Environmental Impact Assessment Intelligence

AI-powered compliance gap detection for Canadian federal and provincial EIA review — legal flags, regulatory flags, dual-model adversarial validation, and full provenance chain on every finding.

ParadigmForge.AI

⚖️ IAA 2019 · CEAA 2012 🌲 BC EAA · EPEA · Ontario EA · REEIE 📎 Audit-Grade · Full Provenance

🚨 EIA Review Is Broken — Dangerously So

Every major infrastructure project in Canada requires an environmental impact assessment. Right now the review of those documents is slow, inconsistent, legally exposed — and getting worse.

🔊 VO-01
6–18
months to review a major EIA
500+
pages in a typical EIA submission
$4M+
average cost when a JRP review fails
30+
section types reviewed manually today
⏱️
Manual Review Bottleneck
Experienced EA reviewers spend 70% of their time on rote compliance checking — finding what's missing — rather than applying judgment to complex assessment decisions. That's not where their expertise adds value.
⚖️
Legal Exposure Is Real
Missing a SARA trigger, an inadequate Haida consultation test, a Fisheries Act s.35 gap — these aren't oversights. They are grounds for judicial review that stop billion-dollar projects cold.
🔄
Inconsistent Outcomes
Different reviewers flag different things. Institutional knowledge walks out the door when staff turn over. Every new proponent reinvents the wheel on what the reviewing body actually expects.
📈
Regulatory Velocity Problem
IAA 2019 timelines. UNDRIP obligations. BC EAA 2018 changes. The regulatory landscape is accelerating faster than any team can manually track. Human-only review cannot scale.

⚙️ Seven Steps: Upload to Decision-Ready

EIA Pro handles the full assessment lifecycle — from raw PDF to a structured review package with flagged compliance gaps, recommended conditions, and a complete audit trail.

🔊VO-02
01
Document Upload
PDF or DOCX accepted up to 500MB. UUID generated, SHA-256 integrity hash computed, metadata stored in EIA Pro's private Neo4j instance.
POST /api/documents/upload · Status: UPLOADED
02
Intelligent Document Parsing
Heading detection and regex-based segmentation classifies every section into one of 30+ recognized types across three categories: Baseline Environment, Impact Assessment Phases, and Regulatory Elements. Each section chunked to 2,048-token segments with 256-token overlap, embedded, and stored in Qdrant.
nomic-embed-text 768d · EIA Pro Qdrant (port 6350)
03
Entity Resolution
Extracts species, chemicals, waterbodies, and Indigenous nations from document text. Resolves each entity against shared canonical registries: SARA Schedule 1, CEPA Schedule 1, DFO waterbody registry. No entity is taken at face value.
Shared Neo4j (port 7691) · Read-only
04
AI Flag Generation
For every section: shared Qdrant semantic retrieval surfaces applicable regulations even when the EIA doesn't cite them directly. The generator model identifies compliance gaps as structured JSON flags with confidence scores and evidence chains.
TV-GRAG architecture · Shared regulatory intelligence (port 6338)
05
Adversarial Consensus Validation
Every CRITICAL and HIGH flag is independently evaluated by a second adversarial model whose job is specifically to challenge and reject weak findings. It sees only the flag and evidence — not the generator's confidence score. CONFIRMED flags proceed; REJECTED flags are discarded.
Dual-model independence · No shared state · 0.90 legal / 0.85 regulatory thresholds
06
Human Review & Triage
Reviewer works flag-by-flag: CONFIRMED / DISMISSED / DEFERRED / NEEDS_INVESTIGATION / ESCALATED. Every decision requires written rationale. Timestamped user attribution on every action.
POST /api/flags/{flag_id}/triage · Full audit trail
07
Summary & Institutional Learning
Structured review output: executive summary, critical issues, risk matrix, recommended conditions from the template library, and information requests. Every completed review calibrates future confidence thresholds and captures effective condition language.
TV-GRAG consensus engine · Template capture · Continuous improvement

📄 Reads Every Section. Classifies It Correctly.

Before any compliance checking can happen, EIA Pro must understand what it's reading. Automatic section classification into 30+ types ensures the right regulations are applied to the right content — even when proponents don't follow standard naming conventions.

