Veridex: Structured Claim Reliability and Moderated Discussion Platform

Veridex is a full-stack web-based platform designed to enhance the reliability assessment of public claims through structured evaluation, authenticated user participation, and robust moderator oversight. Our goal is to provide a transparent and auditable environment for public discourse.

Claim Submission

Registered users can submit public claims with attached supporting references (URLs or document uploads).

Reliability Evaluation

Deterministic, rule-based evaluation generates clear reliability indicators for each claim.

Moderated Discussion

Claim-specific, moderated discussion threads foster constructive dialogue and accountability.

Transparency& Auditability

Veridex ensures transparent processes, auditability, and neutrality in all reliability evaluations.

Project Scope: Building Trust in Public Claims

What Veridex Is Not

To maintain its neutrality and focus, Veridex explicitly defines what it will not do:

No Absolute Truth/Falsehood

The platform does not declare information as absolutely true or false, focusing instead on reliability indicators.

Not a Fact-Checking Replacement

Veridex complements, rather than replaces, professional fact-checking institutions.

No Dissent Suppression

Legitimate dissent is not suppressed without thorough moderation review and justification.

Key Project Objectives

Moderator Oversight

Reliability Evaluation

Structured Claims

Secure Authentication

Our primary objectives are centered around creating a secure, structured, and transparent environment for claims and discussions.

• Develop a secure authentication system with login and signup functionality, including role-based access control (User and Moderator roles).
• Implement a structured platform for claim submission and robust reference attachment, supported by an explainable evaluation engine.
• Integrate AI-assisted structural claim analysis for advisory purposes, ensuring it does not influence deterministic reliability scoring.
• Provide comprehensive moderator access for reviewing, monitoring, and managing system data, fostering transparency and accountability.

Proposed System Architecture: Modular & Full-Stack

Veridex employs a modular, full-stack architecture based on the Model–View–Controller (MVC) design pattern, emphasizing scalability and maintainability. This structure facilitates clear separation of concerns and robust development.

Frontend: Engaging User Experience (Presentation Layer)

Developed using modern web technologies like HTML, CSS, and JavaScript, the frontend provides an intuitive and responsive user experience. It communicates seamlessly with the backend via RESTful APIs, ensuring a dynamic and interactive environment.

• Login & Signup Interface: Secure and user-friendly access points.

​

• Claim Submission Interface: Streamlined process for submitting claims and attachments.

​

• Discussion Interface: Facilitates engaging, claim-specific conversations.

​

• Moderator Dashboard: Restricted access for comprehensive oversight and management.

Authentication tokens, such as JWT, are meticulously utilized to maintain secure and persistent user sessions.

Backend: Robust Application Logic (Express.js)

The backend, powered by Express.js, is structured into several modular components, each handling specific functionalities to ensure efficiency and scalability.

Authentication Module

Manages user registration, login, password hashing, token generation, and implements role-based access control.

Claim Management

Handles claim creation, retrieval, status updates, and stores AI-assisted structural output.

Source Management

Facilitates file uploads and URL references, associating them with specific claims.

Evaluation Engine

Applies deterministic rules for reliability confidence indicators (source count, official source presence, freshness).

Discussion Module

Manages claim-specific discussions, supports posts (text, image, polls), voting, and replies, ensuring user accountability.

Moderator Module

Provides dashboard access for viewing claims, monitoring discussions, reviewing flagged content, and system-level actions.

Data Layer: Reliable Storage (MongoDB)

MongoDB serves as the backbone of our data layer, offering a flexible and scalable NoSQL database solution. Each entity is logically separated into distinct collections, ensuring modularity and efficient data management.

• Users: Stores user profiles, roles, and authentication details.

​

​

• Claims: Contains all submitted claims, their text, timestamps, and processing statuses.

​

​

• Sources: Manages all associated references (URLs, document metadata) and their categorization.

​

​

• Discussions: Holds claim-specific threads, posts, replies, and voting records.

​

​

• Verifications: Stores audit logs and evaluation results for transparency.

Security Layer: Protecting Integrity and Trust

Security is paramount for Veridex, with multiple layers implemented to protect user data and platform integrity.

01

Password Hashing

Utilizes robust algorithms like bcrypt for secure storage of user passwords.

02

Token-Based Authentication

JSON Web Tokens (JWT) are used for secure, stateless session management.

03

Role-Based Route Protection

Middleware ensures that users can only access resources aligned with their assigned roles (User/Moderator).

04

Input Validation

Strict validation protocols are applied to all user inputs to prevent injection attacks and data corruption.

05

Controlled File Uploads

Rigorous handling of file uploads minimizes security risks associated with user-submitted content.

Claim Lifecycle Model — Veridex

Veridex models each claim as a state-driven entity to guarantee governance, traceability, and auditability. The lifecycle enforces deterministic transitions and preserves provenance at every step.

1

1. Draft Creation

Author composes claim; metadata attached (source pointers, timestamps).

2

2. Claim Submission

Immutable submission event recorded; submission signature and hash stored.

3

3. Input Validation

Syntactic & schema checks; metadata normalization; reject on fatal errors.

4

4. Deterministic Evaluation

Parallel rule workers evaluate evidence; deterministic aggregator computes score.

