Data Sources & Methodology — Canonical Stat Reference

This is the ONE SOURCE OF TRUTH for every quantitative claim on the site. All research pages and guides must cite this document for any number listed below. Where a page states a figure, it should carry a (see [Data Sources & Methodology](/docs/DATA_SOURCES_METHODOLOGY/)) note and the VERIFIED / INFERRED label defined here.

Established: 2026-10-07 Last verified against official sources: 2026-10-07 Maintainer: 3mpwrApp Research

1. Why this document exists

The site accumulated several mutually contradictory figures (two WSIAT decision totals, two WSIAT win rates, two HRTO totals, two HRTO rates, and a family of large “total records” numbers that do not reconcile). This document records the canonical value for each, the source it was verified against, the date of verification, and whether it is VERIFIED (traced to an official source, dated) or INFERRED (derived by 3mpwrApp from a source that does not itself publish the figure — e.g. keyword-inferred outcomes, or an aggregate we computed).

Labelling rule (apply on every page):

• VERIFIED — the number traces directly to an official publisher (WSIAT Open Data, CanLII, Tribunals Ontario, Statistics Canada) and is dated.
• INFERRED — the number is computed or keyword-inferred by 3mpwrApp; it is an estimate, not an official statistic, and must be marked as such. Never present an INFERRED number as an official rate.

2. Official sources checked (2026-10-07)

Method note: The WSIAT CSV was downloaded and parsed with a CSV reader that handles embedded newlines in decision summaries. The live file (2026-10-07) contains 96,025 data rows (96,024 distinct DecNum; one duplicate decision number is present), spanning 1986-01-14 to 2026-06-30.

3. Reconciliation table (canonical)

4. Detailed methodology & provenance per stat

4.1 WSIAT decision count — 98,992 vs 99,036 vs 96,025

• Canonical (committed dataset): 98,992. This is the count 3mpwrApp parsed directly from the official WSIAT Open Data CSV on 2026-04-29. Every committed analytics artifact in the repo (data/wsib-comprehensive/data-inventory.json, wsiat-content-updates.json, etc.) is built on 98,992. It is therefore the internally consistent figure for the site’s existing analysis.
• Live official file (2026-10-07): 96,025 decision rows (96,024 distinct DecNum). WSIAT updates this file quarterly and has revised it since April 2026 (the row count dropped by ~2,967). Because the official source is authoritative and changes over time, the live figure is the one to re-extract toward; until a re-extraction is done, cite 98,992 with an explicit “as parsed 2026-04-29” date and note the live file reads 96,025.
• 99,036 is RETIRED. It appears across many pages as “99,036 WSIAT decisions (1987-2026 / 2016-2025)” but does not correspond to a single verified official-source parse and disagrees with both 98,992 and the live 96,025. It must be replaced by 98,992 (or the re-extracted live count) everywhere.

4.2 WSIAT win rate — 63.1% (n=651) vs 73.5% (n=649)

Neither is an official WSIAT-published statistic; both are keyword/ML-inferred by 3mpwrApp from CanLII onwsiat decisions.

• Canonical: 73.5% — from the 2020-2026 CanLII classified subset: 438 granted / 596 decisive outcomes, drawn from 649 classified decisions. This is the larger, more recent, and currently-used classification.
• Superseded: 63.1% — from an earlier 2,000-decision sample in which 651 outcomes were keyword-detected (411 worker victories). Retain only as a labelled historical estimate; do not present it as the current rate.

Both must be labelled INFERRED and never presented as “WSIAT’s official success rate.”

4.3 HRTO total — 9,269 vs 62,093

These are two different scopes, not a contradiction:

• 9,269 = HRTO decisions extracted from CanLII (onhrt) for 2020-2026, individually outcome-classified, and committed as the site’s HRTO dataset (totalDecisions: 9269). This is the figure to use when the site says “HRTO cases analyzed.” VERIFIED against CanLII.
• 62,093 = HRTO applications aggregated from 39 quarterly XLSX statistical reports (2016-2025). These are summary counts, not individual case records. INFERRED from Tribunals Ontario quarterly reports. Use only when explicitly describing “applications 2016-2025.”

4.4 HRTO rates — 43.9% vs 5.4%, and the retracted 0.7%

Computed from the committed 9,269 HRTO dataset (2020-2026):

• Abandonment: 43.9% (4,073 / 9,269) — VERIFIED.
• Allowed / discrimination-found: 5.4% (503 / 9,269) — VERIFIED.

The 0.7% “applicant victory” figure that appeared on the ONCA cross-tribunal comparison and a visualization is RETRACTED. It was computed from a “detected-outcome subset” that does not exist in the committed repository and was audit-corrected (see the 2026-07-20 accuracy audit note in _posts/2026-04-20-hrto-email-crisis-abandonment-epidemic.md). Remove 0.7% wherever it appears (e.g. data/visualizations/cross-tribunal-comparison.json, connecting-the-dots-canlii-keyword-visualization-network.html, _posts/2026-05-08-onca-precedent-overview.md). Use 5.4% as the HRTO allowed rate.

