References and review

Know what is behind your estimate

CostKit creates a first-pass estimate for contractor review. AI selects scope, quantities, and prices using available reference data. It can make mistakes, including when it cites a reference. Check supplier quotes, local labor costs, scope, and quantities before using an estimate for a bid or contract.

What the pricing labels mean

AI reference citation

The AI selected an identifier from the line-item references supplied during generation. CostKit checks that the identifier was present. This does not independently verify that the reference matches the work, or that the quantity or price is correct.

Needs contractor verification

No valid line-item reference citation was recorded. A price may be based on broader context or an AI assumption. Review it against your own quotes and experience.

Your price

You edited the line, or one of your cost adjustments changed it. It no longer carries a reference citation, because the price is now your decision.

Coverage unavailable

Reference lookup failed, the location or trade is unsupported, or the estimate predates coverage tracking.

These labels are for you. They appear on your estimate page, not on the PDF or client link, so their absence from a document your client receives is not evidence of verified pricing. The document tells your client that the estimate is subject to your final confirmation.

Sources and their limits

The US pipeline can supply an NCE 2026 construction cost reference, state labor wage data and material price indices from the Bureau of Labor Statistics, and regional material, permit, climate, and other context. Contractor-reported actuals may contribute aggregated benchmarks where sufficient data is available. Not every source or line item is available for every project.

Coverage is recorded when an estimate is generated. That date is not a source refresh date. Source publication dates vary, and a regional reference is not a live supplier quote or a verified price for a particular city. The line-item citation count is a coverage measure, not an accuracy percentage.

How prices are kept current

NCE 2026 book prices are at February 2026 levels. CostKit adjusts them for cost changes since then: the labor share of each price moves with Bureau of Labor Statistics construction hourly earnings, and the material share with a BLS producer price index. Since February 2026, construction materials are +7.7% and construction wages +3.3%, using data through September 2026. The same adjustment applies to estimates and to the cost guides on this site. Each adjustment is capped, so one unusual month cannot swing every price.

The cost-to-build-a-house guides use US Census data instead: the annual median contract price per square foot for new contractor-built homes in each Census region (excluding land), placed for each state with its building permit values over the latest 12 months and brought current with the national change in permit values. Permit values run low in some states, so each state is kept within 20% below to 30% above its region. Alaska and Hawaii are not represented by the mainland West, so their guides use the adjustment above.

Data sources are downloaded on a schedule, from daily to twice a week, and checked every day. A source is "behind schedule" when its newest data is older than its publisher's normal release lag.

How the internal benchmark works

The saved May 17, 2026 benchmark snapshot evaluated 12 synthetic residential scenarios across trades and US regions, with three generations per scenario. It checked broad total-cost ranges and expected scope items. The expected ranges were derived from the same construction reference used by the estimating pipeline and estimator rules of thumb.

These were not completed, contractor-verified jobs. Passing a range check does not measure error against a final invoice, establish a confidence interval, or guarantee a profitable bid. Historical results are not a measurement of the current model or current data.

We do not yet publish a validated error rate against completed jobs. A stronger benchmark needs independent completed-job costs, matching scope, trade and region breakdowns, and dated model and data versions. Until then, review every estimate before committing to a price.