How Eunosaactually works.

Section 01 · The Research Corpus

697 peer-reviewed studies, curated, not scraped.

Every cited answer Eunosa returns is grounded in this corpus. The evidence base is curated — not scraped — and weighted toward Canadian regulatory, cultural, and healthcare contexts. When the literature is sparse on a question, Eunosa flags the gap rather than inventing a study to fill it.

Canada-first, then comparable jurisdictions.

Studies are weighted by jurisdictional fit on a three-tier hierarchy so the corpus reflects the legal, cultural, and healthcare realities of the Canadian workplace.

1. Tier 1
   Canada-first

   Studies grounded in Canadian regulatory, cultural, and healthcare contexts
2. Tier 2
   Comparable jurisdictions

   Australia, United Kingdom, New Zealand, Nordics, Netherlands
3. Tier 3
   Other international

   United States and other international evidence included only when findings are demonstrably generalizable across systems

Inclusion criteria.

• Peer-reviewed publication
• English-language
• Methodological quality threshold (cohort, RCT, systematic review, meta-analysis prioritized)
• Direct relevance to RTW, accommodation, or workplace mental health
• Canadian or comparable-jurisdiction applicability flagged at intake
• Foundational classics admitted on case-by-case review

Reference institutions.

Sources are drawn from the institutions and standards bodies that govern Canadian workplace research and disability practice.

• Institute for Work & Health (IWH, Toronto)
• Canadian Centre for Occupational Health and Safety (CCOHS)
• WSIB Ontario · WorkSafeBC · CNESST
• NIDMAR (National Institute of Disability Management and Research)
• Mental Health Commission of Canada (MHCC)
• CSA Z1003 / ISO 45003 standards bodies
• APA PsycNET · PubMed · Web of Science
• Cochrane Collaboration

Featured studies.

A representative sample of published studies in the V1.0 launch corpus. Each is peer-reviewed and graded for evidence strength downstream by /ask retrieval.

• Mustard 2024
  Supervisor RTW practices and case duration in healthcare cohort

  Mustard et al. · Journal of Occupational Rehabilitation · strong
• Cancelliere 2016
  Systematic review of factors affecting RTW after injury and illness

  Cancelliere et al. · Chiropractic & Manual Therapies · strong
• van 2015
  Workplace interventions for preventing work disability

  van Vilsteren et al. · Cochrane Database of Systematic Reviews · strong
• Cullen 2018
  Effectiveness of workplace interventions across musculoskeletal, pain, and mental health conditions

  Cullen et al. · Journal of Occupational Rehabilitation · strong
• Steenstra 2017
  Prognostic factors for duration of sick leave with low back pain

  Steenstra et al. · Journal of Occupational Rehabilitation · strong
• Nieuwenhuijsen 2014
  Interventions to improve RTW for workers on sick leave with depression

  Nieuwenhuijsen et al. · Cochrane Database of Systematic Reviews · strong

Section 02 · The Canadian Legal Corpus

Research tells you what works. Law tells you what's required.

Eunosa pairs the curated research corpus with Canadian legal authority — case law and legislation, cited inline by formal legal citation. The combination is what makes a consultant's recommendation defensible: the research evidence speaks to outcomes, the legal authority speaks to obligations. Eunosa is the only Canadian AI research tool that combines both in a single cited answer.

What's covered today

Live now via the Access to Algorithmic Justice (A2AJ) open Canadian legal data project — a research initiative of York University's Osgoode Hall Law School and Toronto Metropolitan University's Lincoln Alexander School of Law.

Expanding to every provincial tribunal

Eunosa is building per-tribunal ingestion for every Canadian provincial human-rights tribunal and workers'-compensation appeal tribunal. Each is added directly from its source site under the province's open-government licence.

1. Phase A

   WSIAT (Workplace Safety and Insurance Appeals Tribunal, ON)

   Single biggest Ontario gap for RTW consultants — workers’ compensation jurisprudence

2. Phase B

   HRTO (Human Rights Tribunal of Ontario, trial-level)

   Most accommodation-relevant tribunal in Canada’s largest province

3. Phase C

   BC Human Rights Tribunal + BC WCAT

   Second-largest English-Canada human-rights jurisdiction; BC workers’ comp

4. Phase D

   Alberta Human Rights Commission + AB WCB Appeals

   Third-largest jurisdiction

5. Phase E

   Saskatchewan, Manitoba, Atlantic provinces (NS, NB, PE, NL)

   Completes English-Canada coverage

6. Phase F

   Quebec (Tribunal administratif du travail + Tribunal des droits de la personne via SOQUIJ)

   Largest distinct legal system in Canada

7. Phase G

   Provincial legislation for AB, SK, MB, QC, NS, NB, PE, NL

   Statutes and regulations for provinces not yet covered

8. Phase H

   Yukon, Northwest Territories, Nunavut

   Territorial coverage completes the national map

How we get the data

Eunosa scrapes source tribunal websites and provincial legislative portals directly — not CanLII. Each source’s robots.txt has been verified to permit scraping. Citations always link back to the official source.

• Per-tribunal ingestion modules respect each source site’s rate limits and robots.txt
• Decisions are stored verbatim with provenance metadata (source URL, scrape timestamp, upstream license string)
• Re-ingestion uses a content hash to skip unchanged decisions and detect content drift
• A separate audit log records every scraper run with status, counts, and any errors

Legal basis for commercial reproduction

What the legal layer is not

• Decision-support, never adjudication. Eunosa explains, cites, and surfaces relevant authority — it never decides a case.
• Every legal-authority answer closes with a "verify with counsel or the relevant regulator" disclaimer.
• Editorial headnotes and case summaries (e.g. CanLII’s commentary) are not included — only the official decision text from the issuing tribunal.
• The corpus is expanding. When a question lands outside current coverage, Eunosa says so explicitly and points to CanLII or the tribunal’s own database — it does not fabricate a citation.

Section 02 · The Risk Model

A rule-based engine that explains itself.

The RTW Complexity Risk Model is a knowledge-engineered, rule-based scoring system. It produces a continuously updated case-complexity signal that recalculates whenever new data enters the platform — and whenever time passes without required actions. Every score change traces back to a specific rule, factor, and source.

Built on a curated subset of the corpus.

The Risk Model draws from a focused subset of 48 catalogued primary sources within the broader corpus. Each is paired with structured evidence extracts and mapped to factors and rules. Both stores share the same audit trail.

Five domains.

Risk is organized into five domains representing the major categories of influence on RTW complexity. Each domain contains four factors. Definitions, measurement methods, and the evidence backing for each factor live in the gated methodology document.

• Culture / Team Climate

  Psychological safety, belonging, peer support, and stigma around disclosure determine whether workers feel safe engaging with the RTW process authentically.

• Supervisor Capability

  Supervisor behaviour is the single most modifiable workplace predictor of RTW outcomes — training, confidence, communication quality, and speed of contact.

• RTW Process Quality

  The timeliness, coordination, and quality of formal RTW processes — planning, accommodation, follow-up — directly predict duration and sustainability of return.

• Work Design / Context

  Job demands, control, role clarity, and the availability of modified duties shape whether the workplace itself supports or undermines recovery.

• Compliance / System Maturity

  Organizational readiness — policy infrastructure, standards implementation, reporting culture — acts as a multiplier on individual case risk.

Three role-based views.

Output is shaped by who is reading it. Privacy is enforced at the field level — supervisors do not see clinical detail; executives do not see individual case data.

Section 03 · Boundaries & standards

What the model does not do.

Scope discipline is part of defensibility. Eunosa is decision support — and only decision support. It is human-in-the-loop by design, intended for qualified professionals to interpret in context.

Aligned with Canadian standards.

The model maps to two CSA standards. Compliance maturity acts as an organizational multiplier — strong systems dampen case risk; weak systems amplify it. The specific mapping logic is documented in the gated methodology document.

Citation Resolution Guarantee · Canadian data residency · No training on your data · PHIPA-aware · 90-day query retention

Section 04 · Governance & version

Versioned, peer-reviewable, honest about scope.

Version lifecycle.

Every model version moves through a documented lifecycle. Released versions are immutable; new evidence and pilot feedback flow into the next draft.

What V1.0 does not yet do.

Six limitations are documented up front so consultants can speak to them when their work is reviewed.

• No empirical weighting in V1.0

  V1 uses expert-assigned rule logic rather than empirically derived weights. Calibration against real outcomes data is planned for V1.1.

• Canadian-centric

  The model is designed for Canadian workplaces. Compliance mapping is specific to CSA standards. Adaptation for other jurisdictions requires re-mapping.

• No diagnostic input

  The model deliberately excludes medical diagnosis as an input, focusing on workplace and process factors. This is a feature, not a limitation, but users must understand its scope.

• Survey-dependent factors

  Culture and team climate factors require organizational survey data. Without it, those factors fall back to default assumptions.

• Single-condition focus

  V1 does not model comorbidity or multi-condition complexity. Future versions may add this.

• Decision-support only

  Every output carries a decision-support disclaimer. The model is human-in-the-loop by design — not a replacement for professional judgment.