How we score: five assessment formats

Every verified TalentProof scorecard is built from transparent rules. Recruiters can chain formats in a multi-step workflow on one job. Scores inform judgment; they are not a hidden platform hire/no-hire cutoff.

Scoring principles

Consistent

Same rubrics and formulas for every candidate on a given role: no interviewer mood swings.

Transparent

Formats, pass bars, and confidence signals are documented here and on the scorecard.

Integrity-aware

Proctoring signals feed Session Confidence or validity context: for review, not silent discard alone.

Assessment formats

Locked labels: Q&A, Rapid fire, Hands-on, Conversation, Psychometric.

Q&A

Open-ended, role-tuned questions graded by AI against a fixed rubric (correctness, depth, relevance).

Scale: Skill score 0-1000 (shown as % on many surfaces)
Pass / use: Pass at 70% (700/1000)
Integrity: Full Session Confidence (camera, paste, tab, AI-text signals when proctored)

Rapid fire

Timed multiple-choice items generated from the JD: auto-scored with negative marking.

Scale: Score = (correct×1 − wrong×0.25 − fast-wrong×0.5) / n × 100
Pass / use: ≥70% overall and ≥40% of hard items correct
Integrity: Timed path: integrity via timing anomalies and hard-question gate (not a separate authenticity meter)

Hands-on

Practical task on the candidate’s own tools (code, PRD, UX by role). Session can be recorded for review.

Scale: Final 0-100 = functional (50) + quality (30) + completion (20)
Pass / use: ≥60% and work detected
Integrity: Integrity / no-work flags and recording context for reviewers

Conversation

Proctored skill dialogue with Alex (AI skills interviewer): not a recruiter phone screen.

Scale: Transcript → answers → same scorecard path as Q&A (0-1000 skill + Session Confidence)
Pass / use: Same performance bar as Q&A (≥70%)
Integrity: Same proctoring / authenticity pipeline as Q&A when enabled

Psychometric

Forced-choice pairs, situational judgment, and cognitive items across trait dimensions: work-style profile, not a skill screen.

Scale: Per-trait %; optional fit vs recruiter ideal profile; validity / consistency checks
Pass / use: No pass/fail skill gate: profile + validity for judgment
Integrity: Response-consistency / validity indicators (speed traps, patterns, contradictions)

Core scorecard metrics

What recruiters see on a verified TalentProof scorecard (format-dependent).

1. Performance

How well the candidate did on that step: Q&A / Conversation skill score, Rapid fire %, Hands-on composite, or Psychometric trait profile.

90-100
Expert

80-89
Advanced

70-79
Proficient

60-69
Developing

<60
Needs work

Letter-grade helpers may appear on some results views for readability. Screening behavior follows what you configure on each role. No platform-wide letter grades (A/B/C/D) or hidden AI cutoffs.

2. Session Confidence (0-100)

Starts at 100 on proctored Q&A / Conversation paths. Adjustments reflect session signals: context for reviewers, not a secret auto-reject.

Heavy paste / clipboard patterns

Larger deductions

Camera denied (when required)

−15

Screen share denied (when required)

−10

Tab / blur bursts

Capped deductions

Face missing / multiple faces

Per-event, capped

AI-text pattern notes

Context on scorecard

3. Job fit (when tied to a JD)

Resume ↔ role capability fit for prioritization and call prep. Job match scores and resume flags are advisory. They surface call prep and do not auto-reject candidates on their own.

Knockout questions and pre-screening criteria you set on a role can auto-filter applicants when auto-reject is enabled. Rules you control per job.

Hiring decisions & trust →

Overall score (when combined)

On Q&A-style scorecards, overall can combine performance, Session Confidence, and job fit for a single comparison number:

With job fit

Overall = (Perf × 0.6) + (Confidence × 0.2) + (Fit × 0.2)

Without job fit

Overall = (Perf × 0.7) + (Confidence × 0.3)

Multi-step workflows report per-step results; each step uses its own format rules above.

Privacy by design

PII is encrypted before storage. AI evaluation uses placeholders instead of real names. See Data Security.

See it in action

Preview a sample scorecard, or try building a role with any of the five formats.

View sample scorecard Try Assessment Builder