2% (CC-06) percentage rules stand.
3. Coverage. field_absent is reported only when a field is absent. The ≤ 0 branch is evaluated.
4. Format §5.3 CC-05 and CC-06 rows state both branches and the tolerance. The Python validator matches.

The accepted tolerance (D1)

Accepted: $500, in the record's currency units, absolute. The precedents:

Check	Compares	Tolerance
CC-09	Stress-test base-case annual debt service vs debt_structure	$500 absolute
CC-10	Sources & Uses purchase price vs valuation	$100 absolute
CC-12	Mixed-use component NOI sum vs property NOI	$1 (rounding)

NOI, like annual debt service, is an annual operating amount that producers round to whole dollars or hundreds. CC-09's $500 is the existing tolerance for an annual operating figure restated in another section, so this RFC reuses it. The other options:

• $1 (CC-12), if restatements are expected to be exact copies.
• $100 (CC-10), the tighter price rule.

A loss of $272,000 restated as $299,200 is caught under any of the three.

Validation impact

Records with NOI ≤ 0 whose restatements differ by more than the tolerance gain a CC-05 or CC-06 warning. The findings retain existing warning severity and CLI exit behavior. No record with positive NOI changes outcome.

Conformance impact

New cases:

• land NOI −272,000, restated −272,000: clean;
• restated −272,400 (within $500): clean;
• restated −272,501: CC-05 warning;
• the same three for CC-06;
• NOI exactly 0 restated 0: clean; restated 600: warning.

Coverage cases show evaluated for NOI ≤ 0.

Reference implementation

Format §5.3; validator.ts (CC-05, CC-06 branches, coverage); the Python validator; the cases.

Alternatives considered

• Keep skipping NOI ≤ 0. This is the defect. Rejected.
• Percentage of the absolute NOI (|a − b| / |a|). It is defined for negative NOI, but still undefined at zero, and its meaning near zero is unstable. Rejected.
• An absolute rule at every NOI. This changes the established positive-NOI percentage rules. Not proposed.

Decisions

#	Question	Proposal	Alternative
D1	Tolerance for NOI ≤ 0	$500 absolute (CC-09's annual-amount precedent)	$100 (CC-10) or $1 (CC-12)
D2	Where the branch starts	Model NOI ≤ 0, zero included	NOI < 0 only, leaving zero skipped
D3	Severity	The existing warning	—
D4	Versions	None allocated; prelaunch 4.x	RFC 0091 D6

Prior art

• Format §5.3 CC-05, CC-06, CC-09, CC-10, CC-12; RFC 0037 (coverage reasons); #491 and bug bash pass 2c.