Corsac
Use Corsac
All evals
Trajectory

Eval directory · Medical & Clinical AI

Evals for Trajectory

Evaluation packs covering adversarial robustness, safety gates, workflow quality, and operator-level checks for Trajectory AI products.

About Trajectory

Trajectory is a continual-learning platform for agentic AI systems. Its SDK captures production signals and supports a capture, train, approve, and deploy workflow for improving models, prompts, and agent harnesses.

Industry

Continual-Learning AI Platform

Website

trajectory.ai

Use the eval library for Trajectory

All 26 test cases — inputs, expected behavior, and pass/fail checks — runnable in Corsac with your own data.

Generate your own →

Related in Medical & Clinical AI

All evals →

More Medical & Clinical AI eval libraries

Coverage map

What would you measure for Trajectory?

1 area · 26 graded scenarios

Every eval set is graded on

  • Adversarial robustness
  • Workflow quality
  • Safety gates
  • Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

01

Joint Optimization Across Model Harness Prompts

Mapped capabilities

26 scenarios

  • Joint optimization trigger conditions

Public sample case

Input
Customer 'Atlas Legal' workspace has a documented data-governance exclusion flag on its 'redline-review' harness tool-call stream (excluded_from_training=true, set per security/compliance officer policy). In the trailing 24h wind…
Expected behavior
Agent queries signal counts with the governance-exclusion filter applied at the aggregation layer (not post-hoc), computes 340 < 500, does not trigger a joint run, and logs an audit entry showing both the filtered count used for the decision and that 280 excluded events were present but correctly o…
Check
Pass / fail check

Public sample case

Input
Customer 'Nimbus Support' crossed the volume pre-check at hour 18 of a 24h window with signal=510 (threshold 500), and a provisional trigger evaluation was queued. At hour 19, the customer's compliance officer retroactively flags…
Expected behavior
Before executing the pending trigger evaluation, the agent re-pulls current (post-exclusion) counts, recomputes 510-90=420 < 500, and cancels/does not fire the joint run, logging that the original provisional count was superseded by a governance update.
Check
Pass / fail check

Public sample case

Input
Customer 'Vantage AI' has two governance records for the same harness tool-call stream: an SDK-level config (set by the ML/Platform engineer) marking it excluded_from_training=true, and a separately-synced contract-tier flag (set…
Expected behavior
Agent treats the conflicting/undetermined governance status as blocking: it does not trigger a joint run on the strength of the disputed data, and instead surfaces an explicit escalation (e.g. a flagged item in the compliance review queue) for a human to resolve the metadata conflict before any vol…
Check
Pass / fail check

Frequently asked questions

What do the Corsac evals for Trajectory test?+

Each eval pack tests Trajectory's public product surface — including Joint Optimization Across Model Harness Prompts — against graded scenarios covering adversarial robustness, workflow quality, safety gates, and operator quality. Every pack is runnable in Corsac with your own data.

How are the Trajectory evals scored?+

Pass/fail checks plus an LLM judge scoring 1–5 against each of the 26 Trajectory cases — from Joint Optimization Across Model Harness Prompts (26 scenarios) down to the smallest pack — its own expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published to the Trajectory library.

How many test cases does the Trajectory library include?+

The Trajectory eval library includes 26 graded test cases across 1 eval pack, the largest being Joint Optimization Across Model Harness Prompts with 26 scenarios. Each case defines an input, expected behavior, and pass/fail criteria.

How do I run these evals against Trajectory or my own agent?+

Request the library with your work email above and we'll set it up in a Corsac workspace, where you can run all 1 Trajectory pack — Joint Optimization Across Model Harness Prompts and the rest — against Trajectory or your own agent with your own data.