Methodology
We show our work, always.
§0 · The premise
A trustworthy listing is one whose claims can be independently corroborated in under three seconds. Lighthouse fans out across public sources — registries, archives, professional networks, compensation databases, scam-report forums — and stitches the evidence into a single annotated report. Every claim cites its source. Every score is decomposable. No black box.
§1
Real
Does this company exist, and did it write this posting?
WHOIS lookupDomain age, registrant continuity, NS stability.
Registry matchSEC EDGAR, Companies House, state filings.
Recruiter graphLinkedIn tenure, prior roles, posting history.
Copy noveltyPosting body diffed against 9M historical listings.
§2
Active
Is the role open and being filled — or is it warehousing résumés?
Snapshot traceArchive.org first-seen, version count, reposts.
Repost diffIdentical copy across LinkedIn, Indeed, niche boards.
Community signalReddit / Blind / HN thread response rates.
Server hintsLast-Modified, ETag, careers-page edit cadence.
§3
Fair
Is the deal coherent — pay, equity, scope, process?
Comp percentileLevels.fyi banding by level, region, function.
Equity sanityCarta median ranges by stage and ownership.
Scope ratioResponsibilities-per-dollar against benchmark.
Process feesDetect $-for-training, $-for-equipment patterns.
§4
Credible
Does the company actually do anything?
Product surfaceLive signup, billing, status page, changelog.
Founder graphPre-company employment history, mutual ties.
Funding ledgerCrunchbase / press, investor verification.
Code activityGitHub org commit volume, last 30 days.
A note on bias
A 14-day-old domain is not automatically suspect; a 14-year-old domain is not automatically safe. Each signal carries a weight, each report shows the full evidence trail, and humans — you — make the final call. We don't blocklist companies. We surface what we found, and let you read it.