How we label our claims
Every claim in a Hylios brief carries a confidence label. The label is visible inline, in the share preview, and in the email digest. Honest uncertainty beats false confidence, labeled claims let you decide what to trust.
The four confidence labels
Fact stated explicitly in a cited article whose body names the company.
Example: Walgreens reported Q3 revenue of $36.7B (cited to Walgreens' earnings release).
How to read it: Treat as established fact. The reference is the article body.
Connection through a named, verifiable supplier, customer, regulator, or input from the company's SupplyContext. The article doesn't have to name the company, the connection does.
Example: McKesson CEO transition affects Walgreens because Walgreens' 2024 10-K names McKesson as its primary pharmaceutical distributor.
How to read it: Trust the connection; verify the magnitude through the cited article.
Connection through industry-, region-, or input-level exposure. The company is in the affected population but the specific link isn't documented per-company.
Example: New tariffs on imported electronics will raise front-of-store COGS, Walgreens sources consumer electronics from Asian suppliers per industry reports.
How to read it: Read as a directional signal. Magnitude is range-bound.
Plausible second- or third-order inference. The link is real but the magnitude or timing is uncertain. Hedging language is intentional.
Example: If consumer credit tightens, Walgreens' discretionary sales may decline alongside other retail.
How to read it: Conversation starter, not a forecast. Validate before acting.
Quantified impact estimates
When a claim affects a measurable outcome (gross margin, COGS, transit time, etc.) the brief includes a range-bound estimate. The range is honest about uncertainty, we publish a low and a high, not a single point.
- Intelligence: qualitative direction only (“higher”, “lower”, “more exposure”). No numbers.
- Intelligence Pro ($19.99/mo): range-bound estimates sourced from industry-average elasticities (BEA Input-Output tables, industry research, public filings). Labeled Likely.
- Twin: estimates derived from your twin’s actual cost history. Same UI, grounded in your data. Likely-tier claims promote to Strong.
What we don’t do
- Never fabricate facts. Every claim must trace to a cited article ID in the brief.
- Never mark a Likely or Speculative claim as Direct to make it look more authoritative.
- Never include Speculative claims in share previews, email subject lines, or outreach copy, only on the brief detail page.
- Never expose a customer’s twin data or private context to other customers. Each Twin-tier estimate is computed on that customer’s data alone.
Track record (live)
Once 90+ days of Likely/Speculative claims have aged, this section publishes hit-rate stats per confidence tier, the percentage of claims that validated against subsequent data. We publish even when the numbers are unflattering. Aggregated automatically from claims; not founder-curated.
Coming online ~90 days after public launch. Until then, treat Likely as range-bound projection, Speculative as scenario starter.