About ALETHEIA
Methodology, sources, and the team behind the API
ALETHEIA is a regulatory-grade chemical safety API. It synthesizes hazard classifications from major government agencies and authoritative databases worldwide into a single statistical reference layer — with the disagreement between sources preserved, not hidden.
The platform is built and operated by Holistic Quality LLC. Coverage today spans 1,886 distinct compounds, 1,287 consumer products across eight product tiers, and 959 raw materials. Every response includes the underlying source positions, per-source weights, and confidence intervals so downstream systems can audit the math.
The problem we solve
Regulatory bodies routinely disagree on the same compound. IARC may classify glyphosate as “probably carcinogenic” while EPA and EFSA reach the opposite conclusion. All three are working from solid methodology and extensive evidence — they simply weigh it differently.
Most safety datasets collapse that disagreement to a single “score,” quietly making editorial decisions about which agencies to trust. ALETHEIA does the opposite: it surfaces the distribution and lets the consumer of the data make the call. That makes the API a defensible input for product safety reviews, supplier screening, and consumer-facing tools where the audit trail matters.
How synthesis works
1. Distribution, not a single point
Each compound query returns the full statistical shape across sources — mean, standard deviation, 95% confidence interval, and the underlying per-source positions. Where consensus exists we say so; where it doesn’t, we expose the disagreement.
2. Mathematical weighting, not editorial weighting
Source weights are computed from three transparent inputs, not chosen by hand:
- Recency (exponential decay, five-year half-life)
- Confidence (sample size, methodology rigor, peer review)
- Evidence hierarchy (case report → cohort → RCT → meta-analysis)
Every weight is included in the response payload.
3. Defensible language
The API never asserts “safe.” It returns context-scoped risk profiles (“low risk at typical adult dermal exposure,” for example), each citing the sources behind it, with explicit gaps where data is thin.
4. Auditable methodology
The synthesis engine is weighted statistics in a five-dimensional space. No neural networks, no proprietary scoring, no opaque pipelines. The dimensions and thresholds are documented below; the code is reviewable.
The five-dimensional coordinate system
Every regulatory classification is projected into five normalized dimensions, each bounded to [0, 1]:
Live demo · prototype
Three compounds, two contexts. Same five lenses, same coordinate system — the shape changes with the (compound, context) pair. Click any cell to open the live Looking Glass at that state.
Read across the rows: Theobromine transforms (mild → severe), Glyphosate stays contested in both contexts (dashed borders on Magnitude), Lead reads severe regardless of context. That's the methodology — same coordinate system, different epistemological situation, different shape.
Magnitude
Severity of the hazard. Each agency’s classification system is mapped to a numeric scale — IARC Group 2A → 0.750, EPA “not likely” → 0.150, and so on — with the mapping table published.
Confidence
How well-supported the classification is, based on sample size, methodology rigor, peer review status, and independent replication. High confidence is not a claim of correctness; it is a claim that the position is well-supported by available evidence.
Recency
Time-weighted relevance with a five-year half-life. A study from a decade ago contributes roughly a quarter of the weight of a current one. Older data is included but discounted.
Methodology
Study design quality, drawn from the evidence-based-medicine hierarchy: anecdote (0.10) → case report (0.25) → cohort (0.55) → RCT (0.85) → meta-analysis (0.95).
Specificity
How closely the source population matches the query context. A study on mammals applied to a dog-context query is weighted lower than a study on the same breed; broad extrapolation carries a cost in the final synthesis.
Consensus detection
Consensus is reported as true when all of the following
hold:
- Standard deviation across source positions is below 0.15
- Maximum pairwise distance between any two sources is below 0.25
- At least two independent sources are available
Consensus strength is reported on a 0–1 scale (1.0 = perfect agreement). When consensus does not exist, the API does not synthesize a false majority — it returns the actual distribution.
Primary data sources
Hazard classifications and supporting evidence are drawn from authoritative regulatory bodies and reference databases:
The underlying compound, material, and product corpus is maintained in the HQ Safety Database, which is open for inspection.
Engineering and operations
Who builds this
ALETHEIA is operated by Holistic Quality LLC, founded by Levi Robey. The company builds safety and quality infrastructure with the explicit goal of being usable by technical teams that need an audit trail, not just a marketing score.
Development is conducted with extensive human-AI collaboration under the internal codename “Hexad.” The methodology, the source list, the weighting rules, and the per-call response payloads are all designed so that an external reviewer can verify the math without trusting the operator.
The name “ALETHEIA” comes from the Greek ἀλήθεια — unconcealment. The thesis the project is built on: the full shape of what is known is more useful, and more honest, than a single number.
Get started
Evaluating ALETHEIA for a commercial deployment, a regulated workflow, or a custom integration? Reach the team directly at levi@holisticquality.io. Bug reports and security disclosures: safety@holisticquality.io (see also our security.txt).