🔊VO-03
Baseline Environment
Atmospheric Acoustic Aquatic Terrestrial Wildlife Species at Risk Land Use Socioeconomic Heritage / Cultural Human Health Navigation
Impact Assessment
Construction Phase Operations Phase Decommissioning Accidents & Malfunctions
Regulatory Elements
Executive Summary Project Description Alternatives Assessment Regulatory Framework Public Consultation Indigenous Consultation Government Consultation Cumulative Effects Mitigation Measures Residual Effects Significance Determination Monitoring Plan Follow-up Program
📝 EIA Pro Owns — Read / Write
Document graph + flag graph (Neo4j port 7694)
EIA document chunks + flag descriptions (Qdrant port 6350)
Review workflows + reviewer annotations
Condition templates + institutional learning data
🔒 Shared EnviroPro Intelligence — Read Only
Federal / provincial EA legislation (Neo4j port 7691)
SARA species schedules + DFO requirements
Duty-to-consult case law: Haida, Taku, Mikisew, Clyde River
Regulation + guideline embeddings (Qdrant port 6338)

🚩 Flag Example: SARA Trigger Unaddressed

An illustration of the kind of Critical legal gap EIA Pro is built to surface — a species-at-risk compliance failure in a mining project's terrestrial wildlife section, with full provenance chain from document text to regulatory source.

🔊VO-04
🚩 Flag Type: SARA Trigger Unaddressed CRITICAL
Section
Species at Risk — Terrestrial Wildlife, Construction Phase
Species
Woodland Caribou (Rangifer tarandus caribou) — Boreal population
Listed as Threatened on SARA Schedule 1. Critical habitat identified in finalized federal recovery strategy. Section 58 protection order in force on federal lands.
Gap Identified
Species presence acknowledged in document. The following SARA obligations require assessment:

Section 79 — The federal authority must identify adverse effects on the species and its critical habitat, consult the competent minister, and ensure mitigation and monitoring are consistent with the recovery strategy.

Section 58 — Critical habitat is legally protected on federal lands. The EIA must explicitly address potential destruction of identified critical habitat.

Section 73 — Where project activities risk contravening SARA prohibitions under s.32 (harming individuals) or s.58 (critical habitat), a permit analysis is required to determine whether authorization is needed.
Confidence
0.97 — Generator flagged · Adversarial validator: CONFIRMED
Severity Basis
A clear record that the responsible authority failed to apply section 79 duties in the face of a known Schedule 1 Threatened species and identified critical habitat is grounds for judicial review — failure to consider a mandatory relevant factor and failure to comply with a statutory duty.
Provenance
📎 Evidence Chain
EIA Text → Species at Risk section: "Boreal Woodland Caribou were observed during winter wildlife surveys in the northern block of the project footprint."
Entity Resolution → SARA-SCHEDULE1 · THREATENED · Boreal population — status confirmed against SARA Public Registry
Regulatory Retrieval → SARA ss.32, 58, 73, 79 · Boreal Caribou Recovery Strategy (2019) · SOR/2019-188 Critical Habitat Order
Generator → Flag generated with 0.97 confidence · Adversarial validator → CONFIRMED independently

📋 Nine Regulatory Flag Types — Confidence Threshold 0.85

Regulatory flags catch technical and methodological deficiencies — gaps that generate information requests, require supplemental studies, and add months to review timelines even when they don't rise to judicial review.

