Applicants.io Applicants.io
Methodology

Company Ingestion Checklist

Company Ingestion Checklist

Goal

Create a company page that is:

  • Accurate
  • Legally safe
  • Machine-parsable
  • Easy for AI + Google to extract

Non-Negotiable Rules

  • Pattern language only. No accusations phrased as fact.
  • Every key claim must have an Evidence row.
  • Keep each H2 section under ~120 words, use bullets.
  • Do not speculate about intent. Document what is reported/observed.

Automated Extraction Pipeline

For processing Reddit JSON and other unstructured sources, use the 4-step extraction pipeline:

  1. Raw Intake: Extract candidates without interpretation
  2. Entity Normalization: Normalize to canonical names
  3. Pattern Scoring: Apply risk scoring dimensions
  4. Publication Gate: Apply thresholds for publication

See server/extraction/README.md for detailed documentation.

Key principle: The pipeline extracts candidates and detects patterns, but does not apply labels or make accusations. DevilCorp characteristics emerge through scoring, not declaration.

Inputs to Collect (Minimum Set)

A) Identity & Entities

  • Official name as listed on website or LinkedIn
  • Website URL (or "not found")
  • Known aliases / DBAs (from job posts, LinkedIn, registries)
  • Operating regions (cities/states mentioned in postings)
  • Parent / network association hints (Smart Circle, Synaptic, etc.) if publicly referenced

B) Role & Recruiting Surface

  • 3–5 job titles repeatedly used
  • Common compensation language (commission-only, "uncapped", "OTE", etc.)
  • Interview format signals (group interview, same-day offer, "orientation")
  • Training phase signals (unpaid training, "shadowing", "student mentality")

C) Report Patterns (What people report)

Collect 6–12 evidence items across sources if possible. Prioritize:

  • Glassdoor reviews (time-separated)
  • Indeed reviews
  • Reddit threads (company + city + "legit/scam")
  • BBB (if present)
  • LinkedIn company page + employee counts / title patterns
  • Any public news or court docs (only if solid)

Evidence Row Rules

Each Evidence item must include:

  • Claim: one sentence, specific, non-interpretive
  • Source: platform + URL
  • Date captured: YYYY-MM-DD
  • Excerpt: short quote or paraphrase (keep short)
  • Credibility score (0.0–1.0) guideline:
    • 0.9: court doc, official filing, direct company page
    • 0.7: review platforms with multiple corroborations
    • 0.5: single anecdote without corroboration
    • 0.3: vague forum claims, no detail, no repeats

Risk Subscore Scoring Guidelines (0–100 Each)

Wage/Payment (W)

  • 0–20: no wage/payment complaints, clear pay structure
  • 21–60: mixed reports, ambiguity, missed commissions alleged
  • 61–100: repeated reports of nonpayment / bait-and-switch pay

Misrepresentation (M)

  • 0–20: job matches listing language
  • 21–60: repeated "not what was advertised" claims
  • 61–100: consistent pattern of misleading role descriptions

Churn/Pressure (C)

  • 0–20: normal pressure
  • 21–60: repeated "high pressure, long hours"
  • 61–100: consistent churn, "recruit everyone", burnout patterns

Recruitment Funnel (R)

  • 0–20: standard interviews
  • 21–60: group interviews, same-day offers, vague roles
  • 61–100: mass interviewing, rapid onboarding, unpaid trial phases

Legal Threats (L)

  • 0–20: none
  • 21–60: reports of threats without documentation
  • 61–100: documented cease-and-desist patterns or intimidation claims across time

Identity Obfuscation (O)

  • 0–20: transparent org identity
  • 21–60: frequent DBA usage, unclear parentage
  • 61–100: repeated rebrands, shell-like footprint, inconsistent naming

Review Pattern Anomaly (A)

  • 0–20: normal distribution
  • 21–60: suspicious clustering, repeated phrasing
  • 61–100: heavy astroturf indicators or sudden rating spikes without explanation

Final Output Requirements

Before publishing, confirm the page includes:

  • ✅ Overview (what it is)
  • ✅ Risk Summary (score + confidence)
  • ✅ Patterns (bulleted)
  • ✅ What to do if applying (action steps)
  • ✅ Evidence table with at least 6 items
  • ✅ FAQ with 3–6 questions tied to candidate intent

Risk Score Calculation

Weights (sum = 1.00)

  • wW = 0.20 (Wage/Payment)
  • wM = 0.18 (Misrepresentation)
  • wC = 0.14 (Churn/Pressure)
  • wR = 0.14 (Recruitment Funnel)
  • wL = 0.10 (Legal Threats)
  • wO = 0.12 (Identity Obfuscation)
  • wA = 0.12 (Review Pattern Anomaly)

Raw Risk Calculation

RiskRaw = (wW*W + wM*M + wC*C + wR*R + wL*L + wO*O + wA*A)

Confidence Calculation

For each evidence item i, define credibility ci in 0,1. Let N = number of evidence items.

Confidence = clamp( (sum(ci)/max(N,1)) * log1p(N)/log(1+10), 0, 1 )

Rules:

  • If N < 3, cap Confidence at 0.45
  • If N = 0, RiskFinal must be 0 and page must show "Insufficient evidence"

Final Risk (Confidence-Adjusted)

RiskFinal = round( RiskRaw * (0.60 + 0.40*Confidence) )

Risk Bands

  • 0–19: Low
  • 20–39: Guarded
  • 40–59: Elevated
  • 60–79: High
  • 80–100: Severe

Template Reference

Use content/companies/_company-template.md as the base structure.

Ensure all required frontmatter fields are populated:

  • title, description, slug
  • industry, website, hq, operating_regions
  • known_aliases, last_updated
  • risk_score (total, confidence, subscores)
  • evidence (array with all required fields)
  • faqs (array)

Quality Checklist

Before publishing:

  • All evidence items have URLs and dates
  • Credibility scores are justified
  • Subscores align with evidence
  • Risk band language explains the score
  • No speculative language
  • All claims backed by evidence
  • FAQ answers are factual and concise
  • Internal links to related guides/networks

Data Sources