How Oink scores job matches without hiding the tradeoffs

A useful job-match score should help you decide what to inspect first. It should not pretend to know whether a recruiter will call, and it should not hide the evidence behind a confident number. Oink’s matching pipeline is designed around that distinction: rank a posting against the job seeker’s current brief, show concrete fit and concern evidence, and leave the application decision to the person.

The profile is more than a resume

Oink separates what you can do from what you want to do. A resume provides evidence of roles, skills, domains, and seniority. The current preferences provide the target: desired role, must-have or avoid skills, work arrangement, location and eligibility, salary floor, languages, and dealbreakers.

That separation prevents a common matching failure. If your last role was in ad-tech but your current brief explicitly excludes ad-tech, career history should not silently override the new constraint. Likewise, absence of a skill in a short resume is not automatically proof that you lack it. The pipeline retains missing or uncertain evidence instead of converting every blank into a confident pass or fail.

If your brief is still vague, use this practical guide to describing the job you want. Better constraints make the reasons more useful; keyword stuffing does not.

A two-stage relevance decision

The production path uses a lower-cost first pass to reduce an oversized pool, followed by more detailed per-job scoring for candidates that survive. Source-grounded checks can identify explicit incompatibilities—such as a clearly disallowed location or work mode—without asking a model to reinterpret the facts. Nuanced cases continue to contextual evaluation.

This design does not mean every fetched posting is guaranteed to receive identical AI enrichment. Providers can fail, source fields can be missing, and some records may proceed with reduced information. A good result makes that limitation visible as an unknown or concern instead of manufacturing details to complete a template.

What the 0–5 score means

Oink stores match scores on a 0–5 scale:

• 0 is a rejection decision for an explicit or very strong incompatibility and is not a normal delivered recommendation.
• 1–2 indicate weak alignment or important conflicts.
• 3 is a plausible stretch: worth inspecting when the reasons make sense to you.
• 4 is a strong “apply today” candidate against the stated brief.
• 5 aligns with the target requirement by requirement based on the evidence available.

These labels rank relevance inside your search. They are not calibrated probabilities. A 5 does not mean a five-in-six interview chance, and a 2 does not prove you are incapable of doing the job. The employer has information and preferences Oink cannot observe; the posting may also be incomplete or stale.

Read the reasons before trusting the number

Each useful verdict should identify specific support and specific gaps: matching role or stack, seniority evidence, work mode, location, salary, visa or language signals when present, plus missing facts that matter. The original posting text and source fields are the evidence boundary. AI-enriched fields can help organize the record, but they should not erase provenance or contradict explicit source text.

The fastest quality check is simple: could you explain the score in one sentence without repeating the score itself? “Strong backend fit, EU-remote, salary above floor, but less MLOps ownership than requested” is actionable. “Great match for you” is not.

Our guide for people who applied to hundreds of jobs without a reply shows how the same evidence-first rubric works manually. The remote eligibility guide explores one of the constraints most likely to be flattened by ordinary keyword search.

Feedback changes evidence, not reality

“More like this” and “Not for me” are additional preference signals. Oink records feedback so later matching can incorporate it, and enough distinct feedback can trigger a profile rebuild. One tap does not instantly retrain a private model, and it cannot change what a posting says. Feedback is most useful when it reveals a consistent preference that was missing or poorly expressed in the original brief.

Marking a role as applied is separate. It updates the application tracker; Oink does not submit the employer’s form. Delivery is also separate from scoring: the strongest eligible results are ranked against a configurable score floor and batch cap before they become Telegram cards.

Use the score as a queue, not a verdict

Open the high-scoring roles first. Read both the strengths and the concerns. Confirm the listing at the original source, then decide whether the unknowns are acceptable and whether the application deserves tailoring. If a reason is wrong, correct the underlying preference or provide feedback rather than treating the number as permanent truth.

That is what explained matching is for: less time reading obvious mismatches, more attention for roles whose evidence survives scrutiny.

Continue with how Oink works end to end or see how those reasons appear in Telegram job alerts.