One province gives you the right to argue with a machine. The other three do not.

Where the four provinces sit

The Transmutarianism quadrant places an agent by two flows. Filtering (F), the vertical axis, is deprivation the agent absorbs without passing it on. Amplification (A), the horizontal axis, is fulfillment the agent emits in excess of what it received. Four archetypes fall out of the signs: a Transmuter (top right) absorbs deprivation and emits fulfillment; an Extractor (bottom left) does neither; an Absorber (top left) takes deprivation in and gives little back; a Magnifier (bottom right) emits fulfillment while failing to absorb, or adding to, the deprivation of the population it acts on.

Three of the four provinces emit fulfillment and pass deprivation on. The fulfillment is independently sourced: CBRE ranks Toronto the third-largest tech-talent market in North America, independent technical work supports the case Alberta's grid managers make for treating flexible datacentre load as a dispatchable asset that helps the system rather than strains it, and BC wrote a datacentre cap into regulation. All of it reaches firms, universities and investors.

Quebec absorbs and emits in near-equal measure: its W moves by 2.6 points across the entire τ band, which no other dot in this series does. A reader who weights filtering twice as heavily as amplification, or half as heavily, gets the same answer about Quebec.

The per-level scores, the stated assumptions, the math, and the full sensitivity set are in the methodology appendix. The excluded-populations denominator holds every placement. The gross-megawatt construction does not, and the appendix publishes what it moves.

Public-interest recommendations

Ontario

• Make the EDSTA regulations. The Enhancing Digital Security and Trust Act has been in force since 29 January 2025. Its AI provisions apply to "such public sector entities as may be prescribed", "in prescribed circumstances", "in accordance with the regulations". A search of the e-Laws regulation index on 8 July 2026 returns zero AI regulations made under it. Until they are made, Ontario's public-sector AI statute binds nobody.
• Define "specified load facility". Electricity Act s. 28.1 forbids connecting one to the grid without meeting regulatory requirements. It has been in force since 11 December 2025 and the term has never been defined, so the prohibition binds nobody and datacentres connect under ordinary load-service rules.
• Publish the Document Verification Service bias testing before go-live, and retest on a representative sample.
• Bring AI within the AODA. A full-text search of the Accessibility for Ontarians with Disabilities Act and its Integrated Accessibility Standards returns zero occurrences of "artificial intelligence", "algorithm" or "automated". The Act was last consolidated in 2016.

Alberta

• Give Albertans a right to human review. Alberta's own Information and Privacy Commissioner recommended it and the government declined. The Commissioner's report records that the new privacy statute "did not incorporate most of our recommendations".
• Publish the Allocated MW Trading Period trades. Qualified developers were permitted to exchange megawatt assignments with each other before final allocation. No record of those trades is public, so nobody outside AESO knows who holds the 1,200 MW.
• Publish a register of government AI systems. Alberta publishes none. Quebec publishes 258.

British Columbia

• Publish BC Hydro's supply agreement with TELUS. The three-site cluster (the Kamloops expansion plus two new Vancouver facilities) takes 85 MW initially, scaling to over 150 MW by 2032. The province's own regulation caps a project at 145 MW, and 85 MW is more than a quarter of the entire two-year AI allocation. BC Hydro has confirmed that the three data centres hold pre-existing energy-transmission agreements and sit outside the competitive call the province had just created; the agreements themselves and each site's allocation are unpublished, withheld as commercially sensitive.
• Publish the Call for Demand application register. BC Hydro has said it received 15 applications totalling close to 800 MW, roughly double the 400 MW ring-fence, so the cap bound; it has not published a register naming the applicants or their individual requests.
• Implement the 2021 recommendations. The Information and Privacy Commissioner and the Ombudsperson jointly recommended a duty to notify individuals of AI decisions and a mandatory AI Fairness and Privacy Impact Assessment. Neither exists, and no government response is on the record.
• Give the anti-racism AI commitment a deliverable. Its current deliverable is a multidisciplinary group that will develop guidance to inform a gap analysis of the government's AI policy framework. It is the only AI-equity commitment with a named lead ministry in any of the four provinces, and it creates no duty, no assessment and no remedy.

