jd-decoder

JD Decoder

Treat every posting as a probabilistic signal, not a promise. Some listings are stale, evergreen, exploratory, duplicated, or no longer funded, but published estimates are not interchangeable and should not be turned into one universal “ghost job” rate. The practical takeaway is to inspect the specific posting and company, then scale effort to the evidence.

Process

Step 1: Score ghost risk before anything else

One point each (0-6):

1. Posted or reposted for 45+ days with no edits
2. No named team, manager, or concrete problem; responsibilities read as template
3. Not listed on the company's own careers page (job-board only)
4. Company had layoffs in the last 6 months in this function
5. No recruiter activity visible (no LinkedIn posts about the role, no outreach happening)
6. Extremely wide or missing comp band with no explanation

0-1: real, tailor deeply. 2-3: probably real, tailor moderately and verify. 4+: verify before investing; offer to draft a short note to the likely hiring manager asking "is this role actively interviewing?" That question is legitimate, cheap, and converts uncertainty into information. A repost alone is NOT proof of fake: budgets change, finalists fall through, some roles are evergreen.

Step 2: Classify every requirement

Four piles: must-have (early, repeated, connected to responsibilities), signal (tells you the team's stack and pain), nice-to-have (long comma lists; candidates matching 100% of a wishlist do not exist and the hiring manager knows it), compliance filler. Tailor only to must-haves and signals.

Step 3: Translate loaded language into diligence questions, not verdicts

Posting-language studies are weak evidence of anything, so convert phrases into interview questions instead of conclusions:

• "Fast-paced" → "What changed in the last two quarters that created this role?"
• "Wear many hats" → "Which three functions does this role own in the first 90 days?"
• "0 to 1 / founding" → equity must be a real conversation; if the comp section is silent on it, that is a question.
• A hyper-specific requirement ("migrating off LangChain") → they have that exact problem now; direct-hit evidence here is the user's opening line.
• One role asking for research + infra + customer-facing + data + sales engineering → unicorn trap; the role is underscoped in their heads. Generate clarifying questions and lower the tailoring investment.