5

5. Moderator Review

Human oversight reviews edge-cases, audits logs, and adjudicates disputes.

6

6. Decision & Archival

Approved or Rejected state recorded; complete audit trail and archival retention.

7

7. Re-evaluation Loop

New sources trigger deterministic re-evaluation; previous states preserved for history.

Diagram suggestion: state-transition lifecycle flow (state nodes + guarded transitions). Emphasize: claims evolve (not static) — every transition is logged for patent-grade traceability.

Working Principle — Stepwise Workflow

1. Claim Submission: structured text + metadata

​

• Input Validation: schema, sanitization, duplicates

​

• Source Attachment: fetch, metadata extraction

​

• Parallel Deterministic Evaluation: multiple rule workers

​

• Confidence Computation: weighted aggregation of rule outputs

​

• Reliability Categorization: categorical label with supporting evidence

​

• Moderator Review: human oversight, dispute resolution

​

• Final State / Archival: immutable record with audit trail

Fork–join concept: concurrent rule execution (fork) then deterministic aggregation (join) to minimize dependence on AI and ensure traceability.

Reliability categorization

Deterministic classification

Claim submission

Structured text and metadata

Fork–Join Evaluation Flow

Confidence computation

Combine results into score

Source verification

Deterministic provenance checks

Metadata checks

Schema and integrity validation

Factual cross-reference

Independent evidence matching

Parallel deterministic evaluation reduces latency and provides independent evidence tracks that are deterministic and auditable.

Fork–Join Deterministic Evaluation Architecture

The evaluation engine applies parallel, independent rule checks (fork) and merges deterministic outputs (join) into a reproducible confidence score. All steps yield auditable artifacts.

Parallel Rule Workers (Fork)

• Source Count Check — verify multiplicity and independence

​

• Official Source Detection — authoritative registry matching

​

• Freshness Verification — timestamp and content delta analysis

​

• Metadata Validation — schema, signatures, and provenance links

Deterministic Aggregation (Join)

Deterministic merge function computes final confidence; reproducible given same inputs and rule set.

System Advantages:

• Reduced latency via concurrent rule execution.

​

• Independent evidence verification reduces correlated failure modes.

​

• Deterministic reproducibility supports formal verification and patent claims.

​

• Transparent logs enable post-hoc audits and regression testing.

Diagram suggestion: fork–join workflow with labeled worker boxes and deterministic aggregator; include sample log snippet showing worker outputs and final score.

Future Scope

Integration of advanced NLP & Machine Learning

For claim similarity and clustering.

Multilingual support

Extending capabilities for global claim analysis.

Browser extensions & plugins

Enabling real-time reliability indication across social media and web.

User reputation scoring

Implementing a robust system to track and verify user credibility.

Optimizing for high-performance, large-scale usage.

Cloud infrastructure deployment

AI Advisory Analysis Module: Intelligence Without Bias

The AI Advisory Analysis Module is designed to assist moderators and users by providing structural insights into claims. Crucially, its output is purely advisory and does not directly influence the deterministic reliability scoring, preserving the neutrality and explainability of the evaluation engine.

​

​

​

​

​

​

​

​

​

​

​

​

• Structural Claim Breakdown: Analyzes claims to identify core components and arguments.

​

• Similar Claim Detection: Flags existing claims that resemble new submissions to identify patterns or recurring narratives.

​

• Potential Conflict Identification: Highlights areas where a claim might contradict established or verified information.

​

• Informational Summaries: Generates concise summaries to aid in understanding complex claims.

Separation of AI Advisory and Decision Authority

Architectural separation enforces that AI components provide contextual assistance only; deterministic engines and human moderators retain exclusive decision authority.

AI Advisory Module

Generates summaries, suggested evidence links, and auxiliary provenance annotations. Outputs are tagged as non-authoritative.

Rule-Based Evaluation Engine

Executes deterministic rules; computes reliability metrics from structured evidence.

Moderator

Final oversight, policy enforcement, and resolution of ambiguous outcomes; can override with documented rationale.

Key assertion: AI outputs are advisory only and never determine reliability outcomes. This separation yields: explainability, ethical AI usage, reduced bias, and auditable decision logic.

Diagram suggestion: parallel blocks showing AI assisting (dashed arrow) while decision authority path remains separate and signed.

Component

Responsibility

AI Advisory Module

Contextual summaries, source suggestions, confidence hints (advisory)

Rule-Based Evaluation Engine

Structural claim analysis, deterministic scoring, evidence aggregation

Moderator

Governance, final adjudication, audit annotations

Summary, Evaluation & Conclusion

Experimental Observations

Source credibility strongly influences confidence; parallel evaluation improved throughput and maintained deterministic reproducibility; governance reduced false escalations.

Advantages over Existing Systems

Explainable rule traces, separation of deterministic rules and AI advisory, lifecycle modeling, multi-level review, and auditable governance records.

Contributions & Future Work

Veridex contributes a transparent, scalable framework for reliability scoring. Future directions: richer rule DSL, federation of governance, performance benchmarking, and formal verification of rule engine.

Veridex is designed to assist reviewers and patent evaluators by providing reproducible, explainable, and governance‑integrated reliability assessments without asserting absolute truth.