Quebec

• Enforce s. 12.1, or say why it cannot be enforced. The right has been in force for thirty-four months and has produced no published decision, penalty, investigation or audit. The Commission's own five-year review identifies the obstacle: the penalty-eligible breaches turn on subjective constitutive elements, and "based exclusively on automated processing" is one.
• Close the exclusivity loophole. As drafted, a human rubber-stamp anywhere in the chain takes a decision outside s. 12.1 entirely. Benefits, hiring and credit adjudication routinely carry a human signature.
• Publish the impact assessments. Quebec mandates a privacy impact assessment for every generative-AI project from conception, and for any other AI project touching personal information, and publishes none of the 258.
• Add a consent requirement for datacentre siting. The AFNQL has published a demand for consultation and consent before datacentres are built on First Nations territories. Quebec's regime contains no consent requirement.
• Make the CCTM recommendations binding. Eighteen recommendations on AI in the workplace were adopted unanimously by employer and union representatives on 29 April 2026. They landed on 15 June 2026 as a voluntary guide.
• Publish a successor AI strategy. The 2021-2026 strategy expires this year and no successor exists, so Quebec may enter 2027 with the only automated-decision right in Canada and no strategy at all.

British Columbia and Alberta together

• Use the jurisdiction you already have. BC, Alberta and Quebec are the only three provinces whose private-sector privacy laws are designated "substantially similar" under the federal Act, which is the vehicle Quebec used to reach every bank, insurer, landlord and employer making automated decisions about its residents. BC and Alberta hold the same vehicle and have put nothing in it: their private-sector acts return zero occurrences of "algorithm" and "artificial intelligence", and Alberta's returns zero occurrences of "automated" as well.

Federal

• Name the two-tier problem. Whether a Canadian can demand the reasons for a decision a machine made about them, and be heard by someone able to review it, currently depends on which province they live in. A benefits claimant in Gatineau has a right that a benefits claimant across the river in Ottawa does not.
• Fill the gap Ontario cannot fill itself. Ontario has no private-sector privacy statute, so automated decisions made about Ontarians by private actors are governed federally by default.
• Require published bias testing for any AI system touching a federally-funded benefit, in any province. No province in Canada publishes the bias testing for the AI systems already in the benefits pipeline.

What this evaluation is

The Transmutarianism framework scores an agent on two flows at each of five levels of human need, from the physiological base to cognitive sovereignty at the top. Filtering is deprivation the agent absorbs without passing it on. Amplification is fulfillment the agent emits beyond what it received. The two combine into a Moral Work value, W, and the signs of the aggregate flows place the agent in one of four quadrants.

The scoring is partly normative. Where a province sits depends on choices about who counts, what horizon is scored, and how public resources are treated. Every one of those choices is numbered in the appendix, and the two that could move a placement are recomputed and published there.

Four provinces were scored blind. Twelve independent scorers, three per province, each given only its own province's evidence, each forbidden from reading any prior field note or looking up any prior score. Nineteen of the twenty cells came back within two points across three different analytical lenses.

The four provinces at a glance

Three things all four have, at four different stages. Every figure is from a primary source.

The governance instrument: five binary questions

None of the four gives a right to human review in the sense the GDPR uses the term, a right to obtain human intervention in the decision itself. Quebec is the only one that gives the person an audience at all.

Declared government AI systems

The energy pipeline, at four stages

Energy: who gets the power, and on what terms

Alberta has allocated 1,200 MW of grid capacity to datacentres and holds a queue above 20,000 MW. Its Large Load Projects page names two assigned projects, GLDC Load at 970 MW and Keephills Phase I at 230 MW, which together account for the whole 1,200. AESO's own September 2025 update describes a funnel of fifteen qualified projects at 4,800 MW narrowing to three assigned at 1,200 MW. The current page names two. The same document discloses an "Allocated MW Trading Period" in which qualified developers could exchange megawatt assignments with each other before final allocation, and no record of those trades is public.

