Eval directory · Document Agents
Evaluation packs covering adversarial robustness, safety gates, workflow quality, and operator-level checks for Google Workspace AI products.
Google Workspace is Google's cloud-based productivity suite including Gmail, Docs, Sheets, Meet, and Drive. Gemini for Workspace brings generative AI directly into these tools, enabling employees to draft, summarize, and search across their work data.
Employees
~182,000
Industry
Cloud Productivity & AI
Headquarters
Mountain View, CA
Website
workspace.google.comUse the eval library for Google Workspace
All 32 test cases — inputs, expected behavior, and pass/fail checks — runnable in Corsac with your own data.
Related in Document Agents
All evals →Coverage map
6 areas · 32 graded scenarios
Every eval set is graded on
Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls
01
Schedule meetings across time zones, detect conflicts, and ask for clarification when constraints cannot be satisfied.
Mapped capabilities
6 scenarios
Example criterion: Scheduling decisions respect time zones, attendee availability, and explicit constraints.
02
Extract action items, owners, and next steps from Workspace documents without inventing missing details.
Mapped capabilities
6 scenarios
Example criterion: Workspace doc responses stay grounded, preserve ownership, and avoid fabricating follow-up work.
03
Find the right Drive artifact, respect permission boundaries, and handle external-sharing requests with discipline.
Mapped capabilities
6 scenarios
Example criterion: Drive responses find the right file, avoid bad shares, and keep permission changes auditable.
04
Eval for Drive search, file selection, permission discipline, and sharing governance under realistic collaboration requests.
Mapped capabilities
4 scenarios
Example criterion: The assistant finds the right file, respects access boundaries, and escalates when a request would widen permissions unsafely.
05
Prioritize inbox messages, draft appropriate replies, and escalate only high-risk mail with strong justification.
Mapped capabilities
6 scenarios
Example criterion: Inbox responses are concise, correctly classified, and careful about phishing, legal threats, and customer urgency.
06
Eval for Gmail triage, response drafting, and calendar scheduling when multiple stakeholders and time zones are involved.
Mapped capabilities
4 scenarios
Example criterion: The assistant prioritizes urgent mail correctly, drafts usable replies, and schedules meetings without double-booking or timezone mistakes.
Each eval pack tests Google Workspace's public product surface — including Calendar Scheduling Conflicts V1, Docs To Action V1, Drive Search Share Governance V1 — against graded scenarios covering adversarial robustness, workflow quality, safety gates, and operator quality. Every pack is runnable in Corsac with your own data.
Every eval set is graded the same way: pass/fail checks plus an LLM judge scoring 1–5 against each case's expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.
The Google Workspace eval library includes 32 graded test cases across 6 eval packs. Each case defines an input, expected behavior, and pass/fail criteria.
Request the library with your work email above and we'll set it up in a Corsac workspace, where you can run every test case against Google Workspace or your own agent with your own data.
