Top Stories
New to mold illness? Start with Mold 101.
Ten chapters on Chronic Inflammatory Response Syndrome (CIRS), the 13 symptom clusters, biomarkers, testing, and the Shoemaker treatment framework. Written for people figuring this out for the first time.
Is mold affecting your health?
The Starter Health Panel checks 3 biomarkers linked to mold-related illness. Blood draw at 2,000+ LabCorp locations. Results delivered to you.
Latest News
View All →Think mold is making you sick?
MoldCo clinicians specialize in mold-related illness. Start with a $129 telehealth consultation. Treatment plans built around your data.
Book a ConsultationMore Stories
Submit a News Tip
Got a story the mold world should know? I review every submission. Published tips get credited.
Heads up before you submit: Don't include personal health info (your medical records, lab results, etc.) and don't share info about anyone else without their consent. By submitting, you grant The Mold Report a non-exclusive license to publish your tip with attribution. See Terms.
Tip Received
Our AI editorial pipeline will review, verify, and publish it. You'll be credited.
AI Journalism, Done Right
Every article passes through a multi-stage AI pipeline before anything goes live. No hallucinations. No fabricated sources. Full transparency below.
Source & Filter
Monitor feeds, reject duplicates, verify sources, screen reader tips
4 gatesScore
AI rates newsworthiness 1-10 for the mold community. Only 7+ advances.
1 gateWrite
Rewrite headlines for clarity. Rewrite summaries in our editorial voice with expert context.
2 gatesVerify & Publish
Claims compliance, Shoemaker alignment, photo assignment, SEO optimization
4 gatesAll 11 Pipeline Gates
1. Dedup
Content-aware duplicate detection catches repeat stories before they publish
2. Freshness
Rejects stale content older than 90 days to keep the feed current
3. Source
Verifies every article has a real, reachable source URL
4. Tip Screening
Reader tips are checked for editorial validity and Shoemaker alignment
5. Interest
AI scores each story 1-10 on newsworthiness; must score 7+ to pass
6. Headline
Rewrites titles for clarity and reader engagement
7. Rewrite
Rewrites each summary in Mold Report voice, adding expert context
8. Compliance
Checks and auto-corrects against MoldCo claims compliance rules
9. Research
Verifies Shoemaker alignment and scientific accuracy for research articles
10. Photo
Assigns images: OG source, topic-specific Unsplash match, or category fallback
11. SEO
Generates optimized meta titles and descriptions for search visibility
Our Weights & Biases
Full transparency. Here's exactly how our AI editor thinks, what it prioritizes, and what it's biased toward.
Model
Two Anthropic Claude models, depending on the gate: claude-sonnet-4-6 for nuanced work (editorial rewrite, research lens) and claude-haiku-4-5 for fast, structured calls (interest scoring, tip screening, headline, compliance, photo, SEO, duplicate detection). Nine agent roles in total.
Source Bias
Weighted toward peer-reviewed journals, government agencies (EPA, CDC), and established news organizations. We deprioritize press releases, affiliate content, and social media.
~31% academic / gov sources, ~51% from established news outlets too
Research Framework
Aligned with the Shoemaker Protocol research framework. Biomarker references prioritize TGF-B1, MMP-9, MSH, C4a, VIP. This is a known bias. We believe it's the most evidence-backed framework available.
Editorial Voice
Neutral, data-driven, journalistic. No em dashes. No alarmist language. No promotional framing. Lead with facts, end with context. 150-250 words per summary.
Known Limitations
AI cannot verify paywalled sources. Cannot access real-time lab data. Cannot interview human subjects. Cannot detect subtle p-hacking in cited studies. We acknowledge these limits openly.
Pipeline Version History
Weekly site-improver pass. Regenerated sitemap.xml — 23 newer articles were missing and 4 stale URLs pointed to deleted shares (200 article URLs now, up from 173). Added og:image, dimensions, and twitter:image to about, category, research, providers, privacy, terms, and mold-101 so social-card previews stop rendering blank. Fixed a broken JSON-LD publishingPrinciples link that was pointing to a non-existent /editorial-standards page; it now resolves to /about.html#editorial. Trimmed 2 oversized SEO titles and normalized 5 duplicate source names (Sci Rep → Scientific Reports, Mol Biol Rep → Molecular Biology Reports, etc.). Backfilled pipeline_config.json so sync_transparency.py won't clobber recent history if anyone runs it.
Weekly site-improver pass. Filled in missing seoTitle and seoDescription on the MARCoNS/alpha-MSH study article — it was rendering with no meta description in Google and Twitter previews. Regenerated sitemap.xml: search engines were missing 6 newer articles (177 article URLs now, up from 171). Re-synced the embedded articles blob in index.html with articles.json after the metadata fix.
Trimmed oversized SEO titles (66) and meta descriptions (32) at word boundaries so Google stops cutting them off mid-sentence. Wrapped newsletter.html in proper HTML with canonical and OG tags (it was a bare fragment). Normalized 8 inconsistent source names (e.g. aol.com → AOL, pbs.org → PBS). Corrected the “Source Bias” transparency card from a misleading 85% to the actual ~31% academic/gov (~51% with established news included). Re-synced the embedded articles blob and added newsletter.html to sitemap.xml.
Weekly site-improver pass: regenerated sitemap.xml to surface 20 newer articles search engines were missing (157 total). Added rel=canonical, og:url, and Twitter Card tags to about.html and category.html. Corrected the stale model reference in the “Our Weights & Biases” transparency card — now reflects the actual two-model setup (claude-sonnet-4-6 for nuanced gates, claude-haiku-4-5 for fast gates).
New composite scoring for Top Stories: articles ranked by featured status, recency decay, source authority, and Shoemaker alignment. Category cap prevents any single topic from dominating the hero. Added deterministic compliance lint gate to catch claims violations before the AI review. Fixed image repetition with pool-based dedup picker. Mobile share-bar overflow fix.
Refactored pipeline into secure two-phase flow: --fetch-only for candidate retrieval, --publish for human-approved articles. Connected tip submission form to Formspree backend (was broken since launch). Added dedicated editorsNote field with styled rendering. Featured MARCoNS research study as lead article.
Standalone article pages with full OG tags, Twitter cards, and JSON-LD structured data. Added sitemap.xml and robots.txt. Upgraded SEO agent with keyword extraction and search-optimized headlines. Major pipeline audit: removed dead code, added parse safety defaults, enforced summary length limits, fixed editor's note generation.
Integrated Shoemaker research corpus (82 papers, 8,000+ chars) as the research agent's knowledge base. Activated Shoemaker Lens as pipeline gatekeeper — off-topic content now killed at the research gate. Added image URL validation to prevent broken images. First batch of fully autonomous publishes.
Category archive pages with search, sort, and filtering. Redesigned AI Journalism transparency section with phase cards and collapsible details. Fixed mobile nav (close on tap, dropdown styling). Added UTM tracking to all MoldCo partner links.
New Gate 0 in pipeline: content-aware dedup with three detection layers. Removed 14 duplicate articles from seed database. Upgraded photo agent with topic-specific Unsplash matching.
Fully autonomous pipeline: RSS fetch, 11-gate AI review, auto-publish. Added PubMed, EPA/CDC/HUD feeds, reader tip intake, and SEO optimization gate.
Added compliance agent enforcing MoldCo claims rules, research verification agent for Shoemaker alignment, and headline rewriting agent.
Articles now scored 1-10 on newsworthiness. Only 7+ articles publish. Added source verification gate and category-based image fallbacks.
Initial launch with seed articles, manual curation, and basic AI rewriting.