# Methodology — Powering Canada's Compute (research brief) Release `2026-09` · dataset `v1.0` · last verified 2026-09. This document covers the **research brief and dataset**. The playable simulation at `powering-canadas-compute.raj-gandhi.workers.dev` has its own, longer methodology (`../METHODOLOGY.md` in the model repo); the two share a source base and the same discipline about what is observed versus assumed. --- ## 1. Research question > **What actually determines whether a Canadian region can host the next > generation of large-scale AI and data-centre development?** The brief answers it through the constraint chain a real site runs into — generation, transmission, interconnection, land, network — and a five-province comparison built on primary sources. ## 2. Scope and boundaries - **In scope:** grid GHG intensity, large-load connection process and status, and a categorical interconnection-access assessment, for five provinces (Quebec, Ontario, Alberta, British Columbia, Manitoba); and a generation technology cost/performance spine. - **Out of scope:** industrial land inventories, fibre-route maps, water/watershed modelling, a project-by-project data-centre register, and any forecast of demand, price, jobs, or investment dollars. Where the underlying research could not source one of these to a primary document, it is named as a limitation (§7) rather than estimated. - **Not an endorsement.** No utility, regulator, or firm has reviewed this. The brief is an independent compilation. ## 3. Source hierarchy In descending preference: Canadian system operators and regulators (IESO, AESO, Hydro-Quebec, BC Hydro, Manitoba Hydro, CER, OEB); Canadian federal statistics and environment agencies (ECCC, StatCan); intergovernmental analysis (IEA, IPCC); U.S. government reference datasets used as documented benchmarks where no Canadian equivalent exists (EIA, NREL, NETL); and company or regulatory filings for project-specific facts (OPG). News is used only to confirm an event, never as a market-wide claim. The full list, with URLs and retrieval dates, is `data/source_manifest.csv` (35 sources). Every source has a stable id (`ECCC_GRID_2026`, `AESO_LARGE_LOADS_2026`, …). Every data value cites one. `data/claims.csv` maps the brief's headline statements to their source and classification. ## 4. Observed, derived, assessed, calibrated The dataset keeps four kinds of value distinct, and the brief labels them: | Kind | Example | How it is treated | |---|---|---| | **Observed** | Quebec grid intensity 1.9 gCO2e/kWh (ECCC 2026) | Transcribed from the source with its reference year. | | **Derived** | Capex per average-MW delivered | Computed from observed inputs; the formula is printed. | | **Source-fact** | "AESO's interim 1,200 MW block is fully allocated" | A dated qualitative statement from a primary source. | | **Assessment** | `interconnection_access_assessment = HIGH` for Quebec | This project's categorical classification against the rubric below. Not a measurement, and deliberately not a 0–100 score. | | **Calibration** | `interconnection_timeline_years_scenario = 4` | A disclosed scenario parameter, anchored to source language, explicitly not a measured queue time. | ## 5. The interconnection-access rubric `interconnection_access_assessment` is a categorical label, not a number. It is assigned against three questions: 1. **Rationing mechanism** — is a new large load connected on request, or allocated through a selection process or competitive call? 2. **Interim-block status** — is there a published interim large-load capacity block, and is it still open or already exhausted? 3. **Formal study process** — does a multi-year system study with no published duration stand between application and energization? `HIGH` — connection is allocated by selection, not available on request (Quebec). `MODERATE-HIGH` — a formal study with no published timeline, or an exhausted interim block (Ontario, Alberta). `MODERATE` — a competitive call cycle, or near-term room with a closing window (British Columbia, Manitoba). `LOW` — defined for completeness; not assigned to any of the five in this release. Two independent reviewers applying this rubric to the same five sources should land on the same labels. That is the test it is meant to pass; a 0–100 "readiness score" would not. ## 6. Why capacity is not the same as available data-centre power A province's installed generation says little about whether a specific 24/7 load can be energized on a developer's timeline. Between the two sit: firm (not just nameplate) supply; transmission to the region; local substation capacity; the interconnection process; and the build schedule for any new generation. The brief's central finding — that access is *rationed* in all five provinces regardless of how clean or how large the provincial grid is — follows from holding those steps separate. The IEA's observation that a data centre can be built in roughly 2–3 years while its power infrastructure takes longer (`IEA_ENERGY_AI_2025`) is the same point from the demand side. ## 7. Limitations - **No practitioner review.** No energy or grid-planning professional has checked the assessments or the rubric. - **Fast-moving.** Large-load connection frameworks in all five provinces are being actively rewritten in 2026; the dataset is a snapshot, versioned by release date. - **System-wide, not local.** Grid intensity and connection process are province-level; an actual project depends on a location-specific utility study. - **No land, fibre, or water data.** These matter and are named in the constraint chain, but standardized Canadian data for them was not sourced for this release. Treated as future work, not estimated. - **No project register.** Announced Canadian data-centre projects are not compiled here; capacity definitions (IT load vs utility load vs campus) differ too much to combine without primary filings. - **Benchmark transfer.** Lead times and capacity factors use U.S. EIA reference data as documented benchmarks; they are not Canadian permitting or performance forecasts. CER Table E.1 carries no stamped base year. - **Scenario timeline.** The 2–4 year interconnection timelines are calibration, not measurement (§4). ## 8. Reproduction ```bash python3 scripts/build_sources.py # data/source_manifest.csv, data/sources.json python3 scripts/build_data.py # regions + technology tables (CSV + JSON) python3 scripts/validate_data.py # schema, source-id, unit and rubric checks python3 scripts/build_brief.py # brief/brief.html from the data above ``` Then render the PDF from `brief/brief.html` (headless Chrome `--print-to-pdf`, or any HTML-to-PDF tool). The website and the PDF read the same generated `data/` files. See `README.md`.