Evidence-based Job Fit Scoring
Evidence-based job fit scoring should answer two questions together: “How strong is this opportunity for me?” and “What facts support that conclusion?” JobCtrl keeps the score, requirement assessments, profile evidence, confidence, blockers, and policy version in one reviewable decision record.
A Score Is A Triage Tool, Not A Verdict
JobCtrl scores jobs for the person conducting the search. It is not an employer-side candidate-ranking system, and it should not be used to decide who gets hired.
The 1–10 score helps a job seeker sort a large set of openings and decide where to spend attention. It does not prove that an employer will interview you. A strong result can still contain a hard blocker, uncertain evidence, or a requirement you interpret differently. A weak result can reflect an incomplete Candidate Profile rather than a poor real-world fit.
That is why the Job Detail workspace presents the evidence and correction history beside the score instead of treating the number as self-explanatory.
Start With Two Separate Sources Of Truth
The scoring boundary deliberately separates employer claims from candidate claims:
- the captured posting and accepted employer analysis own requirements, priorities, and evidence about the role;
- the versioned Candidate Profile owns experience, skills, achievements, preferences, and evidence about the job seeker.
The scorer may classify how those sources relate. It may not turn wording from the posting into a new fact about the candidate. An evidence link for a match must point back to profile data; if that reference cannot be resolved, JobCtrl labels it unavailable rather than displaying it as proof.
Read Candidate Profile for candidate evidence ownership and Discovery for the canonical employer analysis.
Score Requirements Before Resolving The Number
When an accepted employer analysis contains explicit requirements, JobCtrl asks the scorer for a structured assessment of each one. A row records the requirement identity, importance, posting evidence, fit classification, and supporting Candidate Profile evidence.
The useful unit is the row:
| Requirement result | Meaning for review |
|---|---|
| Direct or strong match | Profile evidence substantially supports the requirement |
| Transferable evidence | The profile supports adjacent experience, but not the exact claim |
| Missing | The current profile does not contain supporting evidence |
| Blocked | A hard requirement is not met and constrains eligibility |
| Not assessed | The decision record does not contain a usable classification |
Versioned deterministic code then turns those rows into the saved score. More important requirements contribute more, matched evidence receives more credit than transferable evidence, and blocked requirements constrain the final result. The current weights, rounding rules, compatibility path, and worked example live in the canonical Scoring guide.
This division matters: a model classifies structured evidence, while a named policy resolves the number. The persisted score is not a free-form model opinion.
Score, Confidence, And Eligibility Are Different
Collapsing every concern into one number makes a score look simpler than it is. JobCtrl keeps three concepts distinct:
- Fit score summarizes the policy-resolved relationship between the role requirements and current profile evidence.
- Confidence tells you how much scrutiny the evidence needs. It is a review signal, not a hidden points multiplier.
- Eligibility decides whether downstream work such as automatic material generation may proceed. Hard blockers and the live minimum-fit threshold can affect eligibility without rewriting the saved score.
Changing the minimum-fit threshold therefore does not recalculate old scores. It changes which existing decisions are eligible for later stages. Likewise, a low-confidence score is not automatically low fit, and a high-confidence score is not an application guarantee.
Corrections Create History
If the evidence is wrong or your judgment differs, JobCtrl offers two separate actions:
- Correct the score to record your reviewed decision and rationale. This creates a new version and a calibration anchor.
- Re-score the job to run the current policy against the current canonical inputs. This creates a new model-derived version.
Neither action silently edits the old record. A later policy can mark older scores stale, but adopting it remains deliberate. The history lets you see whether a change came from profile evidence, employer analysis, a policy version, model execution, or human judgment.
The Evidence Map provides the reverse view: start from a profile achievement or skill and inspect where scoring and generated materials used it.
What The Audit Trail Can And Cannot Prove
The current requirement-led resolver is deterministic over the structured response it accepts. The parser validates shapes and requires evidence identifiers for matched rows, but it does not yet prove that every returned identifier exists in the saved profile or reconcile every returned requirement field against the accepted analysis before resolving the score.
That limit is visible in the product contract. An unresolved reference is shown as unavailable, and its storage key remains under technical details. You should not treat it as supporting evidence merely because the numeric score was saved.
Evidence-based scoring improves auditability; it does not eliminate model error, incomplete profiles, ambiguous postings, or human disagreement. Inspect high-value jobs and correct the record when the evidence is weak.
Review Fit In JobCtrl
Use the synthetic live demo to open a Job Detail workspace and inspect requirement rows without connecting a provider. In a local installation, start with a reviewed Candidate Profile and a bounded Discover run, then filter the Jobs workspace by score or fit band.
For exact behavior, read Scoring and the deeper Scoring Policy.
Related reading: