Neutral ADR readiness platform for Canadian air passenger disputes
Portfolio Project 3 | Multi-Agent + Transparent Condition Framework | April 2026
On April 8, 2026, Air Canada announced a limited pilot of an Alternative Dispute Resolution (ADR) process — the first of its kind in Canadian aviation. 500 passengers were randomly selected and invited to transfer their CTA claims to CADR, an independent arbitrator, for a decision within 90 days.
The problem: neither passengers nor airlines have a tool to help them understand what a well-prepared ADR submission looks like, what the APPR evidence standards require, or whether their claim is strong enough to take through the ADR process rather than waiting in the 96,000-case CTA backlog (current average wait: 987 days).
ClaimReady AI fills that gap — as a neutral platform that serves both sides.
-
APPR Eligibility Check — deterministic rule engine based on SOR-2019-150. Checks all four eligibility conditions. No LLM, no ML — just the regulation.
-
Evidence Gap Assessment (Agent 1) — reviews what documents the passenger has against what CADR requires. Flags critical gaps. Blocks submission drafting until gaps are resolved.
-
Claim Readiness Assessment — transparent condition checker that evaluates each of the four APPR eligibility factors, carrier size, and documentation completeness individually. Shows each factor as passed, failed, or a warning — with the specific regulation cited and plain-language context explaining why it matters. No probability scores. No win predictions. Every factor is auditable and tied to SOR-2019-150.
-
ADR Submission Drafter (Agent 3) — drafts the passenger's CADR submission in plain language with APPR citations. Locked until evidence is complete.
-
ADR vs CTA Decision Guide (Agent 4) — personalised comparison of the two resolution paths based on claim strength and the passenger's timeline preference.
User input
│
▼
APPR Rule Engine (deterministic — SOR-2019-150)
│ ineligible → stop with explanation
│ eligible ↓
▼
Agent 1: Evidence Gap Assessor (Claude API)
│ critical gaps → human gate
│ confirmed ↓
▼
Claim Readiness Assessment (transparent condition checker)
│ 6 factors — each mapped to specific APPR condition
│ readiness: Strong / Moderate / Developing (no percentages)
▼
Agent 3: ADR Submission Drafter (Claude API)
│ locked until Agent 1 complete
▼
Agent 4: ADR vs CTA Decision Guide (Claude API)
│ personalised to readiness level + timeline
▼
Plain-language output with regulatory citations and disclaimer
Design principle: The claim readiness assessment is deliberately transparent and rule-based — not an ML black box. Every factor shown to the user maps directly to a specific APPR condition or evidence requirement. The product never predicts whether a passenger will win. Outcomes at CADR are determined by an independent arbitrator, and any tool that implies otherwise is misleading.
All evidence and regulatory logic sourced from publicly available information:
| Source | Used for |
|---|---|
| Air Passenger Protection Regulations, SOR-2019-150 | Rule engine — eligibility and compensation schedule. Claim readiness framework — each factor mapped to a specific section. |
| Air Canada ADR pilot press release (April 8, 2026) | Product trigger and ADR mechanics |
| Globe and Mail / BNN Bloomberg / CBC News (public) | Market context and statistics |
| CBC News, August 2024 | 50% passenger win rate cited in ADR vs CTA decision guide |
No proprietary data. No employer information.
An earlier version of ClaimReady AI used an XGBoost classifier trained on synthetic data calibrated to APPR rules. It was replaced with the transparent condition framework for one reason: a passenger who sees "86%–100% success probability" and then loses has a worse experience than one who sees "your claim meets all four APPR conditions and your documentation is complete." The second framing is more honest, more explainable, and more useful. The outcome at CADR depends on an independent arbitrator — no model can predict it reliably.
| Arrival delay | Large airline | Small airline |
|---|---|---|
| 3h to <6h | $400 CAD | $125 CAD |
| 6h to <9h | $700 CAD | $250 CAD |
| 9h or more | $1,000 CAD | $500 CAD |
| Arrival delay | Any airline size |
|---|---|
| Less than 6h | $900 CAD |
| 6h to <9h | $1,800 CAD |
| 9h or more | $2,400 CAD |
- Node.js 18+
- Python 3.9+
- An Anthropic API key (console.anthropic.com)
# Clone and enter the project
git clone https://github.com/mkumar84/claimready-ai
cd claimready-ai
# Copy env file and add your API key
cp .env.example .env
# Edit .env — add your ANTHROPIC_API_KEY
# Install Node dependencies
npm install
# Run tests — rule engine must pass all 38 before proceeding
npm test
# Start the CLI
npm start| Failure mode | Design response |
|---|---|
| Tool implies a passenger will win | Replaced ML probability scores entirely with a transparent condition checker. No percentage, no verdict, no prediction. The disclaimer is on every screen. |
| User treats output as legal advice | Persistent "not legal advice" framing on every step and embedded in the submission draft body |
| Ambiguous disruption cause | The cause factor card shows "most commonly disputed factor" context and explains how to gather supporting evidence — not a hidden calculation |
| Submission attempted without complete evidence | Agent 3 locked until Agent 1 reports no critical gaps — throws an error, not a warning |
| APPR regulations change | Rule engine and condition framework are both configuration-driven — regulatory updates require a config change, not a code rebuild |
| Passenger confused by readiness label | Each factor is individually explained with plain-language context and a specific regulation citation — the label is a summary, not a black box |
ClaimReady AI helps you prepare. It does not replace a lawyer.
For free legal help with air passenger claims in Canada:
- Community Legal Education Ontario (CLEO): cleo.on.ca
- Pro Bono Ontario: probonoontario.org
- CTA self-help: otc-cta.gc.ca
Built by Mahesh Kumar | AI Product Leader | April 2026
Portfolio: github.com/mkumar84