[v4] Implement predatory journal flagging

[cpitzi]

Summary

Identify references citing journals classified as predatory or questionable by established assessment methodologies.

See roadmap/v4-features.md (Priority 3) for full specification.

Open Questions

• Data source: Cabells Predatory Reports is the gold standard but is commercial. Is API access feasible? Cost?
• Alternative sources: Beall's List (archived, not maintained), community lists, DOAJ membership as positive signal
• Classification language: "Predatory" is contentious. Should the auditor use "potentially predatory," "not indexed in DOAJ," or another framing?

Implementation Approach

1. Check journal against known predatory journal databases/lists via web search
2. Check DOAJ membership as a positive legitimacy signal
3. Check Scopus/PubMed indexing as additional legitimacy signals
4. Flag journals that appear on predatory lists or lack any recognized indexing

Acceptance Criteria

• Identifies journals on known predatory lists
• Distinguishes "confirmed predatory" from "potentially predatory"
• Does not flag journals removed from predatory lists after improvement
• Handles journal name variations (predatory journals frequently rename)
• Classified as Elevated risk (not High — the cited paper may still be legitimate research)

Dependencies

• Requires resolution of the data source question before implementation

[cpitzi]

Data-source decision (2026-06-19): Hybrid — indexing whitelists + free community blacklists

Resolving the open data-source and classification-language questions.

Architecture constraint: the auditor is a single prompt using live web search — no code, no API layer, no database. So Cabells Predatory Reports is out of scope for now (paywalled, and would require the API-integration pipeline #11, which is deferred). All signals must be freely reachable via web search at run time.

Approach

Primary — positive signal (whitelists): for each cited journal, check indexing in DOAJ (api.doaj.org), PubMed/MEDLINE, the Scopus source list, and the Web of Science Master Journal List. Any reputable indexing is a legitimacy signal that clears the journal.

Secondary — corroboration (free blacklists): cross-reference freely-available community lists (archived Beall's List, Stop Predatory Journals). Corroborating evidence only — never the sole basis for a flag.

Flagging: a journal indexed in none of the recognized databases while claiming to be peer-reviewed is flagged Elevated — never High in isolation (the cited paper may be legitimate research in a weak venue); escalate to High only when combined with another heuristic trigger.

Classification language (resolves the contentious-language question): report factually — e.g., "not indexed in DOAJ, Scopus, PubMed, or Web of Science; also listed on [community source]." Avoid the verdict word "predatory"; at most "potentially predatory / unverified venue." This sidesteps the defamation and methodological risk of labeling a named journal predatory.

Must NOT flag: grey literature (government/organizational reports have no journal indexing — exempt, per the existing grey-lit rule); established subscription journals indexed in Scopus/WoS; journals that left blacklists after improving. Must handle journal renaming.

Implementation shape

Ships as a new Heuristic 10 in a new prompts/v6-auditor.md (v5 is frozen by its committed #40 baseline), plus docs/heuristics.md + README + CLAUDE.md updates, a dedicated test set (using clearly-fabricated or well-documented-defunct venues, framed as "unindexed" rather than naming a real journal "predatory"), and a v6 baseline-validation follow-up mirroring #40.

This unblocks #7.