AACE estimate classes for road, rail and transit work: Class 5 to Class 1, accuracy ranges and estimate growth
AACE International classifies a cost estimate by the maturity of the project definition behind it. The scale runs from Class 5 (0% to 2% of full definition, concept screening) to Class 1 (65% to 100%, check estimate or bid). For roads, railroads and transit, its Recommended Practice 98R-18 gives each class an expected accuracy range. For Class 5, the range runs from -20% to -50% on the low side and +30% to +100% on the high side. For Class 1, it narrows to -3% to -10% and +3% to +15%. These ranges hold at about 80% confidence, once a contingency set at a 50% probability has been added, for a given scope. The range describes the estimate: it is not a contingency to add. 98R-18 ties a 30% design review to Class 3, while AACE's 2026 tunneling practice reports that it is most often read as Class 4. Either way, the deliverables decide the class. Mirage Metrics wrote this guide for the estimators who prepare these estimates on transit, bridge and highway projects.
- Published
- Updated
- Reading time
- 14 min
- Written by
- Jules Toussaint
AACE's recommended practices are copyright. The ranges and matrices here are restated from 98R-18 (rev. August 2020), from the sample pages AACE publishes of 18R-97, 56R-08 (rev. August 2020) and 142R-26 (August 2026), and from AACE's PGD 01 (rev. August 2022). The class descriptions are paraphrased.
Who wrote this, and why. Mirage Metrics builds AI agents that read construction documents. From Class 3 down, AACE's methods move from cost per length to unit costs on measured quantities. That takeoff is where an agent helps. It structures the quantities from the plan set and keeps each one tied to its sheet, while the estimator prices, sets the contingency and decides the class. Mirage reads plan sets drawn in metric or imperial units, and documents in English or French. Mirage publishes no estimate accuracy figures. Its published takeoff proofs are John Cockerill, 80 to 90% less time on piping takeoff and a bill of materials ten times faster, and OCBM, 3 agents in production and 100% of quantities tied to a source drawing. See the takeoff and estimating workflow.
The five classes on a road, rail or transit project
What sets an estimate's class is how far the project's definition has gone. The maturity of the project definition deliverables is the only primary characteristic of a class. End use, method and accuracy are secondary, and they follow from it. The percentages are only indicative, because "it is the maturity of the defining deliverables that is the determinant, not the percent" (98R-18, section 3). The table below reads each class as 98R-18 describes it for road and rail, alongside what a transit or bridge estimating team actually has in hand.
| Class | Definition and typical review | What the project has | Method | Accuracy range (98R-18) | Quantities that exist | Takeoff automation adds, and does not |
|---|---|---|---|---|---|---|
| Class 5 | 0% to 2%; planning, concept screening | Capacity or load factor, general concept, routing alternatives on large-scale maps | Cost per length, parametric models, judgment or analogy | L: -20% to -50%; H: +30% to +100% | Lengths of each alternative route | Little to measure; the unit rates come from past projects and are the estimator's call |
| Class 4 | 1% to 15%; feasibility, alternatives analysis | Preliminary road or rail design and geometry, routing corridors, preliminary structure and system features, topographic mapping | Adjusted cost per length, approximate unit or assembly costs for major structures and systems | L: -15% to -30%; H: +20% to +50% | Alignment lengths, structure counts and sizes per alternative | The same measurements on every alternative, on the same rules; not the choice between them |
| Class 3 | 10% to 40%; budget authorization; many owners call for a 30% design review | Completed road, rail and structure concept design, route confirmed by survey, right-of-way holders identified, EIS prepared | Semi-detailed unit costs, assembly-level line items, detailed quantity surveying | L: -10% to -20%; H: +10% to +30% | Measured quantities by assembly from the concept plans | Measured quantities, each tied to its sheet; not the unit rates, the contingency or the class |
| Class 2 | 30% to 75%; control or bid; many call for a 60% design review | Detailed design well advanced, most right-of-way obtained, permits obtained | Detailed unit costs with forced detailed takeoff, often tens of thousands of line items | L: -5% to -15%; H: +5% to +20% | Line-item quantities, with forced detail where the design is missing | Line-item takeoff at that scale and what changed between packages; not the forced-detail assumptions |
| Class 1 | 65% to 100%; check estimate or bid; some call for a 90% review | All maturity-matrix deliverables complete or nearly so | Detailed unit costs on actual design quantities | L: -3% to -10%; H: +3% to +15% | Full design quantities, package by package | A full takeoff checked against the bid schedule; not the price |
Source: 98R-18: Table 1 (maturity, end use, method, range), Tables 2a to 2e (deliverables, reviews, methods), Table 3 (maturity matrix). The last two columns are Mirage Metrics' reading for takeoff.