British Columbia's B.C. Reg. 8/2026 ring-fences 300 MW for AI datacentres and 100 MW for conventional ones over two years, caps any single project at 145 MW, and charges $25,000 per megawatt in bid security to compete. The province has used this lever before: it suspended new cryptocurrency connections in December 2022, catching 1,403 MW across 21 pending projects.

Quebec is going the other way: Hydro-Québec has asked the Régie de l'énergie to set a datacentre tariff near 13 cents per kilowatt-hour, roughly double the current large-power rate, with a decision expected around the end of 2026. Its actual datacentre fleet is about 190 MW.

Ontario's grid gate is Electricity Act s. 28.1, in force since 11 December 2025. It forbids a transmitter or distributor from connecting a "specified load facility" unless regulatory requirements are met. No regulation has been made, and "specified load facility" has never been defined, so the prohibition applies to nothing. The consultation that would have defined it closed on 4 November 2025 with no decision posted. The IESO publishes no datacentre connection queue.

Toronto alone carried 293 MW of datacentre demand in 2023 on the IESO's own figures, most Ontario datacentres sit around 10 MW on the distribution grid below any large-load threshold, and a press mapping counted at least fifteen proposed datacentres totalling 2,202 MW.

Sovereignty: whose data, whose territory, whose decision

Quebec's ministerial framework, mandatory since December 2025 with a compliance deadline of 5 June 2026, binds the public bodies subject to the province's information-resources statute: ministries, government agencies, school service centres and school boards, CEGEPs and universities, and the health and social services network. Municipalities are not subject to that statute, and commercial state corporations such as Hydro-Québec are covered only if the government extends the instruments to them by order. It requires a privacy impact assessment for every generative-AI project from its conception, updated through experimentation, development and deployment, and for any other AI project that collects, uses, communicates or keeps personal information. The annual declaration that follows produced the 258-system inventory.

The Indigenous organisations have published, and each asks for something its province does not provide. The AFNQL, in The Digital Territory of First Nations Quebec-Labrador (2025), demands consultation and consent before datacentres are built on First Nations territories. Quebec's datacentre regime contains no consent requirement of any kind. The Chiefs of Ontario, in First Nations and Artificial Intelligence (2024), ask for consultation on whether to use an AI system at all and a right to object to AI decisions. Ontario grants neither.

The Chiefs of Ontario paper aims both asks at the federal Artificial Intelligence and Data Act, which it mentions thirty times. It mentions the Government of Ontario, Bill 194, and datacentres zero times each.

The BC Assembly of First Nations' resolution packages for its 2025 Annual General Meeting and its 2026 Special Chiefs Assembly contain zero occurrences of "artificial intelligence" and zero of "data centre", while "data sovereignty" appears in both. BC First Nations have resolutions on data governance and none on AI, while the province ring-fences 400 MW. This note reads that silence as silence. It does not read it as consent.

Alberta is where the question is being litigated. Sturgeon Lake Cree Nation is in court against Alberta over the duty to consult on the Wonder Valley datacentre park, after the province's Aboriginal Consultation Office determined that neither the water licence nor the land transfer required consultation. Alberta First Nations are not uniformly opposed: Treaty 6 chiefs have said they want equity rather than only consultation, and Woodland Cree First Nation holds 51% of the Mihta Askiy project.

Labour: one rule, and what it requires

Ontario has the only AI provision in Canadian employment-standards law. Employment Standards Act s. 8.4, in force 1 January 2026, requires an employer who uses AI to screen, assess or select job applicants to say so in the posting.

Ontario's own guidance, updated 24 February 2026, sets the bar: "It is not necessary to provide a detailed description of the artificial intelligence system or the employer's use of the system. It is enough for the employer to state that artificial intelligence is used to screen, assess or select applicants." There is no bias audit, no notice to the individual applicant, no explanation right, no human review, no right to contest, and no recordkeeping of AI outputs. O. Reg. 476/24 exempts any employer with fewer than 25 employees.

