Delivery record
Launched onboarding automation two weeks early; support tickets dropped 31% over the next sprint.
Talent Evidence OS
Evidence-first talent calibration, not automated people judgment.
Turn messy work evidence into a structured assessment with confidence levels, missing proof, safety boundaries, and a 60-90 day validation plan.
Role-fit review
Human review required
4
Evidence levels
90
Validation days
0
Final decisions automated
Working demo
Sample input to audit-ready output
The demo uses synthetic data so judges can reproduce the workflow without exposing private employee information.
Input
Evidence packet
Business goal: improve launch reliability for a growing product team.
Target role: project owner for cross-functional releases.
Period: last 90 days.
Structured feedback
Three peer reviewers cite clear escalation habits and strong handoff quality during cross-functional launches.
Concrete incident
One missed risk update caused design rework; manager log includes date, context, behavior, and impact.
Rejected signal
Private chat tone, online status, and vague culture-fit labels are excluded from the assessment.
Output
Calibrated assessment
Evidence confidenceMedium-high
Two independent A/B signals support role-fit claims; one C-level risk needs validation.
PerformanceStrong
Consistently converts ambiguous operating problems into shipped systems.
PotentialPromising
Learning speed and ownership are visible, but manager leverage needs a broader test.
Decision boundaryHuman review
Do not use this output for final hiring, firing, pay, or promotion decisions.
60-90 day validation plan
- Run a 60-day project-owner trial with explicit escalation checkpoints.
- Collect one calibrated customer or stakeholder review.
- Compare output quality against two peers using the same role rubric.
- Review privacy, bias, and appeal path before any high-impact decision.
How it works
Built around evidence gates
Define role context
Every review starts with business goal, target role, level, period, and decision purpose.
Classify evidence
Work results, structured feedback, incidents, and impressions are separated before reasoning.
Set boundaries
The system rejects private, unauthorized, protected, or high-impact final-decision requests.
Plan validation
Instead of false certainty, the output shows what to verify over the next 60-90 days.
Devpost package
Reproducible by design
The submission package includes this demo, synthetic sample data, setup notes, and a README explaining how Codex and GPT-5.6 were used to turn the framework into a running workflow.
Safety statement
Talent Evidence OS is not a hiring, firing, promotion, compensation, or layoff automation tool. It prepares evidence for accountable human review.