Reading the Footnote
What ALTO’s $60–90 billion cost estimate actually means — and what the AACE Class 5 label in the footnote tells the public that the headline figure does not.
On May 8, 2026, ALTO published a blog post titled How Much Will Alto’s High-Speed Rail Cost Canadians and how is it Funded?. The headline figure is $60 to $90 billion. A footnote attributes the estimate to “the Class 5 guidelines set by the Association for the Advancement of Cost Engineering International.”
That footnote is doing the analytical work in the disclosure. This brief explains what it means and how to read it.
The AACE Class 5 designation in the footnote is the lowest-accuracy cost estimate class in the global standard, intended for concept screening before engineering, geotechnical investigation, station design, or construction contracting have been completed. The accuracy range associated with Class 5 estimates is −50% to +100%.
Applied honestly to ALTO’s stated $60–90 billion range, that means the realistic outturn range is approximately $30 billion to $180 billion — three to four times wider than the headline range suggests, and skewed toward the upper end.
This is not a critique of ALTO for being uncertain about cost at the concept stage. Substantial uncertainty is appropriate at this stage. The question is whether the disclosure communicates that uncertainty in a form the public can act on.
The AACE classification system
The AACE International Cost Estimate Classification System is the global standard for describing the maturity and reliability of capital project cost estimates. It defines five classes, numbered from 5 (the least mature) to 1 (the most mature). Each class is tied to a specific stage of project definition and carries a characteristic accuracy range.
| Class | Purpose | Project definition | Typical accuracy |
|---|---|---|---|
| Class 5 | Concept screening | 0% – 2% | −20% to −50% / +30% to +100% |
| Class 4 | Feasibility study | 1% – 15% | −15% to −30% / +20% to +50% |
| Class 3 | Budget authorization | 10% – 40% | −10% to −20% / +10% to +30% |
| Class 2 | Control or bid | 30% – 75% | −5% to −15% / +5% to +20% |
| Class 1 | Check estimate | 65% – 100% | −3% to −10% / +3% to +15% |
Class 5 is intended for what AACE calls “screening of viable alternatives” — deciding whether to advance a concept to further study, not committing public funds. At 0–2% project definition, there is no detailed alignment, no completed geotechnical investigation, no station design, no electrical engineering, and no signed construction contract. The estimate is built from per-kilometre parametric assumptions drawn from comparable projects, scaled for length, and adjusted judgmentally for context.
The accuracy range is wide for a reason: the engineers preparing the estimate genuinely do not know what they will eventually be building. And the range is asymmetric. The upside risk (+30% to +100%) is roughly twice the downside risk (−20% to −50%) — reflecting more than fifty years of empirical experience that infrastructure cost estimates are more likely to be too low than too high.
Applied to the midpoint of ALTO’s $60–90 billion range, the AACE accuracy band of −50% to +100% produces a realistic outturn range of approximately $37.5 billion to $150 billion. Applied to the upper bound of $90 billion, the upside-risk range extends to roughly $180 billion. The $60–90 billion figure is not a budget envelope; it is the centre of a much wider statistical distribution that current information cannot narrow.
Which side of the range to expect
The asymmetry in the AACE accuracy ranges — more upside than downside — is not arbitrary. It reflects more than fifty years of empirical research on infrastructure megaproject cost outturns. The leading scholar in this field is Bent Flyvbjerg, professor at the University of Oxford’s Saïd Business School, who has spent more than twenty-five years compiling the largest dataset of large-project cost outturns in the world. His findings — summarized in Megaprojects and Risk (Cambridge University Press, 2003), in How Big Things Get Done (Currency, 2023), and in several decades of peer-reviewed papers — are remarkably consistent.
Nine in ten go over
Of every ten large infrastructure megaprojects studied, nine exceed their original cost estimate in real (inflation-adjusted) terms. The pattern is not isolated to any one country, sector, or political system; it holds across the Flyvbjerg dataset spanning more than a hundred projects and seventy years.
Rail averages roughly 45%
For rail projects specifically, the average cost overrun is approximately 45% in real terms. The standard deviation is large, meaning many projects overrun by considerably more than the average; a smaller number come in close to estimate.
High-speed rail tends higher
High-speed rail tends to overrun more than conventional rail, for two converging reasons: greater engineering complexity (tighter alignment tolerances, electrification, signalling, grade separation), and the fact that the political case for HSR often rests on ridership forecasts that subsequently prove optimistic.
Fat tails, not bell curves
The distribution of cost outcomes is “fat-tailed”: extreme overruns are more common than a normal distribution would predict. A small but significant fraction of large infrastructure projects overrun their original estimate by more than 100%. The mean and the median therefore tell different stories.
Flyvbjerg’s framework is now incorporated, in various forms, into the cost-estimation guidance of HM Treasury (the UK Government’s “Optimism Bias” supplementary guidance to the Green Book), the Australian Department of Infrastructure, and a growing number of comparable institutions. The technical name for the practice is reference-class forecasting: instead of building a project cost estimate from the inside out (this is what we think it will cost, based on our project), the estimate is calibrated against the actual outturn experience of comparable past projects.
