Stability & Decision Risk
How Vetriva measures the reliability of a hiring decision.
Why stability matters
A candidate can have a high score but a low stability — meaning the decision is contested or fragile. Stability gives you a second dimension beyond the score itself: how likely is the current recommendation to hold if more signals arrive?
Stability index
The stability index is a value from 0 to 100 that measures the consistency and agreement of signals on a candidate. A high stability means reviewers agree and the recommendation is unlikely to change. A low stability means there is disagreement or too few signals to form a confident view.
| Stability index | Label | Meaning |
|---|---|---|
| 80 – 100 | Stable | High agreement, decision is reliable |
| 55 – 79 | Moderate | Some variance, more signals would help |
| 30 – 54 | Volatile | Significant disagreement, treat with caution |
| 0 – 29 | Critical | High conflict, decision is unreliable |
Volatility
Volatility measures how much the adjusted score has changed over successive signals. A stable candidate has a smoothly trending score. A volatile candidate has large swings — for example, alternating between Strong Yes and Strong No signals.
Collapse probability
Collapse probability estimates the likelihood that the current recommendation will change to a different decision band if one more average-strength signal is added. A high collapse probability (above 40%) means the candidate is near a threshold and a single signal could flip the recommendation.
Risk labels
Vetriva applies a combined risk label based on stability index and collapse probability:
- Low risk — stable and low collapse probability
- Moderate risk — some volatility or moderate collapse probability
- High risk — unstable or high likelihood of recommendation change
How to reduce risk
Add more signals from additional reviewers with clear, high-confidence assessments. Consistent signals converge the score toward a reliable point and increase the stability index.
Decision Reliability
While the stability index measures a single candidate, Decision Reliability measures the consistency of your entire hiring pipeline. It appears on the Decision Insights page as a score from 0 to 100 with a letter grade.
| Grade | Score | Label | Meaning |
|---|---|---|---|
| A | 80 – 100 | Highly Reliable | Decisions are consistent and stable across your pipeline with minimal reversal risk |
| B | 65 – 79 | Mostly Reliable | Minor instability present but signals are broadly aligned — worth monitoring edge cases |
| C | 45 – 64 | Needs Attention | Noticeable volatility or reversals detected — some decisions may not hold under additional feedback |
| D | 0 – 44 | Unreliable | Significant instability across the pipeline — decisions should not be finalised without further review |
How Decision Reliability is calculated
The score is a weighted combination of three factors:
- Average signal stability (40%) — the mean stability index across all candidates
- Flip resistance (35%) — the inverse of average flip probability across candidates
- Reversal-free rate (25%) — proportion of candidates whose recommendation never crossed the hire/reject boundary
Insight signals
Vetriva automatically detects the following patterns and surfaces them as alerts in the reliability panel:
- Late-stage reversal — a candidate moved from Hire to Reject (or vice versa) during the evaluation. This indicates an inconsistent assessment process.
- High-volatility role — a role where the average stability index is below 40%, meaning scores are shifting frequently across candidates.
- Polarization — 80% or more of a role's candidates are in extreme buckets (Hire or Reject) with very few in the middle.
- Low agreement — no single decision bucket exceeds 40% for a role, meaning the team has no clear consensus.
- Unstable pipeline — org-wide reliability score below 45 with five or more candidates evaluated.
Candidates at risk
The reliability panel also shows a Candidates at risk count — the number of candidates whose flip probability exceeds 40%. These are candidates where one or two more average signals could change the recommendation. Prioritise structured review for these candidates before recording a final outcome.