Golden-deal release comparison

Date: 2026-09-21

Scope

A private, independently calculated underwriting corpus was used as an adopter acceptance suite for the released UWMD implementation and the next implementation on main. The private corpus remains outside this repository. This record contains only de-identified counts, outcomes, and defect classification; it contains no source documents, deal names, tenant or resident data, local paths, or deal values.

The corpus was frozen at commit 719ff38. The compared UWMD targets were:

• published v2.12.0;
• main at 83339b5.

The evaluated implementation was subsequently released as v2.13.0 at 7c250c2; this record is the golden-deal acceptance evidence for that release.

Results

Evidence	Result
Frozen Artifact B assertions accounted for	532
Assertions passed	526
Documented source/baseline defects	6
Refusal assertions passed	20
Runnable control/deal cases passing on the 2.13 implementation	8

The six exceptions are frozen source or baseline defects, not UWMD failures. They remain recorded rather than being changed to make the implementation pass.

Two development cases remain explicitly evidence-blocked:

• GD04 requires a frozen development pro forma.
• GD07 requires a frozen purchase price, leases, and development/operating program.

No missing inputs were synthesized to make either case run.

Released-version defect

Seven model-bearing cases failed on published v2.12.0 at the first UW JSON model-fidelity round-trip. Empty frontmatter arrays were serialized as bare YAML keys, which reparsed as null and changed the semantic digest:

[] -> bare YAML key -> null -> different semantic digest

The source, validation, crosswalk, deterministic calculation, and receipt stages completed before that round-trip assertion. The parser-only negative case, which does not perform a model-fidelity round-trip, passed.

main at 83339b5 serializes an empty frontmatter array as []. The full suite then passes, including UW JSON, UW XML, and UW CSV bundle model-fidelity checks. This is an implementation/representation serialization defect. It does not indicate a missing format field, protocol rule, calculation primitive, or other reason to expand the standard.

Independence and limits

Artifact A and Artifact B were independent of the UWMD implementation under test. UWMD did not regenerate financial expected values, and the comparison did not modify the frozen source models, expected outputs, or authored reference records.

This acceptance result does not close the roadmap's RFC 0045 real-deal property cash-flow assembly item. The available annual models do not establish RFC 0045's required complete monthly or quarterly lease-up periods and supplemental cash-flow coverage. That evidence remains pending; no periodic schedule was manufactured for this review.