Quebec produced a labour instrument and did not make it binding. Its tripartite labour committee released an avis on 29 April 2026 with eighteen recommendations adopted unanimously by employer and union representatives, covering algorithmic oversight, mechanisms to prevent discriminatory bias, occupational health and safety, privacy, and training. On 15 June 2026 the Labour Minister published a guide for workplaces. It is voluntary.

Alberta and British Columbia have no AI hiring rule, no displacement measure, and no AI retraining fund. The only AI-workforce item in an Alberta primary document faces inward: a roadmap of AI-driven productivity initiatives to transform the Alberta Public Service.

Urban: land, water, and who was asked

Wonder Valley, in Alberta's Municipal District of Greenview, appears on the province's own major-projects registry at $12.0 billion and 1.4 GW for its first phase, off-grid, status Proposed. Because it is off-grid, it sits outside AESO's queue and outside the 1,200 MW cap entirely.

Sturgeon Lake Cree Nation filed for judicial review in the Court of King's Bench on 7 October 2025, naming Alberta's Aboriginal Consultation Office, the Minister of Environment and Protected Areas, and the municipal district. The Water Act licence at issue (6 million cubic metres from the Smoky River, granted 7 April 2025) is held by the Municipal District of Greenview, not by a Wonder Valley company. Chief Sheldon Sunshine has said the Nation relies on water from the Smoky River, and that it is one of the few areas accessible to exercise their way of life. The Nation's separate appeal to the Environmental Appeals Board was dismissed on 17 April 2026 for lack of standing, on the finding that it was not "directly affected"; that board had no jurisdiction over the constitutional question, which is why the duty-to-consult claim sits in King's Bench.

Alberta's Aboriginal Consultation Office is alleged, in that filed application, to have determined orally that no duty to consult was triggered. The determination has never been produced. The Nation says it requested a copy and was refused, and that it does not know the date of the decision or the reasons for it. This note reports the allegation and does not decide it.

The host agreements are not public in any of the four provinces. Neither are the power purchase agreements, the land transfer consideration, or the grant agreements. The financial-filings class of this note's research protocol came back empty for all four provinces, and the searches are recorded in the ledger.

British Columbia wrote "economic, community, data sovereignty and environmental benefits" into the evaluation criteria for its 400 MW call, and BC Hydro's Request for Applications publishes the weightings they carry: cost and benefits to BC Hydro 30%, data sovereignty 30%, community and provincial economic benefits 20%, environmental impacts and benefits 10%, and First Nation benefits and impacts 10%. The "data sovereignty" criterion means Canadian data sovereignty, defined by the applicant's jurisdiction of formation, degree of foreign control and access, and the share of the facility allocated to digital sovereignty; no primary source links it to Indigenous data sovereignty, and the call carries a separate First Nation benefits and impacts criterion at 10%.

Ethical implications: what happens when the machine is wrong

Quebec's answer is a statute: sections 12.1 and 65.2 of its privacy acts, in force since 22 September 2023 and textually identical but for the words "person carrying on an enterprise" and "public body", require the decision-maker to tell a person when a decision about them was based exclusively on automated processing, to disclose on request the personal information used and "the reasons and the principal factors and parameters that led to the decision", to permit correction, and to give the person "the opportunity to submit observations to a member of the personnel… who is in a position to review the decision". A breach is penalty-eligible under s. 90.1(5).

The duty attaches only where the decision is exclusively automated, so a human rubber-stamp anywhere in the chain defeats the trigger. What must be disclosed stops at the principal factors and parameters: no model, no logic, no weights. And the remedy is an opportunity to be heard by someone with the power to review, which obliges nobody to review, re-decide or reverse. This is a transparency-and-audience right rather than a right to a new decision. The GDPR's Article 22 goes further on both counts, granting a right to obtain human intervention and a right not to be subject to such a decision at all. Quebec grants neither.

And in the thirty-four months it has been in force, it has produced no published enforcement. The Commission d'accès à l'information's own 381-page five-year review cites s. 12.1 three times and s. 65.2 twice, every occurrence descriptive; its penalty register is empty; its decisions database carries no automated-decision case. The regulator names the reason itself: the penalty-eligible breaches include elements that "nécessitent la démonstration d'éléments constitutifs subjectifs", and "based exclusively on automated processing" is exactly such an element. It still scores, because the other three provinces have nothing to test.