98R-18's maturity matrix shows where bridge and transit deliverables sit at each class (98R-18, Table 3). Major structure drawings for bridges and viaducts are started or preliminary at Class 4, and complete, at concept level, at Class 3. Geometric layout (alignment, profile, cross section) is preliminary or complete at Class 3. Roadway or track discipline drawings, and signaling, telecommunication and electrification drawings, are only preliminary at Class 3 and complete at Class 1. Geotechnical and hydrology studies are preliminary at Class 4 and defined from Class 3.
The range is a probability, not a contingency rule
A range tells you how far actual cost may fall from the estimate, which is not the same thing as a contingency. AACE states each range as the typical variation of actual cost from the estimate at an 80% confidence interval, after a contingency set "typically to achieve a 50% probability of project cost overrun versus underrun", for a given scope. A specific project can fall outside it. The range for a given estimate "should always be determined through risk analysis of the specific project and should never be pre-determined" (98R-18, section 3). AACE's guidance document adds that range values cannot be quoted without quantitative risk analysis (PGD 01). So adding 30% because an estimate is Class 3 confuses the range with a contingency.
Where takeoff automation changes the work
Up to Class 4, the estimate rests on cost per length and factored costs, so the work is finding comparable costs, not measuring. From Class 3, 98R-18's methods call for detailed quantity surveying and unit cost line items. At Class 2 there are often tens of thousands of them, with forced detail where the design is missing. That measuring is what an agent can take over. The rates, the contingency and the judgment of what the deliverables support stay with the estimator (see the quantity takeoff guide).
30%, 60%, 90% and IFC: which class?
Here is how the familiar design review milestones line up with the classes. 98R-18 describes Class 3 as the stage where "many call for a 30% design review". It ties Class 2 to a 60% review, and Class 1, for some, to a 90% review. It also warns that a percent design review may not match the deliverables. An estimate at a 30% review that lacks some Class 3 deliverables is "Class 3 with Exceptions", with the missing deliverables listed (98R-18, sections 4 and 5).
AACE's tunneling practice, published in August 2026, reads the same milestones differently. Where AACE classes are applied, it says, 30/60/90 reviews are "most commonly associated with Class 4/3/2", and one reference equates them with Class 3/2/1. Because there is no universal standard for those reviews, they are "an unreliable basis for stage or class delineation" (142R-26, section 3). For a tunnel, geotechnical studies cap the class, so the estimate cannot be more advanced than they support (142R-26, section 2). An issued-for-construction (IFC) set and the bid, with every package designed, are Class 1 territory. By Class 1, 98R-18 notes, all the work is usually tendered.
98R-18 leaves out the parts that drive a transit or bridge program's risk: major long-span bridges and viaducts, major tunnels, and major buildings such as rail stations. Each is a sub-project with its own class. A program is usually rated by its least defined major element, "on the principle that a system is only as strong as its weakest link" (98R-18, section 2). Stations go to 56R-08, and tunnels now to 142R-26. AACE's guide to its classification practices (rev. 2022) lists no practice for long-span bridges (PGD 01). To see at which class each delivery method fixes the price, read the project delivery methods guide.
98R-18 against 18R-97 and 56R-08
Each industry has its own AACE practice, and the ranges do not always match. 18R-97, AACE's first industry practice (process plants), gives the same five ranges as 98R-18. 56R-08, for buildings, is narrower at every class. The generic practice, 17R-97, sets out the principles and a generic matrix, and each industry practice adds its own maturity matrix and ranges (PGD 01). So a search for "AACE 18R-97" on a road or rail project lands on the right numbers but the wrong deliverables. 98R-18's maturity matrix is the one written for routes, structures and systems.
