How each pass performs on the criteria that matter for seeding a cited valuation model.
| Criterion | In-house (adversarial harness) | Gemini | GPT |
|---|---|---|---|
| Sourcing rigor (will not fabricate) | Akills unsupported claims | C+states killed numbers as fact | Acites or abstains |
| Verified vs plausible separation | Aexplicit kill list | Cno separation | Aexplicit "not located" |
| Asking vs closed distinction | Bnoted | B-blended | A+HVS ALIS buyers 8.5-9.5% vs closed 8.2-8.6% |
| Directional accuracy vs primary CBRE | A37-50 bps | B+50-75 bps, proxy-inflated | A40-50 bps |
| Coverage / completeness | B-narrow, 2 gaps open | Amost complete | Bhonest gaps |
| Underwriting usefulness | A-calibrated | A-sophisticated but unreliable points | Abase/downside/upside band + yellow-light guardrail |
| Overall | A- | B+ | A- |
GPT and in-house lead for seeding a cited model. Gemini is the best narrative briefing, not the best source of seed numbers.
All three passes agree. High confidence, cleared to seed pending Warren ratification.
| Finding | Value | Sources agreeing | Confidence |
|---|---|---|---|
| Houston trades wider than Dallas-Fort Worth | ~40 to 50 bpshotel-direct CBRE, not proxy | In-house 37-50 bps (CBRE H2 2025 hotel columns); GPT 40-50 bps (CBRE H2 2025); Gemini 50-75 bps (proxy) | High |
| Texas market ladder (branded limited-service "Other" midpoints) | Austin 7.50% DFW 7.875% San Antonio 8.25% Houston 8.25%tightest to widest |
All three cite the same CBRE US Cap Rate Survey H2 2025 table | High |
| National stabilized anchor | 8.0% to 8.5%HVS, Feb-May 2026 | All three (HVS US Market Pulse) | High |
| National closed transactions | 8.2% FY2025 8.3% Q4 2025 8.6% Q1 2026MSCI RCA via HVS |
All three (MSCI Real Capital Analytics) | High |
| Segment direction | Luxury, favored extended-stay, high-performing economy trade BELOW the norm (tightest); older limited / select / full-service facing major renovation trade ABOVE it (widest)HVS | All three (HVS) | High |
Where the passes conflict. The verdict column is the call for the seed layer.
| Issue | In-house | Gemini | GPT | Verdict |
|---|---|---|---|---|
| Extended-stay premium vs economy split | Unsourceable | 150-450 bpsCoStar economy ES 8.6-13.1% | Unsourceable | Direction only (premium tighter). Do not seed a hard economy-ES cap. Get Greysteel or closed comps. |
| Houston vs DFW magnitude | 37-50 bps | 50-75 bpsretail/MF proxy | 40-50 bps | Use 40-50 bps. Hotel-direct CBRE over proxy. |
| Economy / midscale point estimate | Killed 10.5% | Asserts 10.5% | 8.3-9.5% (PwC) to 10.25% (Tilt) | Widest segment confirmed, exact point uncertain. Seed a range, not a point. |
| Bryan-College Station cap | No public cap, do not estimate | 8.5-9.5%LoopNet MF proxy + zero-supply A&M demand | No supportable public cap | Assumption only. Soft proxy +50 to 100 bps off San Antonio, label clearly. |
Failed the in-house adversarial round. Do not seed. Listed so any future source repeating them is flagged.
Two layers. Product-type bands stay as-is (weakly sourced); the market layer is the new high-confidence unlock.
| Band / Market | Seed | Confidence | Note |
|---|---|---|---|
| Layer A · Product type / chain scale (keep current, medium-low confidence) | |||
| Extended-stay | 7.5% | Med-Low | Eventually split (premium tighter), only when broker data lands |
| Full-service | 9.5% | Med-Low | Ordering supported, precise point not |
| Select-service, upscale | 8.5% | Med-Low | No clean public chain-scale split exists |
| Select-service, upper-midscale / midscale / economy | 9.25% | Med-Low | Seed as a range, not a point (see Section 3) |
| Select-service, upper-upscale / luxury | 9.5% | Med-Low | Ordering supported only |
| Layer B · Market adjustment, bps vs DFW baseline (high confidence, CBRE H2 2025) | |||
| Austin | -62 bps | High | Tightest Texas metro |
| Dallas-Fort Worth | 0 (baseline) | High | Reference market |
| San Antonio | +37 bps | High | In line to slightly wider |
| Houston | +50 bps | Judgment | Set above the +37 survey floor (CBRE city-center +50, proxies 50-75, RevPAR contracting). Sourced range 37-75. |
| Bryan-College Station | +50 to 100 bps | Assumption | Off San Antonio, no public cap; tertiary illiquidity |
The market layer is the high-confidence unlock: add it onto the product-type band. The product-type bands themselves remain Warren's judgment overlay until closed trades validate them.