Questions and scoring

Question library, evaluation guides, AI authenticity and video integrity, and dynamic scoring weights across hiring evidence.

InsightHire's scoring stack connects questions, rubrics, and AI analysis across async video, assessments, and resume qualification. This guide covers the question library, evaluation guides, integrity signals, and how weights combine into ranking.

Key routes

Question library

Dashboard → Questions (/dashboard/questions) is the org-wide reusable library for async video, assessments, and interview templates.

Question attributes

Filter by skill, type, and difficulty when building journeys or assessments.

Create and edit

• Browse library → select question → /dashboard/questions/[id]
• Edit text, metadata, and scoring configuration
• Changes propagate to new sessions; in-flight candidates keep prior wording per journey versioning rules

Interview-specific questions also live at /dashboard/interviews/questions for template building — prefer the org library for content shared across async and live.

Evaluation guides and scoring criteria

AI scoring requires telling the model what "good" looks like. Configure on /dashboard/questions/[id]:

Scoring config types

AI rubric generation

Click Generate scoring criteria to run the AI pipeline:

1. Analyze question text
2. Identify evaluation dimensions
3. Generate weighted rubric with performance levels
4. Save to question record

Review and edit generated rubrics before relying on them for high-stakes gates.

Ideal response

Optional ideal response field provides an exemplar answer for semantic similarity scoring — especially useful for technical short-answer prompts.

Multiple choice

Configure options with correct flags for deterministic scoring — no AI required.

AI authenticity analysis

InsightHire evaluates whether responses appear genuine and candidate-authored, especially for text and video.

Text authenticity

For assessment free-text items, authenticity_analysis records may include:

Authenticity appears on assessment result pages alongside content score. Low authenticity does not auto-reject — it surfaces for recruiter review.

Enable org-level authenticity checks in organization settings where available (authenticityCheck).

Video integrity

Async video responses pass through integrity checks:

Org-wide trends: Analytics → Video integrity (/dashboard/analytics/video-integrity).

Individual flags appear on journey reports via trust summaries and video analysis sections.

Dynamic scoring weights

Composite ranking combines multiple evidence layers with configurable weights.

Layer 1 — Question / item score

Each scorable question produces an item score via its rubric. Item scores roll up to:

• Video node average
• Assessment category average

Layer 2 — Category weights (assessments)

On /dashboard/assessments/[id]/edit, assign category weights:

assessmentOverall = Σ (categoryScore × categoryWeight) / Σ weights

Null or zero-weight categories display but do not affect overall.

Layer 3 — Journey overall score

Scorable journey nodes (video, assessment, culture quiz) contribute to journeySession.overallScore per position/org node weight configuration.

Layer 4 — Resume qualification

Resume criteria on the position (/dashboard/positions/[id]/edit) produce qualification score — parsed fields vs requirements with per-criterion weights.

Layer 5 — Culture blend (flag: culture_fit_scoring)

When enabled:

finalAverage = (1 - w_culture) × skillComposite + w_culture × cultureFitScore

See Culture fit scoring for profile configuration.

Layer 6 — Review queue display

The review queue badge prefers journeySession.overallScore when journey status is COMPLETED, else falls back to application.score.

Reviewer scores

Recruiters can assign manual reviewer scores (1–5 stars per category) on the position pipeline. Reviewer composite is the average across categories — distinct from AI score but visible alongside for calibration.

Scoring workflow diagram

flowchart TB
  Q[Question + rubric] --> V[Video node score]
  Q --> A[Assessment category scores]
  A --> AW[Category weights]
  AW --> AO[Assessment overall]
  V --> JO[Journey overall score]
  AO --> JO
  R[Resume criteria] --> QS[Qualification score]
  JO --> SK[Skill composite]
  QS --> SK
  CQ[Culture quiz] --> CF[Culture fit score]
  SK --> BL[Blended average]
  CF --> BL
  BL --> RQ[Review queue ranking]

Calibration tips

• Generate rubrics, then edit — AI drafts save time; humans calibrate thresholds
• Spot-check first 20 candidates before enabling score-gate auto-routing
• Compare AI vs reviewer scores — Analytics → Scores (/dashboard/analytics/scores) for drift
• Weight video higher for customer-facing roles — adjust on position settings
• Do not use authenticity as sole reject reason — combine with content score and human review

Related guides

• Assessments — category weights and text AI scoring
• Journey builder — video interview nodes
• Review queue — composite formula reference
• Culture fit — culture quiz and weight cap