Data Source & Sample Composition
All 260 studies in this dataset were generated using the Cost Seg Smart engine (version 2.2.0, calibrated 2026-03-14). The engine applies real RSMeans 2024 construction cost data, MACRS classification rules per Rev. Proc. 87-56, and the methodology described in the IRS Cost Segregation Audit Techniques Guide (Pub 5653).
Property characteristics — square footage, year built, purchase price, location, structural features — are drawn from real US metropolitan property profiles representative of typical investor purchases at each property type. The 260 total comprises the original 107 golden_batch fixtures, 20 condo and multifamily 5+ fixtures added 2026-04-27, plus 133 additional fixtures added 2026-04-28 to bring every property type to n=20.
Property type distribution
| Property type | n | Geographic spread |
|---|---|---|
| Single-Family Rental (SFR) | 20 | TX, CA, NY, FL, TN, CO, AZ, NV, GA, NC |
| Short-Term Rental (STR) | 20 | TN, CA, AZ, FL, CO, NC, UT, GA |
| Duplex | 20 | AZ, TX, CA, FL, OH, MI, NC, MO |
| Triplex | 20 | TX, CA, FL, OH, IL |
| Fourplex | 20 | AZ, TX, CA, FL, NM, OK |
| Condo | 20 | IL, FL, NY, CA, WA, MA, DC, CO, TX, GA |
| Multifamily 5+ Units | 20 | OH, TX, AZ, NC, FL, IN, TN, GA, MO, OR |
| Office | 20 | TX, IL, GA, CA, NY, NJ, FL, AZ, OH, NC |
| Medical Office | 20 | TX, FL, NC, AZ, IL, GA |
| Retail | 20 | NV, CA, IL, TX, AZ, FL |
| Restaurant | 20 | TX, NV, FL, AZ, CA |
| Industrial | 20 | OH, IL, TX, IN, NV, GA, AL |
| Mixed-Use | 20 | NY, CA, IL, TX, FL |
Engine Pipeline (Cost Seg Smart v2.2.0)
The engine produces each study via a deterministic pipeline:
- Land valuation: User override → County assessor (with reliability gate) → Statistical (metro → state → national) prior. Land is excluded from the depreciable basis.
- Component library: Base $/SF costs from RSMeans 2024, segmented by residential (27.5-year long-term) or commercial (39-year long-term) construction.
- Multiplier stack: Geographic cost factor × quality multiplier × property-type adjustment × construction era profile × STR intensity (FF&E uplift if applicable) × PPI time index × indirect cost (25%).
- Caps: 15-year residential cap (18%), office 5-year cap (28%, flag-gated). Caps prevent unrealistic over-acceleration on edge cases.
- Reconciliation: S = adjusted_basis ÷ total_unadjusted_RCN. S < 1.0 means the property's market basis is below modeled replacement cost; S > 1.0 means premium-market property exceeds RCN. The diagnostic ratio rf_raw is a related signal that includes cost variance.
- Premium land floor: For statistical-only land valuation with rf_raw ≥ 2.0 (typical urban condo or premium STR), the engine enforces minimum land allocations: STR 50%, SFR 40%, MF 35%.
- QC gate: 16 automated quality-control checks classify each study PASS / REVIEW / FAIL, with a compound-OK downgrade rule for single soft flags.
Year-1 Federal Tax Savings Computation
Year-1 federal tax savings = (5-year basis + 7-year basis + 15-year basis) × bonus_depreciation × federal_marginal_rate.
Locked assumptions used throughout this report:
- Bonus depreciation: 100% — per the One Big Beautiful Bill Act, signed July 2025, which permanently restored 100% bonus for property placed in service after 2024.
- Federal marginal rate: 37% — top US individual bracket. Most relevant for high-income real estate investors who commission cost seg studies. Lower brackets scale linearly.
- State income tax: excluded — varies 0–13% by state. Including would obscure cross-state comparability. Investors should add applicable state effective rate to estimate total savings.
- $500K normalization — for cross-property comparison, year-1 savings are also computed as if the same property were purchased at $500K (linear scaling by purchase_price ratio). This isolates the property-type effect from the price effect.
Quality Control
Each study runs through 16 automated checks before inclusion. Checks fall into five reason-family categories:
- hard_invariant — basis-conservation violations, NaN/negative outputs, schema mismatches
- market_regime — rf_raw outside expected band for property type, market-to-RCN ratio extremes
- calibration_outlier — pct_accel outside historical IQR by property type
- input_quality — missing or implausible source data (sqft, year_built, purchase_price)
- narrative_safety — output language that could be misleading (savings claims, audit assertions)
Studies flagged FAIL are excluded entirely. Studies flagged REVIEW are included with the qc_status field preserved. PASS studies represent the cleanest data; the dataset's QC pass rate is 97% overall (with 100% pass in 7 of 13 property types) (97% if REVIEW counted as pass).
Reconciliation Factor (S) and rf_raw
S is the engine's reconciliation factor: S = adjusted_basis ÷ total_unadjusted_RCN. It scales the unadjusted component cost build-up to match the depreciable basis derived from purchase price minus land. By construction, S * total_unadjusted = depreciable_basis exactly.
rf_raw is a diagnostic ratio similar to S but inclusive of cost variance. It serves as a sanity-check signal:
- rf_raw 0.4–0.7: typical US property pricing relative to modeled RCN. Engine confidence is high.
- rf_raw 0.7–1.2: moderate market premium. Common for STRs and high-cost metros.
- rf_raw 1.2–2.0: significant market premium. Common for urban condos where amenities and view aren't captured in RCN.
- rf_raw > 2.0: extreme market premium. Engine triggers premium land floor to prevent over-acceleration.
Limitations
- Single engine version. Results reflect Cost Seg Smart's engine. Different engineering methodologies may yield different reclassification splits within ±2–4 percentage points for the same property. Cross-engine comparisons are not in scope.
- Sample size variation. Per-property-type sample sizes are 20 in every category. Per-type IQRs are reported in the main report so readers can see real dispersion alongside medians.
- Geographic coverage. National coverage but uneven; metro-level breakdowns are not published in this release due to small per-metro sample size (median <5 studies per metro).
- Audit risk discussed qualitatively. No individual study in this dataset has been subjected to actual IRS audit. The audit-risk discussion in the main report references published IRS guidance and aggregate audit statistics, not study-level outcomes.
- Year-1 savings assume 100% bonus depreciation. Property placed in service in 2023 (80%) or 2024 (60%) yields proportionally lower year-1 deductions; the unbonused portion still depreciates straight-line over the recovery period.
- Construction cost basis: 2024 RSMeans. Will be updated in the 2027 annual refresh using RSMeans 2026.
License
Released under Creative Commons Attribution 4.0 International (CC-BY 4.0). Use, share, and adapt for any purpose, including commercial, with attribution to "Cost Seg Smart Research, Cost Segregation Benchmarks 2026."
Annual Refresh
The Benchmarks Report will be refreshed annually each Q1, with construction cost basis updated to the latest RSMeans release, sample expanded with completed customer studies, and methodology disclosures updated for any engine version changes.
v1: April 30, 2026. v2 planned: Q1 2027.
Contact
Citation requests, methodology challenges, or data inquiries: research@costsegsmart.com.