Remix

Make your own version of Testimony Drift Tracker

Edit the prompt below and generate a customized version in seconds.

Original by MATHUJAA P M CSE-2023

Describe your app

Build a full-stack web application called "Longitudinal Semantic Drift Detection in Witness Testimony".

This is a research-oriented application for analyzing how factual claims made by the SAME witness change across repeated testimony sessions over time.

Core research pipeline:
1. Ingest witness testimony data.
2. Extract atomic factual claims from each statement.
3. Align claims belonging to the same underlying factual event across sessions (T1, T2, ... Tn), even when wording changes.
4. Compare aligned claims across time.
5. Detect multidimensional semantic changes in:
   - entity/actor
   - action
   - object
   - location
   - time
   - value/quantity
   - polarity
   - certainty
   - specificity
6. Represent claim evolution as a longitudinal claim graph.
7. Detect evolution patterns such as:
   - stable claim
   - progressive specification
   - delayed addition
   - omission
   - persistent shift
   - oscillation
   - reversal
   - certainty increase/decrease
8. Produce an explainable report showing exactly what changed between sessions.

IMPORTANT RESEARCH FRAMING:
The system must NOT claim that it detects lies, deception, guilt, innocence, or witness credibility.
It detects semantic/narrative changes and inconsistencies in testimony.
Use neutral terminology such as "claim change", "semantic drift", "inconsistency", "addition", "omission", and "evidence inconsistency".

DATA:
The primary dataset is the Gustafsson et al. eyewitness memory dataset.
It contains repeated testimony from the same participants across Session 1 and Session 2.
The application architecture should support additional future sessions (T3, T4, ... Tn), even though the current primary dataset has two sessions.

DATABASE:
Use a proper persistent database for:
- users
- cases
- witnesses/participants
- testimony sessions
- statements
- extracted claims
- claim alignments
- claim transitions
- evolution signatures
- analysis results

AUTHENTICATION:
Provide working user registration and sign-in with persistent server-side authentication.
Do not implement authentication using only localStorage.
API responses must be proper JSON and must not fall back to index.html.

USER INTERFACE:
Create a professional research dashboard with:
- Login/Register
- Case dashboard
- Witness/participant list
- Session timeline
- Statement viewer
- Extracted claims viewer
- T1 vs T2 claim alignment view
- Semantic drift analysis
- Claim evolution graph
- Evolution signature summary
- Explainable findings
- Exportable research report

For the analysis page, clearly show:
T1 claim → T2 aligned claim → changed dimensions → transition type → confidence/score.

ARCHITECTURE:
Use a maintainable full-stack architecture.
Keep the frontend separate from backend/API logic.
Create clear modules for claim extraction, claim alignment, transition analysis, graph construction, and evolution-signature detection.

The system should be designed so that future versions can replace rule-based components with trained NLP/ML models without rewriting the application architecture.

Do not fabricate research results.
Do not label a claim as false merely because it changed.
Do not claim the system is scientifically validated unless evaluation data and metrics have actually been produced.

Build the application as a functional research prototype, not merely a static UI mockup.

Free to use. No credit card required.