🔊VO-06
Flag Type What It Catches Severity Typical Consequence if Missed
Baseline Data GapInsufficient data for prediction Data collection period, method, or coverage insufficient to support impact predictions made in the assessment HIGH Additional studies required · 6–12 month delay
Methodology DeviationNon-standard approach Deviates from agency-accepted methodology without justification — e.g. non-CCME noise assessment, non-DFO fish passage methodology HIGH Supplemental report required
Outdated Standards ReferenceSuperseded regulation cited Document cites regulations, guidelines, or standards that have since been superseded MEDIUM Information request · Credibility concerns
Mitigation VaguenessCommitments unspecific "Best practices will be followed" type language — no measurable targets, no responsible party, no timeline HIGH Conditions imposed; proponent required to revise
Monitoring Plan GapDoesn't match predicted impacts Monitoring program doesn't address the parameters predicted to change in the assessment HIGH Monitoring plan condition imposed
Cumulative Effects DeficiencyScope too narrow Geographic scope, temporal scope, or project list misses reasonably foreseeable future projects HIGH Most common grounds for JRP rejection
Follow-up InadequacyKey uncertainties unaddressed Follow-up program doesn't verify impact predictions or address key uncertainties identified in the assessment MEDIUM IAA s.82 follow-up program conditions
Transboundary OmissionCross-jurisdictional effects missed Effects crossing provincial, territorial, or international boundaries not addressed HIGH Coordination with other jurisdictions required
Inconsistency DetectedInternal contradictions Factual or analytical contradictions between sections — species present in one section, absent in another; impact magnitudes that don't match significance ratings MEDIUM Credibility concerns · Supplemental clarification required

🧠 Dual-Model Adversarial Consensus

The architectural decision that makes EIA Pro's flags trustworthy rather than just numerous — an independent adversarial validator whose job is specifically to challenge and reject weak findings before they ever reach a reviewer.

🔊VO-07
EIA Section
2,048-token chunk
→
Regulatory Retrieval
Shared Qdrant + Neo4j
→
Generator Model
Identifies gaps · Structured JSON flags
CRITICAL / HIGH flags →
Adversarial Validator
Receives: flag + evidence only · Not generator confidence score
Job: find reasons to REJECT
→
CONFIRMED
Stored with dual provenance · Reaches reviewer
REJECTED
Discarded · Never shown to reviewer
Not a Second Opinion — An Adversarial Challenge
The validator's job is specifically to find reasons to reject a flag — to attack the legal reasoning, challenge the regulatory citation, and determine whether the evidence actually supports the finding. It starts from scratch with only the flag and the evidence chain.
Independence Is Structural
The validator never sees the generator's confidence score or reasoning path. It cannot be primed to agree. A flag that survives this challenge has been independently stress-tested against the actual regulatory record — not confirmed by a model that was primed to agree.
What 0.90 Confidence Actually Means
A 90% threshold on legal flags doesn't mean a score one model assigned itself. It means a finding that withstood an independent adversarial review grounded in the same regulatory sources. MEDIUM and LOW flags bypass the validator — only CRITICAL and HIGH bear the full cost of adversarial challenge.
Rejected Flags Are Never Seen
Flags that don't survive adversarial challenge are discarded silently. Reviewers only see findings that passed both the generator and the validator. This is what keeps false positive rates low enough that reviewers trust the system — and act on what it surfaces.

👁️ AI Surfaces the Gaps. Humans Make the Calls.

EIA Pro augments reviewer expertise — it doesn't replace it. Every flag requires a human decision with a documented rationale. The audit trail is complete, timestamped, and user-attributed.

🔊VO-08
👁️ Reviewer Triage — SARA Trigger Unaddressed CRITICAL
Flag Summary
Boreal Woodland Caribou (SARA Schedule 1, Threatened) present in project footprint. SARA ss.32, 58, 73, 79 obligations not addressed in document.
Decision
✓ CONFIRMED ✗ DISMISSED ⏸ DEFERRED ? NEEDS INVESTIGATION ↑ ESCALATED
Rationale
SARA s.79 requires the federal authority to identify adverse effects on listed wildlife species and their critical habitat and to ensure mitigation is consistent with the recovery strategy. Section 6.3 acknowledges species presence but contains no evidence of s.79 obligations being addressed, no critical habitat assessment, and no s.73 permit analysis. Flag confirmed as written. Information Request IR-047 to be issued.
Audit Entry
Confirmed by: J. Rivard (Senior EA Reviewer) · 2024-11-14T14:22:09Z · IR-047 issued
Mandatory Written Rationale
Every triage decision requires a written rationale. No checkbox-only triage. The rationale is what feeds the institutional learning layer — it's where the organization's judgment gets encoded and made available to every future reviewer.
Dismissals Are as Valuable as Confirmations
When a reviewer dismisses a flag — because the EIA addressed the issue in a section the AI didn't fully parse — that signal calibrates future confidence thresholds. The system gets smarter from every decision, not just the confirmations.
Section Annotations
Reviewers can annotate any document section, not just flagged areas. Free-text observations are stored, attributed, and carried forward into the summary generation step.
Escalation Path
Flags can be escalated to senior reviewers, legal counsel, or decision-makers. The escalation chain is tracked and auditable — critical for JRP and ministerial decision contexts where the record must be complete.

