The Borovoy Coefficient scores Canadian statutes, regulations, and bylaws from 0 to 100 across five dimensions of legislative alignment — each grounded in recognized constitutional doctrine and in the machine-learning alignment taxonomy.
The Coefficient does not ask whether a law is good. It asks, precisely: does the language do what the preamble says? Are the rights-affecting effects proportionate to the salutary objective? Is the conduct captured the conduct drafters targeted? Is the severity of intrusion survivable? Is the bureaucracy administering the law optimizing something different than what the statute chartered?
These are Borovoy's questions from five decades of civil-liberties practice. They are also — with no stretching required — the inner-alignment, outer-alignment, and Goodhart-variant questions that machine-learning researchers ask of learned systems.
Purpose-Effect Gap · Proportionality · Overbreadth · Rights Intrusion · Enforcement Asymmetry
A sample — not a census. The corpus demonstrates the Coefficient's range across federal, provincial, and municipal jurisdictions, from Vancouver's Noise Control Bylaw (BC = 12, aligned) to the pre-Charkaoui Security Certificate regime (BC = 86, unconstitutional-by-design). Click any row for the full reasoning, evidence, drift vectors, and proposed reforms.
Alan Borovoy served as General Counsel of the Canadian Civil Liberties Association for forty-one years, from 1968 to 2009. He was awarded the Order of Canada in 1982 and four honorary doctorates. He wrote five books — When Freedoms Collide, Uncivil Obedience, The New Anti-Liberals, Categorically Incorrect, and At the Barricades.
His diagnostic frame — the intent-effect gap, the power-hoarding fallacy, the mesa-institutional drift of regulatory regimes — carried him from the October Crisis of 1970 through the post-9/11 anti-terrorism statutes to the sustained critique of human rights commissions that he himself had helped establish. He believed governments miss the mark when they legislate broadly in moments of shock and trust judicial review to clean up afterwards. History keeps vindicating him.
The Borovoy Coefficient is an attempt to translate that diagnosis into measurement. It is not a replacement for Borovoy. It is an instrument that he might, I hope, have used.
There is an important distinction between intent and effect. The question we must ask is whether the language, regardless of the intent, is broad enough to cover that activity.
Measurement without reform is a tribute essay. The full methodology is prospective: drafters can embed these five principles at the specification stage to close the outer- and inner-alignment loops before a statute reaches a bureaucracy.
Three published formulas, three depths of legal analysis, one transparent method. Each tier answers a sharper question than the last; each publishes its own math so readers can verify the arithmetic rather than argue with a composite.
Prima facie natural-language inference over preamble–provision pairs. A bipartite entailment matrix is computed between the statute's stated purposes and its operative provisions, then reduced to two scalar features — faithfulness and excess — under a published linear objective. No case law, no adjudicative facts, no doctrinal grounding: only structured textual entailment on the face of the statute.
MTM-L = 100·[α·(1−F) + β·X]
The Charter-violation layer, in two sequential stages. Stage 1 asks whether the statute's purpose, constructed per R v Big M Drug Mart, is itself Charter-offside — a finding that resolves the case without reaching effects. Stage 2 asks, purpose notwithstanding, how likely the statute's effects on face are to violate a Charter right (overbreadth, arbitrariness, gross disproportionality, rights-intrusion severity). Oakes is not engaged at this tier: Oakes presupposes an established violation. Tier II asks whether one exists.
MTM-F = 100·[PV + (1−PV)·EL]
The Oakes justification layer, with facts. Presupposing a Tier II violation, Tier III asks whether the government can justify it under s.1 — pressing-and-substantial objective, rational connection, minimal impairment, balance of effects — with adjudicative and legislative facts, disparate-impact evidence, and enforcement data entered. This is the tier most closely aligned with actual judicial-outcome prediction, but it is the most fact-intensive and therefore the last to automate.
MTM-C = 100·[γ₁·MTM-F/100 + γ₂·DI + γ₃·EF]
Every tier publishes its math. No black box. Licensed Mozilla Public License 2.0. Free regardless of future ownership outcome. Calibration report at calibration/index.html.