Search news articles.
Returns a list of matching articles. Each article includes:
- article_id, classification_id, title, source, date, link, category, rank, total_shares, summary
- bias_values: dict of per-dimension bias scores using plain-text keys (e.g. 'liberal conservative bias'),
same schema as get_bias_from_url and get_all_source_biases (when available)
- bias_analysis_status: 'evidence_ready', 'evidence_unverified', 'evidence_partial',
'evidence_failed' (all scored dimensions' quotes failed verification, so the scores do
not match the article text), 'scored_legacy', or 'pending'
- evidence_ratio: fraction of scored bias dimensions whose supporting quote is verified
(0.0-1.0). Raise min_evidence to demand only articles with verified quotes.
- bias_dimensions when include_evidence=true: a self-contained object joining each score,
scale, evidence status, claim, evidence, counterevidence, confidence, and rationale.
Quotes include verification method and exact character offsets when raw-text matching succeeds.
Dimension evidence_status is one of: verified, provided_unchecked, quote_mismatch,
metadata_incomplete, metadata_only, or missing.
- bias_analysis: contract/schema/model/prompt provenance, generation and review status,
input scope/hash/size, limitations, quote-verification method, and explicit evidence coverage
- context: AI-generated contextual background for the article (when available)
- implicit_assumptions: tacit or unstated premises the article's claims or framing rely on
(list of concise strings, when available)
- extracted_data: structured quantitative/qualitative facts extracted from the article
- raw_data: legacy serialized form of extracted_data
Args:
query: Optional search keywords. Leave empty to return the most recent articles in
scope (use with bias to rank them). e.g. 'NVDA earnings'.
limit: Max results (1-100, default 20).
source: Filter by source name, e.g. 'CNN', 'Reuters'.
category: Filter by category. One of: 'trending', 'tech', 'markets', 'politics',
'business', 'science', 'memes'.
days_back: Only include articles from the last N days. 0 means no date filter.
Default: 90. Widen this (e.g. 720) for older coverage.
min_shares: Minimum total social shares.
sort: Sort order. One of: 'rank' (relevance, default), 'date' (newest), 'shares' (most shared).
bias: Bias dimension to rank by, highest score first. This is a ranking, not a
standalone filter: an empty query still returns other recent articles, ranked
with the bias dimension on top. Any canonical bias key, e.g.
'liberal conservative bias', 'overall credibility', 'conspiracy bias'.
Ranking is scoped to recent articles (the days_back window, or 365 days when
days_back is 0) so one old high-scoring outlier cannot dominate.
include_evidence: Include claim-level evidence, counterevidence, confidence, rationale, and limitations.
Defaults to false to keep search payloads compact.
only_analyzed: Return only articles with valid canonical bias scores.
min_evidence: Minimum fraction of scored dimensions with verified quotes (0.0-1.0, default 0).
Raise this to request only articles whose scores are backed by verified evidence,
e.g. 0.5. Pair it with only_analyzed to get quotable results instead of
pending records with empty bias_values. Returns a 400 if sort or bias is not a valid option.