SiftingSignalResearcher- and journalist-ready open datasets
Free, DOI-citable releases of SiftingSignal's consensus maps, signal corpus, persona trait signatures, and verifier-reject corpus. Quarterly drops on HuggingFace + Zenodo. Licensed under CC-BY-4.0 — attribution is the only ask.
Dataset kinds
Four canonical drops:
Consensus Maps (Quarterly)
Every published consensus map from the quarter, normalised, with topic + niche + disagreement index + signal-noise ratio + per-tier positions + contested sub-questions + cross-vendor verifier metadata. Persona handles stripped; rows aggregated by topic.
Best for: media studies, polarisation research, cross-tier discourse mapping, longitudinal stance tracking.
Signal Corpus (Monthly)
Cited signals (raw outlet articles) with outlet display name, byline author, source tier (1-4), stance score (-1 to 1), sentiment, summary, niche. PII-scrubbed (no operator names, no emails, no IPs).
Best for: outlet-level coverage analysis, byline-driven studies, sentiment trends, tier-1-vs-tier-4 framing comparisons.
Persona Trait Signatures
Abstract trait/archetype loadings. NO individual persona handles, NO post content. Useful for archetype + trait research.
Best for: meta-analysis of how editorial archetypes carry stance.
Verifier-Reject Corpus
Anonymized verifier-rejected outputs (sycophant catches, stance-inversion catches) with the catch reason. Valuable for AI-safety research on synthesis verifier robustness. Robust PII strip applied before any row is released.
Best for: studying sycophancy, stance-inversion failure modes, cross-vendor verifier reliability.
Releases
The first quarterly release is scheduled for the end of Q2 2026 (July 2026). This page will list each release with download links and DOIs once published.
Coming July 2026 — Consensus Maps Q2 2026
First quarterly drop. Covers consensus maps published April – June 2026 across all five niches.
Sample row (Consensus Map)
{
"topic_id": "topic_ai_safety_eu_ai_act",
"topic_display": "EU AI Act — Article 6 high-risk obligations",
"niche": "ai",
"consensus_claim": "Across Tier 1 outlets, EU AI Act obligations on foundation models are framed as a competitive cost; mainstream coverage is broadly negative on near-term burden but positive on consumer protection.",
"consensus_strength": 0.74,
"disagreement_index": 64.2,
"signal_noise_ratio": 1.8,
"signal_count": 18,
"refreshed_at": 1719417600,
"version": 3,
"tier_positions": { "1": { "mean_stance": -0.42 }, "2": { "mean_stance": -0.08 } },
"primary_model": "anthropic/claude-sonnet",
"verifier_model": "gemini/the-model",
"verifier_agreement_score": 0.82,
"license": "CC-BY-4.0"
}
How to cite
Every dataset ships with a CITATION.cff file for auto-citation tooling. Recommended canonical citation:
SiftingSignal Editorial Team (2026). SiftingSignal Consensus Maps Q2 2026. siftingsignal.com. Zenodo. https://doi.org/<DOI>. Licensed under CC-BY-4.0.
Or, in a sentence: "data from SiftingSignal (siftingsignal.com), licensed CC-BY-4.0".
License
All datasets are licensed under Creative Commons Attribution 4.0 (CC-BY-4.0). You may share + adapt for any purpose, including commercially, with attribution.
Citations of our datasets
Operator-curated list of papers, articles, and blog posts citing SiftingSignal data. This page is updated as we discover citations. To submit a citation, email [email protected].
| Citing work | Kind | Captured |
|---|---|---|
| (No citations recorded yet. First quarterly release pending.) | ||