| 98R-18 road and rail | 18R-97 process | 56R-08 building and general construction | |
|---|---|---|---|
| Class 5 | L -20% to -50%, H +30% to +100% | L -20% to -50%, H +30% to +100% | L -20% to -30%, H +30% to +50% |
| Class 4 | L -15% to -30%, H +20% to +50% | L -15% to -30%, H +20% to +50% | L -10% to -20%, H +20% to +30% |
| Class 3 | L -10% to -20%, H +10% to +30% | L -10% to -20%, H +10% to +30% | L -5% to -15%, H +10% to +20% |
| Class 2 | L -5% to -15%, H +5% to +20% | L -5% to -15%, H +5% to +20% | L -5% to -10%, H +5% to +15% |
| Class 1 | L -3% to -10%, H +3% to +15% | L -3% to -10%, H +3% to +15% | L -3% to -5%, H +3% to +10% |
| Class 5 method | Cost per length, parametric, judgment or analogy | Capacity factored, parametric, judgment or analogy | Area factoring, parametric, judgment or analogy |
| Covers | Roads, highways, railroads, transit rail; not major long-span bridges, major tunnels or stations | Process plants; 98R-18 sends substations here | Buildings, stations included, and repetitive site and civil work |
Source: 98R-18, Table 1 and section 2; 18R-97, Table 1; 56R-08, Table 1; scope of 56R-08 from PGD 01.
Canada: Class D, C, B and A
Canada's federal government uses its own estimate terms. The Treasury Board distinguishes an indicative estimate, "an order-of-magnitude estimate that is not sufficiently reliable to warrant Treasury Board approval as a cost objective", from a substantive one. A substantive estimate is "an estimate of high quality and reliability", based on detailed specifications (TBS glossary). For building projects, Public Services and Procurement Canada (PSPC) defines four classes by the documents available and the design allowance each may carry (PSPC definitions, as reproduced by the RAIC).
| PSPC class | Based on | Format | Design allowance, at most | Nearest 56R-08 class by end use |
|---|---|---|---|---|
| Class D (indicative) | Statement of requirements, outline of potential solutions or functional program | Elemental analysis | 20% | Class 5 (functional area, concept screening) |
| Class C | Schematic or conceptual design, or a comprehensive list of requirements | Measured quantities, elemental analysis | 15% | Class 4 (schematic design, concept study) |
| Class B (substantive) | Design development drawings and outline specifications, preliminary design of all major systems | Measured quantities, elemental analysis | 10% | Class 3 (design development, budget authorization) |
| Class A (pre-tender) | Completed construction drawings and specifications before tender | Fully measured quantities, elemental and trade formats | 5% | Class 1 (check estimate or pre-bid) |
The last column is a reading of 56R-08's end-use terms, not an official mapping. The two systems also measure different things. A design allowance is money carried in the estimate for design still to come. An AACE range, by contrast, is a statement about how far actual cost may fall from the estimate.
US references: GAO and FHWA, and where FTA sits
In the US, GAO's Cost Estimating and Assessment Guide (GAO-20-195G, March 2020) is the federal best-practice reference for the process itself, from the technical baseline and work breakdown structure to sensitivity and risk analysis, documentation and updates. It is about how an estimate is built, whatever its class.
For a federal-aid highway project of $500 million or more, 23 U.S.C. 106(h) requires a project management plan and an annual financial plan. FHWA expects at least one Cost Estimate Review, "an unbiased risk based probabilistic review", within 12 months before the Initial Financial Plan, whose cost should reflect the 70th percentile, in year-of-expenditure dollars (FHWA financial plan guidance). A P70 figure (70th percentile) is a different statement from an AACE range around a P50 (50th percentile) estimate. Check which one a number is before you compare it.
For transit, 98R-18 makes two observations. The Federal Transit Administration (FTA) does not make its national decision at the Class 5 regional-planning gate. And the distinction between New Starts and Small Starts can imply different accuracy expectations (98R-18, sections 3 and 4). For the cost format FTA asks for, see the FTA standard cost categories guide.
Why estimates grow between classes, and what overrun studies measure
An estimate that rises from Class 4 to the bid has not necessarily missed, because growth has more than one source. AACE's ranges hold for a given scope and "explicitly exclude scope change" (98R-18, section 9; PGD 01). Scope added later is therefore growth that no range covered. Prices move too. FHWA asks for major-project costs in year-of-expenditure dollars, so escalation sits inside the figure, and 98R-18 lists escalation among the deliverables to define by Class 3 (Table 3). What is left is definition, which the class range describes.