The other three provinces have no such right, and the statutes say so in a way anyone can check. Alberta's private-sector privacy act returns zero occurrences of "automated", zero of "algorithm" and zero of "artificial intelligence." Its public-sector act, in force since June 2025, gives the individual exactly one thing directly: s. 5(2)(d) requires a public body to disclose, at the moment it collects your information, its intention to feed that information to "an automated system to generate content or make decisions, recommendations or predictions." That is wider in scope than Quebec's trigger. The act's s. 6 adds duties on the body to keep the information behind an automated-system decision accurate and to retain it for at least a year, and Alta Reg 143/2025 requires privacy-management-program policies for automated and AI systems. None of it gives you anything afterwards: no explanation, no disclosure on request, no audience, no remedy.

British Columbia's two privacy acts return zero occurrences of "algorithm" and zero of "artificial intelligence." Each contains a single instance of the word "automated", and in both it is a power granted to the Commissioner to comment on the implications of automated systems: verbatim in PIPA s. 36(1)(g) ("comment on the implications of automated systems for the protection of personal information"), while FIPPA s. 42(1)(g) reads "comment on the implications for access to information or for protection of privacy of automated systems for collection, storage, analysis or transfer of information". It confers nothing on the person the system is used against.

Ontario has no private-sector privacy statute at all. BC, Alberta and Quebec are the three provinces whose private-sector laws are designated "substantially similar" under the federal Act; Ontario's only such designation is for health information. Automated decisions made about Ontarians by a bank, an insurer, a landlord or an employer are governed federally. Of the three provinces that hold the power to legislate an automated-decision right into their private sector, only Quebec used it.

Ontario's Auditor General measured the gap between the paper and the practice, and it is the only such measurement in Canada. Of the 400 AI websites Ontario Public Service staff accessed between April and August 2025, 244 were rated unsafe or unsecured by the government's own security tooling. Access was not blocked. Twelve thousand staff were involved, and unapproved generative-AI sites accounted for 94% of usage. Three percent of staff had completed the responsible-use training, which is not mandatory. In one procurement, eleven of twenty approved vendors submitted no third-party security audit and five submitted no privacy impact assessment, despite the tender requiring them. They were approved.

Ontario's first externally-facing AI system is the Document Verification Service. It uses facial recognition to verify a person's identity so they can register for government services online, and the Auditor General records that it is expected to extend to services "such as social assistance or health-related programs". Its go-live has not been announced. It was procured on vendor test reports from 2022 and 2023 whose limitations the Auditor General documents: the vendor's own 2022 report called its testing group of 214 individuals too small and not representative, and the Auditor General found the 2023 testing covered a limited sample of two age cohorts (18 to 30 and 51 to 70). The Auditor General found that the ministry did not challenge those limitations and that, at the time of the audit, there was no plan in place to conduct more representative testing or to monitor bias risks after deployment. The ministry agreed with the recommendation.

The Auditor General does not say the system is biased. It says the testing was inadequate to know, and that the consequence, if it goes wrong, is that "certain groups may experience higher rejection rates or delays when verifying their identities online to access government services."

Ontario's disability statute does not reach any of this. A full-text search of the Accessibility for Ontarians with Disabilities Act and its Integrated Accessibility Standards returns zero occurrences of "artificial intelligence", zero of "algorithm", and zero of "automated". The Act was last consolidated on 19 April 2016. A reader can re-run the search.

What changes the placement

Both mandatory alternative constructions were computed. The excluded-populations denominator flips nothing. The gross-megawatt diversion flips two provinces into Extractor. The four placements survive the whole τ band and all three weight vectors.

Count only the excluded, and the gap widens

Re-score each province against its low-income, unhoused, Indigenous, disabled and displaced-worker residents only, and every dot deepens while none flips. Ontario deepens hardest, from W = −28.99 to −49.50, because its amplification collapses from A = +12.7 to +2.7.