What comparable projects have cost
International high-speed rail provides the relevant reference class for any forecast of ALTO’s eventual cost. Three large representative HSR projects in democracies with mature engineering and procurement institutions illustrate the pattern:
| Project | Initial estimate | Outcome |
|---|---|---|
| California High-Speed Rail | US$33 billion 2008 | Most recent California Legislative Analyst’s Office and Authority business plan estimate for full Phase 1: ~US$128 billion. The line is not yet operating. |
| HS2, United Kingdom | ~£33 billion 2010, for the full Y-shaped network | Pre-cancellation full-network estimates reached ~£100 billion or more. The northern phases were cancelled in 2023; the truncated London–Birmingham line continues, at lower total but higher per-kilometre cost. |
| Channel Tunnel | ~£4.65 billion 1985 | Final cost: ~£9 billion. Real overrun of roughly 80%. Among the most extensively studied infrastructure cost outturns in the academic literature. |
These are not handpicked outliers. They are large representative HSR megaprojects in democracies with mature engineering and procurement institutions. The pattern they show is consistent with Flyvbjerg’s broader dataset, and is the empirical basis for the asymmetric accuracy band in the AACE classification.
For ALTO at $60–90 billion Class 5, a reference-class-adjusted central estimate — using the historical outturn distribution of comparable HSR projects — would place the expected outturn meaningfully above the stated upper bound. The exact figure would depend on which reference class is chosen and which adjustment factor is applied; but a defensible central estimate is well into nine figures, and the upper tail of the distribution is materially higher still.
What ALTO’s post does not say
ALTO’s May 8 blog post discloses a Class 5 cost range, a brief description of the funding model, the existence of risk-sharing with the Cadence consortium, and the federal investment commitments made to date. What it does not disclose — and what would be necessary to evaluate the project on the merits — falls into four categories.
Reference-class adjustment
The post does not say whether the $60–90 billion range is itself a reference-class-adjusted estimate or a bottom-up Class 5 estimate prior to such adjustment. The distinction matters: if the range is bottom-up, the empirical literature would place the expected outturn substantially above the stated upper bound.
Sensitivity analysis
The post does not show how the estimate moves in response to specific parameters — ridership, modal shift from car and air travel, construction cost intensity, financing cost, fare-revenue assumptions. A megaproject cost discussion without sensitivity analysis cannot support an informed public judgment.
Benefit-cost framework
A cost figure cannot, on its own, answer whether a project is a sound public investment. The standard framework — benefit-cost ratio and net present value — requires both quantified benefits and quantified costs, evaluated against alternative uses of the same capital. ALTO’s blog post discloses neither.
Funding model in quantified terms
“A blended model of private capital, fare revenues, and targeted public investment, with construction and operating risks shared with Cadence” describes a structure but does not quantify any of its components. The basic question — what share of the project’s lifetime cost is borne by the taxpayer versus the fare-paying passenger versus the private partner — cannot be answered from the post as written.
None of these omissions are unique to ALTO. They are common features of project-promoter disclosures at the concept-screening stage of capital projects. But the public interest is in having them addressed, not in having them omitted from the only published cost statement.
What “self-sustaining” leaves out
The definitional-line dynamic that runs through the AACE footnote also appears in the government’s parliamentary answers about whether public subsidies will be required.
In response to Order Paper Question Q-923, answered April 22, 2026, the Minister of Transport stated that “operations are expected to be financially self-sustaining, with revenues covering operations and maintenance costs and eliminating the need for ongoing operating subsidies.” Independent academic analysis published by Transportation Research at McGill (Zhang, Negm, El-Geneidy, 2025) — a Queen’s- and NSERC-funded study that describes HSR throughout in favourable terms — reaches the same narrow conclusion about operations using ALTO’s own published cost figures, and then continues the calculation. The McGill model projects average annual public subsidies of approximately C$1.23 billion to cover capital-repayment obligations, totalling C$61.62 billion before the system reaches full cost recovery in Year 48. The “self-sustaining” framing is technically correct for operations narrowly defined; what it leaves out is the roughly C$3.66 billion in annual capital repayments the public pays separately.
The structural pattern is identical to the AACE footnote: a technically accurate statement at the top, a definitional line drawn in language most readers will not unpack, and the substantive public obligation kept out of the headline. A reader who acts on the headlines alone — “$60–90 billion” and “self-sustaining operations” — arrives at a picture of the public commitment materially different from the picture the underlying technical material supports.
Three questions to ask
Class 5 estimates are not, in principle, inadequate for public communication. They are part of how megaprojects are normally discussed at the concept stage. What is inadequate is presenting a Class 5 estimate as if it were a budget envelope, and burying the methodology in a footnote. The next time a federal infrastructure project releases a cost statement — from ALTO, or from any other proponent — three questions are worth asking.
- What is the AACE class of the estimate, and what accuracy range does that imply when applied to the stated figure? A Class 5 figure with a −50%/+100% band tells the public something very different from a Class 3 figure with a −20%/+30% band.
- What reference class of comparable past projects has been used to calibrate the estimate, and what does the historical outturn distribution for that reference class suggest about the realistic outturn range?
- What benefit-cost analysis accompanies the cost estimate, and what does it show about whether the project is a sound use of the same capital that could otherwise fund alternatives?
ALTO’s May 8 post answers none of these questions clearly. Whether the answers, when disclosed, support proceeding with the project on the terms now contemplated is a separate question — but the public cannot evaluate that question from the information currently available.