The baseline you measure from decides the headline. In all the national systems 98R-18 compares, the first estimate reported to the public is the Class 4 one. Given its range, that confuses overrun discussions, and measuring overruns from the Class 3 estimate "would be a better basis" (98R-18, section 4). The best-known study took 258 transport projects worth $90 billion. It measured in fixed prices, from the estimate at the decision to build. Costs were underestimated in almost 9 projects in 10, and actual costs averaged 28% above estimate: 44.7% for rail, 33.8% for fixed links such as bridges and tunnels and 20.4% for roads (Flyvbjerg, Holm and Buhl, 2002). The class of that decision estimate varies between systems. Norway funds at Class 3, while other systems wait for tender prices at Class 2 (98R-18, section 4).
Why costs are underestimated is disputed. 98R-18 records the view that politicized decisions breed "strategic misrepresentation", and also research that has not supported it. Its own ranges assume a base estimate checked against empirical cost metrics, and a contingency from empirically valid risk quantification. For weak project systems and complex projects, the research it cites puts the high range at two to three times Table 1 (98R-18, section 3).
Illustrative example, invented figures: one project, two baselines
A light rail extension is announced on a Class 4 estimate of $250 million, contingency included. At its wide bounds, 98R-18's Class 4 range puts the 80% interval between $175 million (-30%) and $375 million (+50%). At its narrow bounds, the interval runs between $212.5 million (-15%) and $300 million (+20%). After the 30% design review, a Class 3 estimate gives $320 million. The project is bid and built for $340 million, with no change of scope.
Measured against the announcement, the final cost is 36% over ($340 million divided by $250 million). That sits inside the wide Class 4 band and outside the narrow one. Measured against the Class 3 estimate, it is 6.25% over ($340 million divided by $320 million), under even the narrow Class 3 high bound of +10% ($352 million). The project and the cost are the same, yet the result reads as a 36% overrun or as a result inside its budget range, depending on the baseline.
What the estimating team controls is the record, and it has two parts. The first is a basis of estimate that states the scope, the methods, the deliverables used, the assumptions and exclusions (98R-18, section 7). The second is a set of quantities that trace to the drawing revision they were measured from. With both, growth between classes can be split into design, scope and price (see plan revision comparison).
Limits of estimate classes
Five limits are worth keeping in mind before you rely on a class.
- Percent design is not the class. The deliverables set the class (98R-18, section 4), and the 30/60/90 milestones have no universal standard (142R-26).
- A class is not a quality specification. In AACE's words, "Class alone is not a valid contract specification for estimating services". A contract has to define the estimating requirements, processes and methods (PGD 01).
- The range is not a contingency. It comes from the project's risk analysis and is never pre-determined (98R-18, section 3).
- 98R-18 leaves out the big pieces. Long-span bridges, major tunnels and stations carry their own class, and the program usually takes the weakest one (98R-18, section 2).
- Long-range plans are not Class 5. An estimate made decades ahead, whose scope will not be updated, is Unclassified or "Class 10", with no accuracy range (98R-18, section 9).
Questions readers ask
The AACE estimate classes are five classes, from Class 5 to Class 1, set by the maturity of the project definition deliverables behind an estimate. Class 5 rests on 0% to 2% of full definition, and Class 1 on 65% to 100%. End use, method and accuracy follow from that maturity. For road, rail and transit work, the industry practice is AACE's 98R-18. 18R-97 covers the process industries, and 56R-08 covers buildings and general construction.
A Class 3 estimate is the estimate that supports full budget authorization or funding and becomes the first control estimate. Under 98R-18, it rests on 10% to 40% of full project definition, a completed concept design for road, rail and structures, and semi-detailed unit costs at assembly level. Many owners call for a 30% design review at this class. Its expected range is -10% to -20% on the low side and +10% to +30% on the high side.
Under 98R-18, a Class 5 estimate's range is -20% to -50% on the low side and +30% to +100% on the high side, and a Class 4 estimate's is -15% to -30% and +20% to +50%. AACE states these at about 80% confidence, after a contingency set at a 50% probability of overrun, for a given scope. The range for a specific project comes from its risk analysis, not from the table.
A 30% design is not automatically a Class 3 estimate. 98R-18 notes that many owners call for a 30% design review at Class 3, but the class is set by the deliverables. An estimate at a 30% review that lacks some of them is a Class 3 with exceptions. AACE's 2026 tunneling practice reports that 30/60/90 reviews are most commonly read as Class 4/3/2, and it calls them an unreliable basis for class.