4.5 The large “total records” family — 230,392 / 127,600 / 50,161 / 238,767 / 34,960 / 137,252

• 230,392 — headline grand total of the 3mpwrApp dataset. Stated composition: 127,600 tribunal decisions + 102,792 employer/other records (raw, pre-dedupe). VERIFIED as the site’s stated composition.
  • Caveat / internal inconsistency: a second, contradictory decomposition also appears — “99,036 WSIAT + 130,736 employer records” (e.g. research-data-sources.md L427) — which sums to 229,772, not 230,392. The 130,736 figure = NEER 91,814 + CAD-7 38,922 employer records; the 102,792 figure is the number used in the 230,392 decomposition. Retire the “99,036 + 130,736” line; keep 230,392 = 127,600 + 102,792.
• 127,600 — Ontario tribunal decisions = WSIAT 99,036 + HRTO 9,269 + ONSBT 13,798 + ONWSIB 463 + ONCA 5,034. INFERRED composition; note it uses the retired 99,036 component, so if WSIAT is updated to 98,992 the sum becomes 127,556.
• 50,161 — v3.0 outcome-classified Ontario cases (2026-05-15), 85.1% classified. VERIFIED via ontario-classification-stats-v2.json / -v3.json (totalProcessed: 50161).
• 238,767 — NOT FOUND anywhere in the current repository. Provenance unverified; do not cite. (Listed in the reconciliation request but absent from all source files.)
• 34,960 — 2020-2026 Ontario tribunal decisions used in the keyword-network visualization subset (connecting-the-dots-canlii-keyword-visualization-network.html). INFERRED (period-limited viz subset).
• 137,252 — AI outcome-predicted decisions in the legacy research backup (research-old-backup.html). INFERRED (historical model-predicted subset).

5. Historical context preserved (do not delete)

This document adds methodology; it does not erase prior numbers. Where a page currently shows a retired/superseded figure (99,036, 63.1%, 0.7%, 130,736), keep the sentence but append the reconciliation pointer and the correct canonical value, e.g.:

“We analyzed 99,036 WSIAT decisions (historical extraction). The current canonical count is 98,992 (WSIAT Open Data CSV, parsed 2026-04-29; live file 2026-10-07 = 96,025) — see Data Sources & Methodology. All win rates are INFERRED (keyword-inferred), not official WSIAT rates.”

6. Citation targets — pages that must cite this document

Each entry below names a file, the exact line(s) where a contradictory/headline figure appears, and the action for the coordinator (add the (see [Data Sources & Methodology](/docs/DATA_SOURCES_METHODOLOGY/)) note and replace the figure with the canonical value where indicated). This list was generated 2026-10-07; it covers the research pages and all guides (the in-scope set). Additional contradictory files exist outside this scope (data/knowledge-base/*.md, _posts/*.md, several HTML visualizations, docs/WSIB-APPEAL-GAP-ANALYSIS.md) and are listed separately in §6.2.

6.1 In-scope research pages & guides

research.html

• L377 — headline <h1>We Analyzed 230,392 Records…</h1> → add note; 230,392 VERIFIED composition (§3/§4.5).
• L380 — “Ontario v3.0 complete May 15, 2026: 50,161 cases, 85.1% classified” → add note; 50,161 VERIFIED.
• L398 — decomposition paragraph mixing 230,392 / 127,600 / 99,036 / 98,992 / 62,093 / 9,269 → add note; replace 99,036 with 98,992; keep 62,093 only as “applications 2016-2025 (INFERRED)”.
• L410 — “built from 99,036 real decisions” → replace with 98,992.
• L451 / L457 — “230,392 records” / “127,600 Ontario tribunal decisions + 102,792 employer records” → add note.
• L471 / L475 — stat-number blocks 230,392 / 127,600 → add note.

• L493 — list “WSIAT: 99,036

  HRTO: 9,269

  …” → replace 99,036 with 98,992.

• L536 — badge “99,036 Decisions” → replace with 98,992.
• L544 — “9,269 decisions” (HRTO) → add note (9,269 VERIFIED classified 2020-2026).

research-data-sources.md

• L49 — “99,036 decisions (2016-2025)” → replace with 98,992 (VERIFIED, parsed 2026-04-29).
• L115 — “62,093 applications (aggregate counts, 2016-2025)” → add INFERRED label + note (scope = applications, not cases).
• L356 — “All 230,392 records…” → add note.
• L366 / L369 / L373 — Cross-Tribunal total “230,392”, “WSIAT 99,036”, “HRTO 62,093” → reconcile to 98,992 / 62,093(INFERRED).
• L389 — “WSIAT: 68.7% success rate (99,036 decisions)” → this 68.7% is an additional unsupported rate; replace with canonical 73.5% INFERRED and 98,992.
• L424 / L427 — “230,392 records” / “99,036 WSIAT + 130,736 employer records” → retire the 99,036+130,736 line; keep 230,392 = 127,600 + 102,792.
• L534 — “99,036 decisions, 100% coverage” → replace with 98,992.
• L620 — “230,392 records from WSIAT, HRTO, ONSBT, and WSIB” → add note.

