How the dataset is built, and where every figure comes from

Sources

The base layer is the eight provincial licensing registries: British Columbia, Ontario, Quebec, Alberta, Saskatchewan, Manitoba, New Brunswick and Nova Scotia. Each province publishes its own list of licensed programs under its own rules, in its own format, on its own schedule. We ingest the province’s own file or portal, never a third-party aggregator, and we keep the identifier the province assigned to each record. The remaining provinces and the territories are not covered yet; when a registry becomes available, it arrives as a new edition rather than a silent backfill.

Registry scope differs, and the difference matters more than any single count. British Columbia and New Brunswick license small family child cares one at a time, so each home is a row. Ontario licenses home child care through agencies, so thousands of homes sit behind a few hundred agency records. Quebec’s home-based care runs through coordinating offices and does not appear as individual listings at all. Program counts are therefore counts of licensed points of access a parent can look up. They say nothing about how many spaces exist behind those licences, or how many children are served.

Website counts are frozen to the quarterly snapshot (metric_snapshots, 2026Q3) at 7,747; MapleCub’s website-discovery process keeps running between editions, so the live directory shows more — 8,369 when it was read on 2026-08-11. The frozen figure is the citable one, and the gap is published rather than quietly closed. Website presence is not capped by Google matching (a website can arrive from a registry field or from our own harvest) but it is conditioned by it: 62.2% of matched listings carried a website against 25.3% of unmatched ones.

Cleaning and identity

• Deduplication. 290 duplicate records were identified and merged in the July 2026 dedup pass and remain excluded from every published figure. Duplicates arise when a province re-issues a licence under a new number or when the same facility appears under both a legal and an operating name.
• Registry identifiers. All 20,516 records carry the provincial registry identifier they were ingested with, so any row can be traced back to the province’s own listing.
• Geocoding, and what “geocoded” is worth. 20,481 of 20,516 records (99.8%) carry coordinates — and that number on its own is close to meaningless, so we no longer publish it on its own. Of those, 9,400 (45.9%) are address-level and 11,081 (54.1%) are the centroid of a postal district or a city. A point counts as a centroid when more than 3 visible listings on one identical lat/lng, resolving to more than 1 distinct street address: one address carrying several licences is a building, several addresses on one point are not. The threshold errs toward calling a centroid an address, so the centroid share is a floor.
• Where the coordinate is an address, and where it is not. Address-level in British Columbia, Nova Scotia, New Brunswick, Alberta — Alberta everywhere except its two largest cities — and a postal-district centroid in Ontario, Quebec, Saskatchewan, Manitoba. Saskatchewan's S0K resolves 57 listings in 27 separate towns to one point; Alberta pins 314 Calgary listings and 256 Edmonton listings to one point each. No count in the flagship report is computed from coordinates. The one analysis that is — the distance from a one-option community to the nearest other licensed programme — runs only in the four address-level provinces, and says so. An earlier edition of this page said “a small number of Manitoba records geocode to a municipal centroid rather than a street address,” which was wrong three times over: not a small number, not Manitoba alone, and not municipal.
• Coordinates that are missing entirely. Manitoba is the only province with records carrying no coordinate at all (1,104 of 1,139); the gap is addresses the registry publishes in a form no geocoder resolves. That is a separate and much smaller problem than the centroid split above, and listing it alone used to read as though the other seven provinces were fine.
• Communities. A community is a normalized city name — lower-cased and trimmed, empty values dropped. Manitoba and Saskatchewan both ship case variants of the same town, which a raw grouping would count twice; 47 Manitoba rows and 8 Saskatchewan rows carry no city at all and sit outside the community counts, and 35 of those 47 Manitoba rows carry no coordinates either. A 2026-08-10 repair merged registry-supplied city abbreviations in Manitoba (e.g. “Wpg”) into their proper communities wherever the community was determinable, and set the 12 that stayed unrecoverable to null; those 12 are part of the 47.
• Google matching, and why it is never a province ranking. Websites, ratings and review counts are matched conservatively: name agreement is required, so the figures undercount rather than overcount. The undercount is not uniform, which is the part that matters. We hold a Google place record for 9,252 of the 19,569 programmes in the seven enriched provinces — from 89.3% in Quebec down to 6.5% in Saskatchewan. Since a listing cannot carry a rating we never looked for, the published “share with a rating” is bounded by the match rate, and in three provinces it IS the match rate to within a rounding point.
• The floor for a claim about a province. With match coverage m and a rated-among-matched share r, the true share lies between m·r and m·r + (1 − m). We state a finding about a province’s sector only where that interval is under 40 points wide — match coverage of at least 60%, which admits Quebec, Alberta, Nova Scotia and nobody else. Provinces that have not been through the matching step at all are excluded from presence figures rather than reported as zero. New Brunswick is in that state today, which is why the presence denominator is 19,569 rather than 20,516 — and the floor above is that same rule, applied to every province instead of one.
• Age bands are a coverage field too. Where a registry publishes age groups we can separate out-of-school-care programmes from early-years ones. Coverage ranges from every row in Quebec and New Brunswick to roughly one row in six in British Columbia, Ontario and Saskatchewan, so a count of school-age-only programmes is a floor in three provinces and close to a census in five. The flagship report publishes the per-1,000 rate on all licensed programmes for that reason, with the exclusion shown beside it and ranked nowhere.