Quebec's Transmuter placement survives (W = +16.26). It survives because a right to be told and to be heard is held by the person it is used against, so a benefits claimant carries it exactly as a corporation does.

Book the megawatts gross, and Ontario is rewarded for inaction

The central construction scores the net subsidy: the tariff discount below cost, the tax credits, the land, the grid upgrades socialised onto ratepayers. Megawatts sold at a fair tariff are a commercial transaction. The alternative rejects that and books the gross megawatts as a negative physiological flow.

This construction flips quadrants. Booked the way the framework requires, as a negative physiological amplification flow, Alberta becomes an Extractor at four of the ladder's five rungs (100, 200, 300 and 400 MW per unit; W from −62.23 to −33.94), and British Columbia becomes an Extractor at the harshest rung (100 MW per unit, A_agg −3.3). Only the 600 MW rung leaves every placement standing. Quebec stays a Transmuter at every rung.

The ranking of Alberta and Ontario also inverts, and this is one of the two constructions in which Ontario is not last; the other is Alberta's band floor, published in the appendix. Ontario's shift is zero at every rung, so it climbs the table by the full distance the other three fall.

Ontario's zero is a bookkeeping artifact: its grid carries at least 293 MW of datacentre load in Toronto alone, and because "specified load facility" was never defined, distributors connect datacentres under ordinary load-service rules and nothing records it. A construction that scores the act of allocating rather than the fact of load hands its best outcome to the only province that never built the instrument that would have made its allocations visible.

The four numbers in that ladder do not even mean the same thing, and the appendix says so: Alberta's 1,200 MW is capacity assigned, BC's 400 MW is ring-fenced with nothing yet awarded, Quebec's 190 MW is metered load actually flowing, and Ontario's zero is what the province counted.

The power handicap

Every one of the four provinces controls a resource the counted population needs, so the framework requires a power band. Extractor membership needs both coordinates negative against the shifted origin, and the handicap moves the origin on both axes, so a province is an Extractor only past the later of its two crossings.

At a handicap of Δ_M = 2.0 per-level units, not one of the four holds its placement. British Columbia (W = −44.85), Alberta (−53.33) and Ontario (−58.99) are all Extractors. Alberta crosses first, above Δ_M ≈ 1.32. Because Quebec's filtering score is positive, the handicap strips that credit before it strips its amplification: Quebec leaves the Transmuter quadrant at Δ_M ≈ 1.89, becoming a Magnifier, and reaches Extractor at ≈ 2.83.

What would move the dots for real

Four documents are still to land, and they do not land together. Ontario's first AI strategy was promised in Budget 2026 for summer 2026; it is outstanding and still inside its window. Alberta's Phase 2A bring-your-own-generation framework is targeted for August 2026, proposing 1,600 MW of interruptible bridging capacity. BC's power awards land in September 2026, with supplemental applications closing 25 August. The Régie hears Hydro-Québec's tariff case 1 to 9 October 2026 and has published no decision date. Quebec's own AI strategy expires this year with no successor published.

The Foundation commits to re-running these four dots in Q4 2026 against what the provinces actually did. That revisit will not capture the Régie's decision, which is the single document most capable of moving Quebec's physiological cell.

A note on framing

A Magnifier is not a villain. All three of these provinces emit fulfillment, and the sources for it are independent of the provinces themselves. Toronto is the third-largest tech-talent market in North America. Alberta's grid operator has independent technical work behind its case for treating flexible datacentre load as a dispatchable asset. British Columbia wrote a datacentre cap into regulation and suspended 1,403 MW of crypto connections in 2022.

The fulfillment reaches firms, universities, investors and the general economy. The deprivation stays with the person whose benefit application was declined by a system nobody audited, whose job application was screened by a model nobody tested, whose territory was allocated without consent, and who has no one to ask.

Quebec did not solve this by spending more. It spent less on compute than Alberta by an order of magnitude, and it has asked its regulator to roughly double the price of datacentre load. What it did was give the person on the receiving end a door, and write down what its own machines are doing where anyone can read it. Two hundred and fifty-eight systems, declared. A right to the reasons. A person on the other side of the door who is in a position to review it.

