Scrape Trust Score — know what your data is worth before you act on it

Your scraper doesn't lie. Your data does. Every dataset you pull is a mix of the real, the stale, and the planted — dead records, bought reviews, disposable emails, and honeypot rows dropped in to poison anyone scraping. Feed that to a sales team and you waste a week. Feed it to an AI agent that acts automatically, and you've automated a bad decision at scale.

Scrape Trust Score is the quality gate between "scraped" and "acted on." Point it at any dataset and every record comes back with a trust score and the evidence behind it — so you keep the gold and quarantine the garbage.

One field to wire it up: pass a datasetId from any actor's run. Fields are auto-detected. That's it.

The two questions nobody else answers

🕒 "When was this last real?" — Recency

Stale data is a silent killer — the lead moved, the business closed, the page is gone. Trust Score gives every record a recency score + lastSeen, built from record timestamps, source freshness signals, and a live liveness re-check. No timestamp exposed? We flag freshnessConfidence: low and say so — we never invent a date.

🪤 "Was this planted to poison me?" — Decoy / honeypot risk

Sites seed fake rows to trap scrapers, and AI now floods the web with plausible fakes. Trust Score gives every record a decoyRisk 0–100 with a list of evidence — non-resolving domains, dead links, disposable/no-MX emails, fake phone ranges, placeholder text, duplicate-farm clusters. It's a calibrated score, not a coin-flip verdict: a legit-but-unusual record (say, an email-only domain with no website) lands mid-risk with reasons, not a false "fake!" — so you can trust the flags you do get.

What you get

Every record is returned unchanged, plus a _trust block:

{
  "...your original fields...": "...",
  "_trust": {
    "trustScore": 88,
    "recencyScore": 74, "lastSeen": "2026-08-30T...", "freshnessConfidence": "ok",
    "decoyRisk": 10, "corroborationChecked": true,
    "flags": ["role-email"],
    "evidence": ["source URL live (200): https://acme.com", "email domain has working MX: acme.com"]
  }
}

Plus a dataset-level SUMMARY in the key-value store: junkRate, decoySuspectPct, lowFreshnessPct, medianTrust, medianRecency, and a flag breakdown — the one number that tells you how much of this scrape you can trust.

Chain it onto anything (zero config)

• After any Apify actor: run your scrape, then run Trust Score with that run's datasetId. Fields auto-detect — emails, URLs, phones, timestamps — no mapping.
• In one Zap / Make / n8n flow: Scraper → Trust Score → Google Sheets, and filter to trustScore >= 70.
• For AI agents (MCP): the agent scores its own scraped data and gates on decoyRisk before it acts — exposed via the Apify MCP server.

Pricing that matches the value

Cheap enough to sprinkle on every scrape: a tiny per-record fee for the fast signals, and a deep-verify event that only fires when you turn on live MX + URL corroboration — so you pay for deep verification only on the rows that need it.

🔗 Works with

• Upstream: Data Source Finder finds where the data lives; a scraper pulls it; then you point Trust Score at the result.
• Pairs with Contact Email Finder and YouTube Email Finder — enrich a list, then verify it here before outreach.

Honest about the method

Trust Score uses public, generic verification heuristics — MX and liveness checks, disposable-domain and fake-pattern detection, duplicate clustering, timestamp analysis. It scores risk with evidence; it does not claim to read intent or prove a row was maliciously planted. Use the flags and evidence to decide — that's the point.