Skip to content
Stetwright

Changelog

Rule-set releases.

Read from the rule-set when this site is built. Every report records the version it was checked against.

  1. 0.8.0

    • New HHS, OIG, CMS and ACA sources stored (65). Regulations: 42 CFR 1001.952, 1003 (1003.100 to 1003.1600), 411.350 to 411.357, 422.2266, 422.2267, 422.2272, 422.2274, 423.2264 to 423.2267, 423.2272, 423.2274, 438.10, 457.1224, 45 CFR 155.220, 156.225; 42 U.S.C. 1320a-7a, 1320b-10, 1395nn. OIG: Special Fraud Alerts on Medicare Advantage marketing payments (2024), speaker programs, laboratory payments and DME telemarketing, the 2002 gifts bulletin, the 2016 nominal-value policy statement, the 2023 General Compliance Program Guidance, and 25 advisory opinions on referral services, online platforms, advertising and beneficiary incentives. CMS: the Medicare Communications and Marketing Guidelines (3-16-2022). Also AHA v. Becerra (N.D. Tex. 2024).
    • HHS pages that refuse scripts were read through Internet Archive copies and stored as saved by hand: the OCR online tracking technologies bulletin (with its notice about the June 20, 2024 vacatur), the 17 OCR marketing FAQs, and OCR resolutions on online-review replies and website testimonials (Manasa, New Vision Dental, Elite Dental, Complete P.T., Cadia). Four DOJ press releases on marketing kickbacks were stored the same way.
    • 42 CFR 422.2268 and 423.2268 are not stored: govinfo returned the 2020 edition, and the sections were removed in later editions.
    • New rules: HIPAA marketing and tracking (CG-HIP-101 to 104), OIG fraud and abuse in marketing (CG-OIG-101 to 107), Medicare Advantage and Part D marketing (CG-CMS-101 to 110), ACA marketplace marketing (CG-ACA-101 to 103) and Medicaid and CHIP managed care marketing (CG-MCD-101, 102). Each cites text held in the source library, and all need counsel review.
    • Platform policy text stored for Meta, LinkedIn, X, Reddit, Microsoft Advertising, Amazon Ads, Spotify, Taboola, Outbrain, Nextdoor, Apple Ads, Yahoo and more of Google (41 sources, 122 excerpt sections). Internet Archive copies are dated on the Sources page.
    • CG-PLT-001, 004, 005 and 006 now cite stored Meta pages and are verified. CG-PLT-002 (engagement bait) stays unverified.
    • New platform rules CG-PLT-020 to 031 (telehealth, click-bait, body shaming, wearable weight loss, weight loss without lifestyle change, guaranteed weight loss, discouraging conventional medicine, improper FDA claims on supplements, addiction treatment, minors, unsafe supplement ingredients, health data in tracking events).
    • Existing platform rules (007 to 011, 013 to 019) now cite the new platform pages as well and name their platforms.
    • New "Ad platforms" picker: a platform rule runs only for the platforms you tick; with none ticked every platform's rules run. Reports record the platforms.
    • New state sources stored and indexed for 22 states (AL, AZ, CO, GA, HI, IN, KY, LA, MD, MI, MN, MO, NJ, NM, OH, OK, OR, PA, SC, TN, UT, WI) and for Texas Administrative Code chapter 164 (physician advertising). Text came from official sites that allow scripts, the Public.Resource.Org release of the Georgia Code, Internet Archive copies of the official pages, and, for Indiana, the built-in browser. The first line of each stored file records where it came from and the snapshot date.
    • 139 new state rules, written from the stored text and each carrying a verbatim quotation of the provision it rests on, checked against the stored text. They cover false or misleading advertising by licensees, referral and fee-splitting payments, professional-conduct advertising rules, unfair or deceptive practices acts, and a few health-data and Medicaid kickback provisions. Texas gains CG-TEX-013 to 019 from 22 TAC chapter 164 (board eligible, permanent cure, testimonials, free-service claims, unique skill claims, paid-for labels, patient photo consent). Most are warnings because they bind licensees. All need counsel review.
    • Ohio and Michigan pages were re-read live in a browser and match the stored copies, except for a hyphen and the history list on MCL 333.16221.
  2. 0.7.0

    • New federal sources stored, FDCA 503A and 503B (21 U.S.C. 353a, 353b), Ryan Haight Act provisions (21 U.S.C. 829 and 802, 21 CFR 1300.04), 42 CFR 438.104, FTC's Endorsement Guides FAQ, and two FDA compounding pages. 21 CFR 202, 21 CFR 101.93 and the FTC Health Products Compliance Guidance were already stored.
    • New state sources stored, Massachusetts (c. 93A s. 2, c. 175H s. 3), North Carolina (G.S. 90-14, 75-1.1), Virginia (Code 54.1-2915, 59.1-200, 59.1-578, 32.1-127.1:03), Nevada NRS 603A (consumer health data), Connecticut chapter 743jj and Washington RCW 19.373.020. New Jersey, Pennsylvania, Georgia, Ohio and Michigan pages are registered but not stored, because the official sites refuse scripted access or render the law with JavaScript.
    • New rules for compounded-drug advertising (CG-CMP-001, 002), controlled substances with no in-person visit (CG-TLM-001), Medicaid managed care marketing (CG-MCD-001 to 003), ambiguous endorsement disclosures (CG-END-001), Nevada and Connecticut consumer health data (CG-NEV-001 to 003, CG-CTD-001, 002), Virginia (CG-VIR-001, 002), North Carolina (CG-NCA-001, 002) and Massachusetts (CG-MAS-001, 002). Each cites text held in the source library, and all need counsel review.
  3. 0.6.0