data-limitations.md

• L24 / L29 / L39 / L67 / L71 / L143 / L192 / L327 / L400 / L447 / L453 / L456 — every “99,036 WSIAT decisions” and “230,392 records” → replace 99,036 with 98,992 and add note + VERIFIED/INFERRED labels.

guides/index.md

• L74 / L230 / L322 / L395 / L416 / L453 / L476 — “99,036 WSIAT decisions” → replace with 98,992; add note.

guides/hrto-complete-guide.md

• L18 — “Allowed: 503 (5.4%)” → add VERIFIED note.
• L21 — “Abandoned: 4,073 (43.9%)” → add VERIFIED note.
• L27 — narrative referencing 43.9% → add note.
• L580 — “HRTO quarterly statistics (2016-2025) - 62,093 applications” → add INFERRED label + note (scope = applications).

guides/onca-appellate-guide.md

• No contradictory figure currently present in this file (the retracted 0.7% lives in data/HTML/post artifacts, see §6.2). No change required, but add a general “(see Data Sources & Methodology)” note near its data-source statements for consistency.

guides/wsiat-complete-guide.md

• L3 (description), L12, L18, L58, L84, L301, L395, L452, L629, L661 — every “99,036 WSIAT decisions” → replace with 98,992 and add note; L58 references “85-95% yearly win rates” → label INFERRED and point to canonical 73.5%.

guides/wsiat-back-injury-guide.md — L16, L29, L162, L259, L296 (“99,036”) → replace with 98,992 + note. guides/wsiat-chronic-pain-guide.md — L10, L330 → replace with 98,992 + note. guides/wsiat-nel-benefits-guide.md — L274 → replace with 98,992 + note. guides/wsiat-nel-chronic-pain-strategy.md — L137, L315, L352 → replace with 98,992 + note. guides/wsiat-loe-benefits-guide.md — L344, L362 (the L362 “Verified — computed from the WSIAT Open Data export of 99,036 decisions” should read 98,992). guides/healthcare-wsiat-industry-guide.md — L24, L330 → replace with 98,992 + note. guides/construction-wsiat-industry-guide.md — L25, L353 → replace with 98,992 + note. guides/manufacturing-wsiat-industry-guide.md — L24, L352 → replace with 98,992 + note. guides/onwsib-skip-strategy-guide.md — L99, L335 (“99,036 decisions 2020-2026” comparative) → replace with 98,992 + note.

6.2 Out-of-scope but also contradictory (flagged for a follow-up pass)

These were found to carry the same figures but are outside the research/guides scope of this phase. They should be reconciled in a subsequent pass:

• data/knowledge-base/*.md (≈20 files) — repeatedly cite “99,036 ONWSIAT/WSIAT decisions (2020-2026)”. Replace with 98,992 + note.
• _posts/2026-04-15-wsib-exposed-*.md, _posts/2026-04-25-feature-spotlight-*.md, _posts/2026-04-30-suppression-gap.md, _posts/2026-04-16-wsib-transparency-gap-*.md — carry 99,036, 73.5%, 63.1%, 0.7%/5.4% figures. Reconcile to 98,992 / 73.5% INFERRED / 5.4% VERIFIED / 0.7% retracted.
• _posts/2026-05-08-onca-precedent-overview.md — HRTO row still shows “0.7% applicant victory”; replace with 5.4%.
• connecting-the-dots-canlii-keyword-visualization-network.html — uses 98,992 (OK), 9,269 (OK), and a retracted “0.7% of detected outcomes” (L792) and “73.5% abandonment” (L787) for HRTO; replace 0.7% with 5.4% and label 73.5% abandonment as the 43.9% committed figure or remove.
• cross-tribunal-success-rates.html — uses 98,992 (OK) and “62,093” (L182, INFERRED applications); add labels.
• data/visualizations/cross-tribunal-comparison.json (L49) — “HRTO: 0.7% applicant victory / 73.5% abandonment” → replace 0.7% with 5.4%.
• docs/WSIB-APPEAL-GAP-ANALYSIS.md — L188 “62,093 HRTO decisions analyzed (2016-2026)” (mislabels applications as decisions) and L102/L109 “73,640 / 72,466 allowed (30.7%)” WSIB figures (verify separately against wsib.ca).

7. Re-extraction action item

Because the live WSIAT Open Data CSV (2026-10-07) = 96,025 decisions, the committed 98,992 dataset is stale. Before the next data refresh:

1. Re-download https://www.wsiat.ca/opendata/WSIATDecisions.csv.
2. Re-parse and update data/wsib-comprehensive/data-inventory.json and dependent artifacts to the new count (currently 96,025).
3. Propagate the new canonical WSIAT count to all pages via this reference.

Until then, the canonical WSIAT decision count remains 98,992 (as parsed 2026-04-29 from the official WSIAT Open Data CSV), with the live 96,025 noted as a freshness caveat.

Document established 2026-10-07. Verify against official sources before each publish cycle. All INFERRED figures are estimates and must never be presented as official tribunal statistics.