The other three provinces have published principles, frameworks, directives, guides, action plans and strategies. A resident of any of them who is refused by a machine has, as of today, nowhere to take it.

How the placements were scored

Per Maslow level, F and A are scored from −10 to +10. The moral work at each level is M = [τ(F − F₀) + (A − A₀)] / √(τ²+1), the canonical normalized functional of Transmutarianism v19. The weighted total is W = Σ(w·M) with w = {5, 4, 3, 2, 1}. The aggregate dot is F_agg = (Σ(w·F)/15) × 10 and A_agg = (Σ(w·A)/15) × 10. Every number below was computed by a script whose self-test reproduces Field Note 01's published values from its own table.

Assumptions, stated

1. Scope. Each province as an AI-policy-making agent: strategy, public-sector deployment, compute and datacentre policy, privacy and regulation, labour, and measures for excluded populations. Federal programmes are excluded except where a province's own contribution is the thing being scored. The TELUS Sovereign AI Factory is a federal project selected under an Innovation, Science and Economic Development Canada initiative; no BC provincial dollar figure, funding instrument or co-signature appears in any primary source, so only BC's grid access is scored against BC.
2. Time horizon. As of 13 July 2026. Only what is in force and operative today earns positive F, and that rule is applied to all four provinces alike. The four load-bearing documents still to land do not land together: Ontario's AI strategy was promised for summer 2026 and is outstanding but not yet late; Alberta's Phase 2A framework is targeted for August 2026; BC's power awards fall in September 2026; and the Régie hears Hydro-Québec's tariff case 1 to 9 October 2026 with no decision date published. A revisit is committed for Q4 2026, and it will not capture the Régie decision.
3. Asymmetry coefficient (τ). Central τ = 1. Sensitivity band 0.8 to 1.5.
4. Maslow weighting. w = {5, 4, 3, 2, 1}. Lower-level deprivation is heavier. Geometric {16,8,4,2,1} and flat {1,1,1,1,1} are recomputed below, reported as W/Σw because Σw differs across vectors and raw W is not comparable. Every quadrant holds on every vector, and Ontario is lowest on every vector. The Alberta/Ontario ranking moves only under the gross-megawatt construction, and below τ ≈ 0.385.
5. F and A scale. −10 to +10 per level, central estimate, substitutable. Derived blind by three independent scorers per province (a material lens, a rights lens, and an excluded-populations lens), each given only its own province's evidence and each forbidden from reading any published field note or looking up any prior score. Nineteen of twenty cells came back within two points. The tables were fixed and committed before Field Note 03's per-level scores were opened. The adversarial gate then moved ten values across eight cells, and every one of those changes is logged with the evidence that compelled it in the ledger.
6. Denominator. All residents of the scoring province. Chosen because it keeps the four dots comparable with each other and with the federal comparator. Alternative: excluded populations within that province (low-income, unhoused, Indigenous, disabled, displaced workers). Recomputed and published below.
7. Treatment of the diverted resource. Central: net subsidy only. The tariff discount below cost of service, tax credits and grants, land, and grid upgrades socialised onto ratepayers. Megawatts sold at a fair tariff are a commercial transaction, and the framework caps the negative where a subject uses an existing public asset. A province charging datacentres above cost is not diverting. Alternative: gross megawatts allocated, booked as a negative physiological flow. The framework fixes no conversion from megawatts to per-level units, so the whole magnitude ladder is published below rather than one rung.
8. Power asymmetry (ρ and Δ_M). Central ρ = 1 (parity). The operational trigger holds for all four: every province controls grid capacity, and Alberta controls public land, that the counted population needs. A Δ_M band is therefore mandatory and is published below for all four provinces at the same three values. Extractor membership requires both shifted coordinates to be negative, and the handicap moves the origin on both axes, so the threshold is the later of the two crossings: Δ_M* = max[ (F_agg/10)·√(τ²+1)/τ , (A_agg/10)·√(τ²+1) ]. It is weight-dependent through both aggregates.
9. Amplification. Raw jobs, adoption, GDP and talent-ranking figures score near zero as amplification. This is Field Note 03's assumption 7, adopted here because FN03's dot appears on this note's chart, and two dots on one chart cannot be built on opposite rules for what counts as fulfillment emitted. It is the single most placement-determining choice in this evaluation, and it costs Ontario the most.