By end use, Canada's Class D, C, B and A estimates sit closest to AACE Classes 5, 4, 3 and 1, but their percentages measure something different. PSPC defines its classes for building projects by the documents available and the design allowance the estimate may carry. Class D rests on a statement of requirements (20%), Class C on schematic design (15%), Class B on design development (10%) and Class A on completed drawings before tender (5%). The allowance is money in the estimate, not an accuracy range.
Sources and scope
Checked on . All sources were read on 9 November 2026. AACE's recommended practices are copyright: the tables restate their published values and the text paraphrases them; check the latest revision at web.aacei.org before relying on a value.
- AACE International, Recommended Practice 98R-18, Cost Estimate Classification System: As Applied in Engineering, Procurement, and Construction for the Road and Rail Transportation Infrastructure Industries, Rev. August 7, 2020: scope and exclusions (major long-span bridges, major tunnels, major buildings); program rated by its least defined major element; Table 1 and its 80% confidence and contingency note; systemic risks; the bias debate; ranges never pre-determined; Class 4 as the first public estimate; 30/60/90 reviews and Class 3 with Exceptions; Tables 2a to 2e; Table 3 maturity matrix; basis of estimate (section 7); long-range estimates and scope change (section 9)
- AACE International, Recommended Practice 142R-26, Cost Estimate Classification System: As Applied in Engineering, Procurement, and Construction for the Tunneling Industries, Rev. August 6, 2026 (sample pages): geotechnical studies cap a tunnel estimate's class; 30/60/90 reviews most commonly read as Class 4/3/2, by one reference as 3/2/1, and an unreliable basis for class
- AACE International, Recommended Practice 18R-97, Cost Estimate Classification System: As Applied in Engineering, Procurement, and Construction for the Process Industries, Rev. August 7, 2020 (sample pages): Table 1 for the process industries
- AACE International, Recommended Practice 56R-08, Cost Estimate Classification System: As Applied in Engineering, Procurement, and Construction for the Building and General Construction Industries, Rev. August 7, 2020 (sample pages): Table 1 for building and general construction
- AACE International, Professional Guidance Document 01, Guide to Cost Estimate Classification Systems, Rev. August 29, 2022: 17R-97 as the generic RP and the list of industry RPs; 98R-18's scope; class is not a contract specification for estimating quality; ranges come from quantitative risk analysis; accuracy values exclude scope change
- Treasury Board of Canada Secretariat, Glossary of Terms for TB Submissions (date modified January 5, 2017): indicative and substantive cost estimates
- Royal Architectural Institute of Canada, Canadian Handbook of Practice, Appendix A: classes of estimates used by PSPC for construction costing of building projects (reproducing PSPC's Cost Estimate Definitions, accessed September 14, 2020): Class D, C, B and A: basis, format and maximum design allowance
- US Government Accountability Office, Cost Estimating and Assessment Guide: Best Practices for Developing and Managing Program Costs, GAO-20-195G, March 12, 2020: the estimating process, from the technical baseline and work breakdown structure to sensitivity and risk analysis, documentation and updates
- FHWA, Major Project Financial Plan Guidance, issued December 18, 2014 (PDF, OMB control number 2125-0648): 23 U.S.C. 106(h) major projects of $500 million or more; the Cost Estimate Review, an unbiased risk-based probabilistic review within 12 months before the Initial Financial Plan; 70th percentile cost; year-of-expenditure dollars
- Bent Flyvbjerg, Mette K. Skamris Holm and Søren L. Buhl, Underestimating Costs in Public Works Projects: Error or Lie?, Journal of the American Planning Association, vol. 68, no. 3, Summer 2002 (author's version on arXiv): 258 projects worth $90 billion; costs underestimated in almost 9 of 10; actual costs on average 28% higher; rail 44.7%, fixed links (bridges and tunnels) 33.8%, roads 20.4%; fixed prices; estimate taken at the decision to build
This guide explains cost estimating practice for educational purposes. It does not replace the AACE recommended practices, an owner's own estimating manual or a qualified cost engineer's judgment.
Show us how your estimators take off design packages
Leave your work email, then pick a time for a 30-minute call on the design packages your estimators take off by hand, and at which class. You leave with a written plan: what an agent could prepare and what stays with your team. No commitment.