9 - Document AI Validation in DRS

# Epic 9 — Document AI Validation in DRS (MVP)

## Problem (WHY)
Registry analysts are cautious about relying on uploaded documents in the DRS. Without a trusted validation process, they worry about accepting altered or incomplete records. This undermines confidence in moving away from paper and slows adoption of fully digital workflows.

## Epic Description (WHAT — MVP)
Introduce **Document AI** into the DRS to provide analysts with automated checks on uploaded documents.
- **Value (who benefits & how):**
- **Registry Analysts:** Gain confidence that uploaded documents have been scanned for authenticity and tampering before review.
- **Operations:** Build a consistent, defensible process for accepting online documents.
- **Filing Parties (Qualified Suppliers/clients):** Benefit from faster, trusted processing without needing to mail physical documents.
- **Outcomes:**
- Each uploaded document has a **validity indicator** in DRS (e.g., “Validated / Flagged for Review”).
- Analysts know documents have been pre-checked and can proceed with greater assurance.
- Builds institutional trust in digital-only processes by embedding validation into the workflow.

## Goals (outcomes)
- Analysts can see clear validation results on every uploaded document.
- Confidence in digital-only processes increases, reducing reliance on mailed-in documents.
- Establish a trusted, repeatable method for verifying authenticity.

## Product Personas (in scope)
- **Registry Analyst** — primary consumer of AI validation flags.
- **Operations** — benefit from reduced reliance on paper and disputes.
- **Filer (Qualified Supplier)** — indirectly benefits from faster, more reliable turnaround.

## Context
- **Where:** Document Registry Service (DRS), at the point of document upload and when analysts review supporting documents.
- **Purpose:** Build analyst confidence by showing that uploaded documents are automatically validated for authenticity.
- **Flow:** Filers upload → Document AI validates → Analysts see validation status in DRS → Analysts proceed with more confidence.

## Design Reference Link (Design-owned)
_TBD — Design to paste Figma/overview link before Dev estimate._

---

## MoSCoW — MVP scope clarity
- **Must (MVP):**
- AI performs automated checks on uploaded documents and flags suspected tampering.
- Display a simple **validity flag/score** to analysts in DRS.
- Preserve all original documents in DRS (unaltered).
- **Should (near-term, if capacity allows):**
- Provide a **visual highlight/heatmap** of suspected tampering regions.
- Enable export of AI findings into audit logs.
- **Could (nice-to-have, safe to defer):**
- Analyst feedback loop to “confirm/override” AI results for model improvement.
- Confidence scoring explained (why the AI flagged the document).
- **Won’t (out of scope for this epic/MVP):**
- Automatic rejection of documents without analyst review.
- Full integration with external verification services (future candidate).

---

## Roles & Triggers
- **Roles:** Registry Analysts (primary), Operations (secondary).
- **Triggers:** Fires whenever a document is uploaded to DRS (filing request, supporting docs).

---

## Risks & Notes
- **False positives/negatives** could create analyst mistrust; mitigate with clear messaging that AI is advisory.
- **FOIPPA & privacy** — no external sharing of documents outside approved environments.
- **Change management** — analysts will need orientation to interpret and trust AI results.

---

## Dependencies & Systems
- **Dependencies:**
- DRS (Document Registry Service) must store both original and AI-analyzed results.
- Integration with chosen **Document AI service** (vendor selection TBD).
- **Auth/audit/compliance guardrails:**
- FOIPPA compliance — staff-only view, no public exposure of AI analysis.
- Minimum audit record: who uploaded, when, AI score/flag stored.

## Dependencies — **BCROS products** (cross-product)
- **DRS** — inbound integration, blocking.
- **MHR/PPR** — no direct dependency at this stage; documents still flow via DRS.

---

## Compliance & Accessibility (guardrails)
- **FOIPPA:** All documents remain in province-compliant storage; no AI processing outside approved secure environments.
- **Accessibility:** Results must be text-based (not colour-only), keyboard reachable, focus visible, announced to AT.
- **Auditability:** Record AI analysis outcome (flag/score, timestamp) alongside user ID who uploaded.

---

## Post-MVP Candidates
- Analyst feedback loop → Epic 10
- Integration with third-party document authenticity registries → Epic 11

- **Epic Parking Lot (revisit later):**
- Proactive watermarking or blockchain-backed verification.
- Automated notification to filer if tampering is detected.

- **Epic Exclusions (explicitly not doing):**
- Removing analyst discretion — AI is advisory only.
- Full automation of document acceptance/rejection.

---

## Linked MVP Stories (IDs and titles)
- **9.1 — Document AI runs on every uploaded document (spike + integration)**
- **9.2 — Analyst sees AI tampering flag/score in DRS**
- **9.3 — Preserve AI output in audit logs (staff-only)**