# Data dictionary — Powering Canada's Compute research dataset

Release `2026-09` · dataset `v1.0`. Two tables, both regenerated by
`python3 scripts/build_data.py`. Every `*_source_id` resolves to a row in
`data/source_manifest.csv`.

Provenance tags used below:

- **OBSERVED** — a value taken directly from a named primary source.
- **DERIVED** — computed from OBSERVED inputs; the formula is given.
- **SOURCE-FACT** — a dated qualitative statement quoted or paraphrased from a
  primary source.
- **ASSESSMENT** — this project's own categorical classification, made against a
  documented rubric (see `docs/METHODOLOGY.md`). Not a measurement.
- **CALIBRATION** — a disclosed scenario parameter, not an empirical value.

---

## `canada_compute_regions.csv` / `.json` — 5 rows

One row per province with a system operator that publishes large-load connection
guidance: Quebec, Ontario, Alberta, British Columbia, Manitoba.

| Column | Type | Meaning | Provenance |
|---|---|---|---|
| `region_id` | string | Stable key (`quebec`, `ontario`, …) | — |
| `region` | string | Province name | — |
| `province_code` | string | Two-letter code | — |
| `system_operator` | string | The entity that runs large-load connection: Hydro-Quebec, IESO, AESO, BC Hydro, Manitoba Hydro | SOURCE-FACT |
| `power_market_structure` | string | Coarse label: regulated vertically integrated utility, regulated wholesale market, or deregulated energy-only market | SOURCE-FACT (coarse) |
| `grid_ghg_intensity_gco2e_kwh` | number | 2026 electricity-consumption GHG intensity | **OBSERVED** — ECCC 2026 emission factors, Table 5.3 (`ECCC_GRID_2026`) |
| `grid_ghg_reference_year` | integer | Reference year for the intensity value | OBSERVED |
| `grid_ghg_source_id` | string | → `source_manifest.csv` | — |
| `large_load_connection_process` | string | How large loads connect in this province, as described by the operator | **SOURCE-FACT** (`large_load_source_id`) |
| `large_load_status_2026` | string | The one-line status of large-load access in 2026 | SOURCE-FACT |
| `large_load_source_id` | string | → `source_manifest.csv` | — |
| `large_load_source_date` | string | Date of the large-load source | — |
| `interconnection_access_assessment` | enum `LOW` / `MODERATE` / `MODERATE-HIGH` / `HIGH` | How hard it is, today, to get a new large load connected — categorical | **ASSESSMENT** (rubric in METHODOLOGY.md) |
| `interconnection_timeline_years_scenario` | integer | Indicative years from application to energization | **CALIBRATION** — a disclosed scenario value anchored to source language, not a measured queue time |
| `primary_bottleneck` | string | The single factor most likely to stop or delay a large development in this province | **ASSESSMENT** |
| `firm_power_context` | string | One line on the province's generation mix and what it means for a 24/7 load | SOURCE-FACT + ASSESSMENT |
| `notes` | string | Caveats, especially on the scenario timeline | — |

### Assessment rubric — `interconnection_access_assessment`

A three-part categorical rubric (documented in `docs/METHODOLOGY.md`):

1. **Rationing mechanism.** Is connection available on request, or allocated
   through a selection process / competitive call?
2. **Interim-block status.** Is there a published interim large-load capacity
   block, and is it still open or exhausted?
3. **Formal study process.** Does a multi-year system study with no published
   duration stand between application and energization?

`HIGH` = allocated by selection, not on request (Quebec).
`MODERATE-HIGH` = formal study with no published timeline, or an exhausted interim
block (Ontario, Alberta).
`MODERATE` = competitive call cycle, or near-term room with a closing window
(British Columbia, Manitoba).
`LOW` is defined but not currently assigned to any of the five.

---

## `technology_costs.csv` / `.json` — 7 rows

Generation technologies relevant to serving a large compute load.

| Column | Type | Meaning | Provenance |
|---|---|---|---|
| `tech_id` | string | Stable key | — |
| `technology` | string | Technology name | — |
| `overnight_capex_per_kw` | number | Overnight capital cost | **OBSERVED** — CER Canada Energy Future 2023, Electricity Supply Model, Table E.1 (`CER_ESM_2023`). Verified exact against the live CER page for all seven technologies. |
| `overnight_capex_units` | string | `CER EF2023 Electricity Supply Model benchmark, $/kW` | — |
| `overnight_capex_source_id` | string | → `source_manifest.csv` | — |
| `fixed_om_per_kw_yr` | number | Fixed operations & maintenance, per kW per year | OBSERVED — CER Table E.1 |
| `fixed_om_source_id` | string | → `source_manifest.csv` | — |
| `capacity_factor_2025` | number (0–1) | 2025 empirical utility-scale capacity factor | **OBSERVED** — U.S. EIA Electric Power Monthly 2025 (`EIA_CAPACITY_FACTOR`, `EIA_CAPACITY_FACTOR_FOSSIL`). A documented reference table, not a forecast for any Canadian site; live monthly values drift. |
| `capacity_factor_source_id` | string | → `source_manifest.csv` | — |
| `ref_lead_time_years` | integer | Reference-plant construction lead time | OBSERVED where `ref_lead_time_source_id` = `EIA_COST_PERF_2023` (U.S. EIA AEO2023 reference plants); `ILLUSTRATIVE` for solar, which the source table does not cover — treat as a planning placeholder, not sourced. |
| `ref_lead_time_source_id` | string | → `source_manifest.csv`, or the literal `ILLUSTRATIVE` | — |
| `lifecycle_ghg_gco2e_kwh` | number | Representative lifecycle GHG intensity | OBSERVED (gas: NETL 2025) / literature midpoint (low-carbon: `NREL_LCA_HARMONIZATION`). These are ranges; a single point is a representative value, not a site forecast. |
| `lifecycle_ghg_source_id` | string | → `source_manifest.csv` | — |
| `provides_firm_output` | boolean | Whether the technology can serve a continuous 24/7 load without separate storage or backup | ASSESSMENT — wind and solar are `False` |
| `capex_per_avg_mw_delivered` | number | Overnight cost per average MW actually delivered over a year | **DERIVED** = `overnight_capex_per_kw / capacity_factor_2025`, ×1000 by units. Arithmetic only. It does **not** account for firmness — a 24%-capacity-factor plant delivering an average 24% is not equivalent to a 91%-capacity-factor plant for a data centre that must run continuously. |
| `capex_per_avg_mw_delivered_note` | string | Restates the caveat inline | — |

### What CER's base year is

CER publishes Table E.1 as model benchmark `$/kW` without stamping a base
calendar year on that table. The dataset carries the values as transcribed and
labels the unit accordingly. Do not silently treat them as current-year CAD or
as U.S. dollars.