Original data report · dataset v2026.08.16
A build-time analysis of the currently indexable Hippie Scientist research library: evidence grades, structured study/source records, human-study records, reported participant counts, safety cautions, and explicit disagreement between supporting and unfavorable or null evidence relationships.
Strong / moderate
16%
51 of 318 indexable ingredients are currently A or B.
Preliminary / insufficient
76%
C, D, or Avoid/Insufficient profiles remain visibly separate from stronger evidence.
Structured safety cautions
84%
267 profiles contain caution signals; this is a screening statistic, not a risk rate.
Human trial records indexed
110
Across 922 deduplicated structured study entities. This is a record count, not a count of proven-independent underlying trials.
Human evidence source records
282
Publication/study entities classified as human evidence. 1 deduplicated study record with conflicting concrete design classes is conservatively classified as Other / unclear and excluded from human/trial counts until reconciled. Multiple publication records can still originate from the same underlying trial, cohort, or dataset.
Reported participants
~100,176
Only deduplicated human-evidence records with one unambiguous structured N are included. Overlapping publications may still count some participants more than once.
Endpoint heterogeneity
0
Deduplicated study entities with mixed directional endpoint contexts for the same ingredient. Cross-ingredient variation is excluded from this headline count.
Evidence distribution
Category comparison
Categories are ranked by the share of indexable ingredients currently graded A/B, then by structured human-trial record coverage. Human source and trial counts are deduplicated publication entities within each category, not citation incidences or proven-independent underlying studies. Category names come from canonical runtime records; small categories can be volatile.
| Category | Ingredients | A/B | Preliminary/insufficient | Unassigned | Human source records | Human trial records |
|---|---|---|---|---|---|---|
| Uncategorized | 310 | 51 | 239 | 20 | 282 | 110 |
| Copper-binding tripeptide | 1 | 0 | 0 | 1 | 0 | 0 |
| Dual GIP/GLP-1 receptor agonist (prescription drug) | 1 | 0 | 0 | 1 | 0 | 0 |
| GHRH analog peptide | 1 | 0 | 1 | 0 | 0 | 0 |
| GLP-1 receptor agonist (prescription drug) | 1 | 0 | 0 | 1 | 0 | 0 |
| Growth hormone secretagogue (ghrelin receptor agonist) peptide | 1 | 0 | 1 | 0 | 0 | 0 |
| Melanocortin receptor agonist peptide | 1 | 0 | 0 | 1 | 0 | 0 |
| Synthetic peptide (gastric-derived sequence) | 1 | 0 | 1 | 0 | 0 | 0 |
| Synthetic thymosin beta-4 fragment | 1 | 0 | 1 | 0 | 0 | 0 |
Claim-discipline signal
0
This counts only records carrying an explicit structured overreach or evidence-language violation flag. It intentionally avoids manufacturing a more dramatic statistic from weak heuristics.
Directional evidence relationships
0
supporting
0
mixed
0
contradicting
0
no clear effect
Evidence Change Tracker
Grade changes are append-only events with the old grade, new grade, date, reason, and source IDs when available. The tracker starts from recorded events rather than inventing historical upgrades from today’s database state.
Cite / reuse
The Hippie Scientist. The Hippie Scientist Public Evidence Dataset 2026. Version 2026.08.16. https://thehippiescientist.net/evidence/evidence-report/
Downloaded data includes stable study IDs, source identifiers, study class, reported sample size, dose, duration, population, outcome, limitation, safety outcome, ingredient relationships, evidence grades, and directional labels where structured data exists.
Metadata reconciliation
These are raw conflict magnitudes, not subjective severity scores. Candidate values remain available in the downloadable dataset so discrepancies can be reconciled at the source level.
Ambiguous studies
1
Canonical publications with a year, participant-count, or concrete study-class conflict.
Cross-family class conflicts
0
Class disagreements that cross evidence families rather than staying within one design family.
Largest year span
0 yr
Maximum difference between observed publication-year candidates for one canonical study.
Largest participant mismatch
0×
Largest ratio between conflicting structured participant-count candidates; maximum absolute spread is 0.
Evidence independence
The canonical research graph first deduplicates publication identities across profiles, then collapses publications only when explicit trial-registration, cohort, dataset, or parent-study lineage shows they belong to the same underlying evidence unit. Missing lineage stays unresolved and is never treated as proof that publications are either dependent or independent.
Primary-human publications
55
Unique publication identities within the matched public research-profile scope, classified as primary human research.
Independence-adjusted human units
55
Primary-human evidence units remaining after explicitly proven cross-publication dependence is collapsed.
Human publications collapsed
0
Publication identities removed from the adjusted count because shared underlying-study identity was positively established.
Primary-human metadata coverage
20%
Share of unique primary-human publications with explicit registry, cohort, dataset, parent-study, or same-trial metadata available for independence assessment.
Because unknown relationships are left separate, the adjusted study count is a conservative upper bound with respect to unobserved publication dependence—not proof that every remaining evidence unit is independently recruited or generated. These counts are narrower than the report’s “human evidence source records” metric: syntheses and reviews remain publication-level evidence records, while the independence-adjusted figure above is restricted to primary human research. It is not an estimate of unique participants and is not an RCT-only count. Across all study classes in the matched public research-profile scope, the current topology retains 734 evidence units after explicitly proven dependence collapse.
Grade accounting
24 of 318 indexable ingredients (7.5%) currently remain Unassigned. This includes records where evidence varies by outcome, a single grade is not applicable, or the authored evidence signals cannot be reconciled honestly. These records are kept separate from both stronger grades and Avoid/Insufficient rather than being forced into either bucket.
| Category | Ingredients | Unassigned | Share unassigned |
|---|---|---|---|
| Uncategorized | 310 | 20 | 6.5% |
| Copper-binding tripeptide | 1 | 1 | 100% |
| Dual GIP/GLP-1 receptor agonist (prescription drug) | 1 | 1 | 100% |
| GLP-1 receptor agonist (prescription drug) | 1 | 1 | 100% |
| Melanocortin receptor agonist peptide | 1 | 1 | 100% |
Evidence-design mix
Primary human evidence includes trials, observational studies, and case reports. Evidence synthesis includes systematic reviews and meta-analyses and is shown separately so review-heavy categories are not mistaken for categories with a deep primary-human study base. “Preclinical” means mechanistic, animal, or in-vitro evidence. Categories with fewer than three structured relationships are omitted to reduce tiny-sample noise.
| Category | Evidence relationships | Primary human | Synthesis | Preclinical | Other / unclear |
|---|---|---|---|---|---|
| Uncategorized | 994 | 118 · 12% | 184 · 19% | 4 · 0% | 688 |
Learning context
Original data report on evidence grades, structured study/source records, independence-adjusted primary-human research counts, preclinical reliance,… Learning pages explain the reasoning layer behind the herb and compound library. They are designed to make mechanisms, evidence quality, safety tradeoffs, and product claims easier to interpret.
Use State of Supplement Evidence 2026 to build better questions before choosing a supplement: what outcome is being targeted, what mechanism is claimed, what human evidence exists, what dose was studied, and what risks could change the answer for a specific person?
Mechanistic plausibility is useful, but it should be weighed against trial design, safety history, product quality, and the possibility that a simpler intervention may be more appropriate.