ADR-001: Hybrid Forensics Methodology

Status

Accepted - 2026-04-20

Context

The project needs a clear baseline method for forensic processing while remaining easy to test and evolve. Early implementation should avoid overfitting to one source type or one analysis method.

Decision

Adopt a hybrid methodology that separates the workflow into four deterministic stages:

1. scrape raw source material
2. extract lexical/structural features
3. analyze features into confidence labels
4. report findings in human-readable markdown

Consequences

• Positive:
  • Stage boundaries remain stable while implementations improve.
  • Unit and integration tests map cleanly to each stage.
  • Outputs are auditable by intermediate artifacts.
• Negative:
  • Initial implementation may appear simplistic until richer models are added.
  • Repeated stage execution can duplicate work in scaffold mode.

Follow-Up

• Introduce cache/reuse between stages once artifact contracts are finalized.
• Add eval cases that assert behavior across realistic claim/evidence scenarios.