Two exposures, disclosed

Anti-anchoring is compromised. Field Note 03's published placement was read out of the master quadrant explorer's source earlier on the day this note was designed, and the design then called for that federal dot to appear on this chart. The four provincial blind tables were derived by scorers who were given none of it, and were fixed and committed to git before Field Note 03's per-level scores were opened. The ledger records what was seen and when.

British Columbia is the Foundation's home province. The Foundation is based in Vancouver. BC is also the thinnest record of the four and the only one with no independent defender: searches of Pembina, the CCPA, Power Advisory, the academic literature and the BCUC found independent commentary running against BC from both directions and nothing running for it. BC's dot is therefore published with the widest band of the four (W = −27.58 to −1.41, Magnifier throughout, though it becomes an Extractor at the harshest rung of the gross-megawatt ladder), and BC's placement is the most weight-sensitive: under geometric weights its F_agg rises from −24.0 to −11.9. The gate cut BC's physiological cell from +5 to +2 and its safety cell from −7 to −5, moving the home province in both directions on the evidence.

The four per-level tables

M and w·M on the unit scale, at τ = 1. Totals summed unrounded. Cells shown as (F, A). Ten values across eight cells were moved by the adversarial gate after the blind tables were fixed; the ledger carries the blind tables, these tables, and the evidence that compelled each change.

The sensitivity set

Every row below was computed by the derivation script. Two rows flip a quadrant: the power handicap, and the gross-megawatt diversion. Two rank claims are weaker than the placements, and both are stated here.

The two published bands

British Columbia and Alberta each carry a published band rather than a point, because each has one cell the record cannot pin down. Neither band crosses a quadrant at ρ = 1. Alberta's band floor (W = −37.48) sits below Ontario's central placement (−28.99), so the "Ontario lowest" claim holds against Alberta's central value and not against Alberta's floor.

The gross-megawatt construction's own defect. Ontario's zero is a bookkeeping artifact. Toronto alone carried 293 MW of datacentre load in 2023, most Ontario datacentres sit around 10 MW on the distribution grid below any threshold, and a press mapping counts at least 2,202 MW proposed. Because "specified load facility" was never defined, the connection prohibition binds nobody.

The audit protocol

Seventeen primary documents would replace the estimates in this note with measurements. The four that would move a dot furthest: BC Hydro's supply agreement with TELUS for the three-site cluster's initial 85 MW, which two of the three BC scorers named unprompted as the document that would change their table; BC Hydro's Call for Demand application register, which BC Hydro summarizes as 15 applications totalling close to 800 MW and has not published; AESO's Allocated MW Trading Period trades, without which nobody knows who holds Alberta's 1,200 MW; and the Document Verification Service's bias test reports, which Ontario's Auditor General has seen and the public has not.

One gap is common to all four. No province in Canada publishes the bias testing for the AI systems already touching benefit recipients. Quebec mandates a privacy impact assessment for every generative-AI project and for any AI project touching personal information, and publishes none for its 258 declared initiatives. The other three mostly do not require one.

Sourcing for this note is public record only. No request for comment was made and no freedom-of-information request was filed. Every absence above is recorded in the ledger with the searches that produced it.

The Foundation offers framework, scoring template, methodological support, and the published report to any party willing to supply them.

To dispute any placement: substitute different F and A values per level, with a source for each, and the dot moves. The assumption-challenge ledger records the research coverage by source class, the twelve blind scorer tables with their spread, the ten values the adversarial gate moved and the evidence that compelled each, the eleven measured zeroes with the exact searches that produced them, the missing-document register, and every challenge raised against this note before publication.