Source: MapleCub Canadian Childcare Dataset — Aug 2026 · computed 2026-08-10 · N = 20,516 · CSV

External figures

Child population denominators are Statistics Canada, Table 17-10-0005-01 (July 1, 2025 estimates) — the July 1, 2025 estimates for ages 0 to 4, total gender. Fee and subsidy rules are verified against each province’s own regulation, funding guideline or parent-facing page (2026-07-05 for the seven provinces in our audited rules module; New Brunswick’s 2026 Parent Fee Grid is cited to the Government of New Brunswick directly). Inspection and enforcement counts are the provinces’ own published records, reported as dated extracts with the window they cover, because re-downloading those files today will not reproduce our numbers.

What MapleCub does not do

We do not compute safety scores. We do not rank providers by inspection history, and we do not turn violation counts into stars, grades or league tables. Provinces publish inspection data on wildly different terms — three as bulk open data (as dated extracts of files that keep moving), three one facility at a time, two not at all — and even the three per-facility routes are not the same route: New Brunswick publishes inspection reports, British Columbia publishes summary reports of inspections and substantiated complaints through five separate health authorities, and Ontario publishes only confirmed violations and offences while withholding its own file of assessment results. Any comparative score would therefore mostly measure which government publishes more, not which centre is safer. “No violations on record” in a province that publishes no records means nothing, and we say so rather than letting a blank field imply a clean history.

The same discipline applies to our own coverage fields, which is the change this edition makes. A field that records how far a MapleCub pipeline has reached — how many listings we matched to Google, whether a coordinate is an address or a postal-district centroid, whether a registry told us a programme’s age band — is published as coverage, with the denominator that says how much of the province it observed, and no province is ranked on one. We publish what we measured, labelled as what it is.

Cadence and versioning

We snapshot the dataset quarterly. Each edition carries the date it was computed, and published editions are not rewritten in place — when a province is added or a registry reissues its file, the next edition states what moved. Citations should name the edition date, so a figure quoted in July still checks out after an August edition changes the national total.

Licence and citation

The published province-level dataset is available under CC BY 4.0 — reuse it commercially or otherwise, with attribution.

MapleCub, “The State of Licensed Child Care in Canada, 2026,” 2026-08-10 edition. https://doi.org/10.5281/zenodo.21909631

That DOI always resolves to the newest edition. To pin a figure to the exact deposit it came from, cite this edition instead: https://doi.org/10.5281/zenodo.21909632. The archived deposit — CSVs, JSON, the report PDF and a README setting out how these numbers are most often misread — is at Zenodo.

Machine-readable JSON for developers: /data/state-of-child-care-2026.json — same figures as the CSV, same licence.

Source: MapleCub Canadian Childcare Dataset — Aug 2026 · computed 2026-08-10 · N = 20,516 · CSV

Looking for the product side instead — how search results are ordered, what we verify before a listing appears, what providers can and cannot pay for? That lives on MapleCub’s methodology page. Questions about the data itself: hello@maplecub.ca.