Methodology
Rubric v0.3 (proposed). Weights may change before ratings are published.
Principles
- Firm-level only.
- Cited facts only (public or badged user-submitted).
- Deterministic (the LLM never scores).
- Versioned.
- Unknown is neutral.
- Every point links to its fact(s) and source(s).
Rating = 50 + positive signals − negative signals, clamped 0–100.
Weights live in config and the site owner may tune them.
The AI only extracts cited facts; it never scores.
Positive signals
| Signal | Points |
|---|---|
| Domain age ≥ 10 y / 5–10 / 2–5 | +12 / +8 / +4 |
| Active state registration, formed ≥ 5 y | +8 |
| Verified commercial address (not CMRA/virtual/residential) | +8 |
| Official government/vendor record (SAM.gov active, SBA cert, state contract or license) | +12 |
| Third-party public evidence of work with the named client | +6 |
| Email domain matches website domain | +4 |
| Recognized trade/credential listing (e.g. ASA, NAPS, Joint Commission HCSS), typed third-party | +4 |
| Firm claimed profile and responds to disputes | +2 |
Negative signals
| Code | Signal | Origin | Severity | Points |
|---|---|---|---|---|
RF-IMPERSONATION | Lookalike domain or another firm's branding | public | Critical | −40 |
RF-REGULATOR | Public regulator/court/AG action for recruiting fraud | public | Critical | −40 |
RF-FEE | Asks the candidate to pay | public or user-submitted | Critical | −40 |
RF-SSN | Requests SSN digits/DOB/driver's license before a confirmed client interview | public or user-submitted | Critical | −35 |
RF-FAKEPROFILE | Coaches fake or embellished profiles/experience | public or user-submitted | Critical | −30 |
RF-NEWDOMAIN | Domain < 12 months | public | High | −15 |
RF-RESIDENTIAL | Only residential-type addresses | public | High | −12 |
RF-MAILBOX | Only CMRA/PO box/virtual office addresses | public | Medium | −8 |
RF-DOMAINMISMATCH | Email domain has no site, or differs from the website domain | public or user-submitted | Medium | −8 |
RF-NAMEMISMATCH | Brand vs legal name mismatch, unexplained | public or user-submitted | Medium | −6 |
RF-NOREG | No business registration found | public | Medium | −6 |
RF-SHAREDADDR-RISK | Shares an address/phone with a High-risk firm | public | Medium | −8 |
RF-REQMISMATCH | Posting contradicts the official req (duration/rate/location) | public or user-submitted | Low | −4 |
RF-BCCBLAST | Mass email with an empty To:, no specific req | user-submitted | Low | −3 |
RF-PRESSURE | Urgent "call me now" pressure | user-submitted | Low | −3 |
User-submitted flag rules
User-submitted flag rules (proposal).
- Detected by rule-based detectors on redacted text, then the evidence hash is de-duplicated (the same email submitted twice counts once).
- 1 independent submitter: non-Critical flags publish with the badge after bot approval. A Critical flag is held unpublished for the site owner's decision. While held, it has no effect on the rating, label, or rankings. Once approved it caps the label at Medium–High (it doesn't trigger High by itself).
- ≥ 2 independent submitters (different accounts, different evidence hashes, each account ≥ 7 days old), or corroborating public evidence: full points and Critical → High.
- A flag counts once, at its highest instance.
Informational, not scored
| Item | Points | Detail |
|---|---|---|
| Shared-recruiter firm link (SHARED_RECRUITER, Phase 2) | 0 | Requires ≥ 1 system-verified source. Legitimate recruiters often change firms or work for primes and subs. |
| Submission volume (per domain/email) | 0 | 0 points. |
| X-derived facts | never scored | Facts derived from X are never scored. |
| LinkedIn public-page citations | low weight | Never the sole basis for a negative flag. |
| Candidate outcomes (opt-in aggregates) | 0 | Informational only (final). Deferred: a possible "strong outcomes" badge after data evaluation, not v1. |
| Stale facts | excluded until re-verified | A source that is no longer valid (stale after 2 consecutive failed refreshes) is excluded from the rating until re-verified, and removed after 180 days stale. Negative flags with archived evidence never auto-stale (bot review instead). |
LinkedIn public pages may be cited and are not fetched by this site. LinkedIn citations are low weight and are never the sole basis for a negative flag. X-derived facts are never scored.
Labels and rankings
| Rating | Label |
|---|---|
| ≥ 75 | Low risk |
| 60–74 | Low–Medium |
| 40–59 | Medium |
| 25–39 | Medium–High |
| < 25 or any qualifying Critical flag | High risk |
| < 2 sourced facts | Insufficient data |
Labels are risk assessments per the published rubric.
Labels are worded as "risk assessment per published rubric", never "scam" or "fraud".
- Highest-Risk Firms: High label, evidence ≤ 18 months, no dispute pending past its SLA. Firms only.
- Top 10 Most Trusted: highest ratings with ≥ 4 independent sourced positive signals and no Medium+ flags in 24 months.