📈 Every Review Makes the Next One Smarter

EIA Pro's institutional learning layer transforms individual review decisions into organizational intelligence — reducing false positives, capturing effective condition language, and preserving the expertise that would otherwise walk out the door.

🔊VO-09
🎯 Confidence Calibration
Dismissed flags don't disappear — they calibrate. When reviewers consistently dismiss a flag type in a specific project context, that pattern adjusts the confidence threshold for future flags of the same type in the same context. False positive rates drop over time without any manual tuning.
📝 Condition Template Library
When a reviewer writes well-crafted condition language — specific, measurable, enforceable — EIA Pro captures it as a reusable template. Future reviewers are offered relevant templates during summary generation. The organization's best condition language propagates automatically.
🔍 Review Pattern Detection
Recurring deficiencies across a proponent, project type, or EA consultant are surfaced as organizational intelligence. Patterns like "this consultant consistently produces thin cumulative effects assessments" get encoded — not lost when the reviewer who noticed it moves on.
📋 Review Templates
Per-project-type, per-jurisdiction review checklists built from aggregated triage history. A hydroelectric project under BC EAA 2018 gets a different template than a pipeline under IAA 2019. Templates evolve as the institutional knowledge base grows.
Staff Turnover Resilience
When an experienced reviewer leaves, their judgment doesn't leave with them. Their confirmed flags, dismissed patterns, and condition language remain in the system — available to the next reviewer from day one. Twenty years of expertise stays in the organization.
Custom Sensitivity Rules
Organizations can define their own flag rules and severity overrides. A First Nation reviewing proponent EIAs against their own protocols. A province with jurisdiction-specific priorities. EIA Pro adapts to how your organization actually reviews.

💼 The ROI Is Measured in Months Saved

Every information request adds weeks. Every supplemental study adds months. Every missed legal gap risks judicial review — and years. EIA Pro shifts the math at every stage of review.

🔊VO-10
70%
of reviewer time on rote compliance checking
$4M+
average cost of a failed JRP review
18
flag types across legal + regulatory domains
6
EA regimes supported — federal + provincial
Without EIA ProWith EIA ProImpact
Manual section-by-section gap checking Automated classification + flagging in hours Weeks → Hours
Reviewer expertise leaves with the reviewer Institutional learning retains all judgment Resilience
SARA / DFO / navigation triggers missed under pressure Mandatory entity resolution against canonical registries Zero missed triggers
Inconsistent condition language across reviewers Condition template library with proven enforceable language Consistency
Indigenous consultation record fails Haida test at JRP Haida / Taku / Clyde River tests applied to every consultation section Legal exposure eliminated
New reviewer needs 2–3 years to become effective Institutional templates + pattern library accelerates onboarding Effective from week one
Who It's Built For
Federal and provincial EA agencies. First Nations conducting project review under their own protocols. Proponents performing internal pre-submission review. EA consultants doing quality assurance. Law firms advising on EA compliance and JRP proceedings.
Supported EA Regimes
Federal: IAA 2019, CEAA 2012. Provincial: BC EAA 2018, Ontario EA Act, Alberta EPEA, Quebec REEIE. Project types: Mining, Pipeline, LNG, Wind, Solar, Hydroelectric, Nuclear, Transportation, Industrial, Marine, Waste Management.
Canadian Data Sovereignty
All EIA documents remain on Canadian infrastructure. Organization data is isolated — no cross-organization data sharing. Regulatory intelligence is shared across the platform; your review decisions and documents are yours alone.
Audit-Grade by Design
Timestamped user attribution on every action. Mandatory written rationale on every triage decision. Full provenance chain from document text to regulatory source on every flag. Built to survive judicial review of the review process itself.
Ready to See EIA Pro on Your Documents?
Book a personalized demo with a real EIA from your review queue. See exactly what EIA Pro flags — and what it doesn't.