Where SpecStream looks
The places SpecStream reads, how it finds one record several ways, and how every answer is checked.
The places it looks
Where SpecStream looks depends on what you asked for. A clinical trial lives in a registry; a company’s open sales roles live on its own careers page and a job board. Most requests use more than one.
| Where | Good for | For example |
|---|---|---|
| Official records and feeds | Facts the publisher answers for, in the fields they publish them in. | Companies House, SEC EDGAR, GLEIF, ClinicalTrials.gov, Crossref, SAM.gov, Grants.gov, npm, RSS |
| Lists, rankings and directories | Read whole, forty entries at a time, each entry judged on the line that names it, with next pages followed. | Industry rankings, member directories, award lists, roundups |
| Public websites | What a business, creator, or company says about itself. | Contact pages, careers pages, product pages, About pages |
| Pages that only load in a browser | Sites that build themselves with JavaScript. SpecStream renders them first, so nothing is missed because it hadn’t loaded yet. | Storefronts, job boards, map listings |
| Structured data | Facts already sorted into labeled fields. | Schema.org markup, OpenStreetMap, Overture Maps, Wikidata |
| Documents | Reports, manuals, and filings, read in full, including scanned pages. | PDF, Word, Excel, PowerPoint, EPUB, OpenDocument, scanned pages |
| Archives | What a page said before it changed. Always labeled historical. | Common Crawl |
One record, several ways in
SpecStream tries several searches across different sources and follows useful references back to the original publisher. It considers the candidates it finds before selecting the results, rather than treating the first search results as the finished list.
A company can turn up in a business registry, on a job board, and on its own website. SpecStream follows more than one path to find candidates, then merges what it found into one record, matched on stable identifiers such as the website or the registry number. A record found twice is still one result, and you pay for it once.
Facts are checked where they’re published. The address might come from the shop’s contact page and the opening hours from a map listing, and each keeps its own source. When two sources disagree, you see both. A cross-check doesn’t quietly pick one.
What counts as evidence
A result counts when every condition in your request is met, each with its source. Here’s what SpecStream reads as evidence, and how each answer is checked.
Results with a quote for every condition are marked Verified and listed first.
Evidence
- A fact on the page or record that states it, with that page and the time it was seen
- A field in an official record or feed
- A passage in a document, quoted
- Other pages that name the result, such as filings, listings, profiles and news
- An archived copy, for what was true on the day it was captured
Checked
- Every condition gets an answer and a one-line reason
- Numbers, dates and places are compared by code against the published value
- Every quote is confirmed on the page it cites
- When a source contradicts the model, the source wins
- Archived copies keep the day they were captured
Each fact keeps its source and the time it was observed, so you can open it and check. Read the evidence
Reading documents
SpecStream reads the file itself: PDF, Word, Excel, PowerPoint, EPUB, and OpenDocument files. Scanned pages are read with optical character recognition (OCR). When a passage was hard to read, the fact is marked, so you know to check the scan. The quoted passage comes back with the fact.
Old copies, clearly labeled
When a page has changed or disappeared, an archived copy can still answer a question about the past. SpecStream uses archives such as Common Crawl for exactly that, and every fact from one is a historical copy with its capture date. For what’s true now, only a current source counts.
When a source doesn’t answer
Sources time out, block visitors, and move. SpecStream retries within limits and tries another route where one exists. Retries and fallbacks never raise the price. A requirement it still couldn’t check stays unresolved. It’s neither a yes nor a no, and it isn’t charged as a match.
SpecStream reads public sources. It doesn’t sign in as you, get past paywalls, or read private groups.
Sources by record type
| Record type | Sources |
|---|---|
| Local businesses | Google Maps, OpenStreetMap, Overture Maps, Yelp and more |
| Companies | Companies House, French, Norwegian and Finnish registers, GLEIF, SEC EDGAR and more |
| People | LinkedIn, Personal sites, Company team pages, Press releases and more |
| Creators | YouTube, TikTok, Twitch, Vimeo and more |
| Apps | App Store and Public app listings |
| Job postings | Greenhouse, Lever, Ashby, Workable and more |
| Papers | Crossref, PubMed, Europe PMC, PubMed Central and more |
| Tenders | SAM.gov, Open Contracting (OCDS) publishers, and Published procurement notices |
| Grants | Grants.gov and Published funding opportunities |
| Repositories | GitHub, GitLab, and Codeberg |
| Packages | npm, PyPI, crates.io, Maven and more |
| Datasets | DataCite and Crossref |
| Clinical trials | ClinicalTrials.gov |
| Regulations | Federal Register, eCFR, and Legislation.gov.uk |
| Vulnerabilities | NVD |
| Books | Open Library |
| Podcasts | Apple Podcasts and Publisher RSS feeds |
| Pages and documents | Public websites, News sites, Forums, Pages that only load in a browser and more |
| Encyclopedia entries | Wikidata and Wikipedia |
Each record type’s page has the full list, the details you can ask for, and requests to start from. See every kind of record