    • Broad-claim coverage. A condition-and-outcome detector flags a sentence that names a disease or symptom and says it was cured, reversed, beaten or removed. About 120 new rules (broad_claims and train files) cover vague superlatives, safety absolutes, conspiracy framing, stem cell and addiction claims, Spanish-language claims, review gating, kickbacks, Medicare scams, controlled-substance and quiz-prescribing shortcuts, tracking and data sales, and influencer disclosure.
    • Plain negations and disclaimers ('does not replace your doctor', 'has not been shown to treat', 'not meant to diagnose') are masked before checking, which cut blocking false positives on fresh compliant copy.
    • Severity raised from warn to block on 32 rules that rarely fire on compliant copy (mostly state rules, comparative claims and Medicare or referral rules).
    • The AI second read is now required before approval (report version 2). It rates each issue high, medium or low and judges each blocking rule hit as real or a false alarm. High-confidence issues must be fixed or overridden with a reason, or the read waived with a reason. The read is stored in the signed report and bound to the approval; it is model output, kept as reviewer-attested.
    • The standard FDA dietary supplement disclaimer is masked before checking, so it no longer triggers disease-claim rules.
    • Fewer false positives. Negated phrasing such as does not treat, no one cure, not endorsed by and no guarantee is now allowed, and accurate FDA-registered facility wording is allowed.
  4. 0.5.0

    • Medicare plan marketing rules (CG-CMS-001 to 007) from 42 CFR 422 and 423 subpart V, and controlled-substance advertising rules (CG-DEA-001 to 003) from 21 U.S.C. 841(h) and 21 CFR 1306.04. The CMS rules bind plan sponsors and their agents, not every advertiser.
    • Exported reports are signed with a key kept in the browser, so an edited report cannot be re-hashed to pass. Report format adds an optional signature field.
  5. 0.4.1

    • New CG-DIS-015 (cancer-growth claims, successful in cancer treatment, and condition plus treatment or cure phrasing). CG-DIS-010 now also catches ideal for, good for and effective for a condition, and symptom-relief wording.
    • Custom policy rules are rebuilt from their phrases wherever they are loaded, so a crafted report cannot supply its own pattern.
  6. 0.4.0

    • State rules for California (CG-CAL-001 to 012), New York (CG-NYS-001 to 010), Texas (CG-TEX-001 to 012) and Illinois (CG-ILL-001 to 007), written from statute text stored in the source library. Washington and Florida are unchanged.
    • Illinois text comes from the General Assembly's file repository. California, New York and Texas text was read in a normal browser and saved by hand. Three of those pages were re-read and matched exactly. The rest were not re-checked.
    • The state rules cover only provisions that show up in the wording of an ad. Most apply to licensees.
  7. 0.3.0

    • State law. Washington (CG-WAS-001 to 008) and Florida (CG-FLA-001 to 006) rules written from the stored statute text. A check runs state rules only for the states chosen.
    • Platform policy rules (CG-PLT-001 to 003), written from memory of Meta and Google policy pages and marked unverified.
    • Rules from real regulator claims. New or widened rules for disease-marker, antiviral, pain-relief, condition-list and regenerative claims, safe-and-effective claims, replacing medicines, and compounded GLP-1 sameness claims (CG-DIS-010 to 014, CG-EFF-009 to 011, CG-RPL-003, CG-REG-007, and wider CG-DIS-002, 007, 009, CG-ABS-008, CG-REG-006).
    • Rules can name a notice that must be absent from the text (the unless field). CG-FLA-001 and 002 use it.
    • Federal sources added, covering CAN-SPAM, TCPA, 42 CFR Part 2, the Health Breach Notification Rule, 21 CFR Part 99, the anti-kickback statute and OIG's telefraud alert.
  8. 0.2.0

    • New content type, research recruitment, and a research_recruitment category with 8 rules (CG-RES-001 to 008) written from FDA's Recruiting Study Subjects and Payment and Reimbursement information sheets and 21 CFR 312.7 and 812.7.
    • Every rule now carries the frameworks it draws on (FTC, FDA, HIPAA, IRB, OIG) and citations verified against the source library. A check can be limited to chosen frameworks.
    • Corrected the citation in CG-TES-002 from 16 CFR 255.2(d) to 255.2(b), where the typical-results text is.
  9. 0.1.1

    • Audit fixes. New rules CG-ABS-008, CG-EFF-007 and CG-EFF-008 (effortless-results, clickbait, "never worry again" claims).
    • Wider coverage in CG-DIS-004 (conditions ending -itis, -osis), CG-EFF-001 ("clinically shown"), CG-PRV-003 (more reply phrases) and CG-STA-001 ("chance of", "odds of").
    • Typographic quotes, dashes and non-breaking spaces are folded before matching.
  10. 0.1.0

    • First rule-set. Reimplements the 16 Cadence claim rules and the statistic, testimonial, price and patient-confirmation checks. Not diffed against the Cadence test suite.
    • Adds patient-identifier checks, content-type-specific severity, citations, suggested wording and allow patterns.
    • All citations are from general knowledge and have not been confirmed by counsel.

Next step

Spot something that should have changed?

Wrong finding or missed claim, send the rule id and the phrase. That is how rules get fixed.