Changes

The accuracy check of 14 October moved potassium citrate from tier 1 to tier 2 because no primary absorption study comparing it with potassium chloride could be verified. We have put it back to tier 1: it is widely reported as equivalent in absorption to potassium chloride and the downgrade rested on a missing citation, not on evidence that it absorbs worse. The form page still says that no primary comparison was verified.

Separate AI agents re-checked 842 claims (the glossary and the reference data for all 69 compounds and their forms) against primary sources, starting from each claim rather than the existing citation. 616 held up, 36 were wrong, 127 imprecise and 63 unverifiable; all 232 cited PubMed IDs exist and every RDA and upper limit matched. Corrected: a vitamin K trial cited under the wrong PubMed ID, a zinc form rationale that named the wrong comparator, a potassium paper credited to the wrong author, NAC and glycine citations, and form rationales that said no human comparison existed when small ones do (magnesium glycinate, citrulline malate, alpha-lipoic acid, lycopene sources, chlorella, creatine forms). Dose ranges the cited papers did not support were changed: magnesium 250-400 to 241-300 mg, iron high 200 to 120 mg, selenium low 100 to 200 mcg, iodine high 300 to 150 mcg, vitamin C high 1,000 to 500 mg, CoQ10 low 100 to 300 mg, NAC high 2,400 to 3,000 mg, berberine low 300 to 1,000 mg, citrulline high 8,000 to 5,800 mg and biotin low 100 to 300 mg, with fairy-dust thresholds recomputed. Glossary entries were fixed where they misdescribed our own momentum score, near-twin match and quality tiers, and where examples used figures from before the compound list grew; the formula-twins and CoQ10 posts were updated to the current numbers. This was a second AI pass, not human review.

After Google updated its guidance on AI-generated and scaled content (October 1, 2026) and Search Console showed a drop, we audited the site. Pages that are a short template around a table were marked noindex and left out of the sitemaps: formula-twin cluster pages under 25 products, individual glossary terms (the glossary page now carries every definition), insight permalinks, compound-form pages under 200 products, and product pages with no dose we can place in the market or that are off the market. Only a first tranche of product pages is listed in the sitemap, richest first. We also fixed pagination links that redirected, made product URLs without a trailing slash redirect, and added a statement of how the site is made, including what AI does and what has not been reviewed, to every page and the methodology page.

The brand-page figure for products with a certification statement matched the letters NSF, USP and GMP anywhere in a statement, so words like "transfer" counted. It now uses the same whole-word matching as the Certifications report (USP Verified, NSF, Informed Sport/Choice, BSCG, ConsumerLab, GMP), which lowers most brands' figures substantially.

Added molybdenum, boron, inositol, taurine, L-glutamine, L-arginine, L-citrulline, L-carnitine, beta-alanine, L-lysine, alpha-lipoic acid, hyaluronic acid, chondroitin, MSM, astaxanthin, lycopene, zeaxanthin, resveratrol, grape seed extract, green tea extract, ginger, spirulina, chlorella, phosphatidylserine, bromelain, saw palmetto, milk thistle, cranberry and L-tyrosine, each with cited reference data. Huperzine A was researched but not added: its label rows mix micrograms of the marker with milligrams of plant extract, so any dose figure would be wrong. Dose figures for every compound and the Stack Checker now cover the larger set.

Probiotic amounts on labels are declared as billions or millions of CFU, milligrams of culture, or only as a blend total, and were stored without conversion to one unit, which produced a meaningless median. Probiotic dose statistics (median, deciles, fairy-dust share, gummy comparison, percentile widget) are now suppressed; counts, forms and blend share are unchanged.

Brand state fields are free text ("CA", "California", "Ca."). The Where Supplements Come From report used to count each spelling separately; it now merges exact state names and postal codes into one state and leaves anything it cannot classify out, so the top-five share moved from 42.9% to 49.3% of the 4,187 brands with a US state.

The Q-SYMBIO reference summary gave all-cause mortality as 9% vs 17%. The paper's abstract reports 10% vs 18% (P=.018) for all-cause mortality, 9% vs 16% for cardiovascular mortality, and 15% vs 26% for the primary endpoint; the entry now cites PMID 25282031 with those figures.

First DSLD dump ingested (214,780 labels, 100% parse success). Reference data curated for the initial 40 compounds and their common forms, with elemental-fraction stoichiometry and bioavailability tier citations. 155 canonical excipients aliased. All 12 spec metrics computed for the first time.

Confirmed empirically that DSLD declares mineral amounts elementally when the ingredient row is named after the bare element (e.g. "Iron"), and as salt weight when named after the salt itself (e.g. "Calcium Carbonate") — see amount_conventions.md. Also fixed a unit-normalization gap where gram-declared amounts weren't being converted to the compound's milligram basis before comparison.

The initial Claims Census ranked phrases by raw occurrence count, which let one high-SKU brand's repeated boilerplate dominate the top 10. Changed to rank by distinct-brand adoption instead.