Downloadable AI work tests
We test bounded tasks such as employee scheduling and invoice reconciliation. Every approved case keeps the exact prompt, untouched first result, one correction, separately run checks, downloadable evidence, and limits together.
Featured result
We tested whether AI could create a valid 24/7 employee shift schedule, preserved the exact prompt, and checked the Excel file against five fixed rules.
Every published case includes
Evidence by task
Task areas start as transparent evidence notes. Cross-case comparisons and averages appear only after at least three distinct, search-approved records support the same decision.
Guide 01
Compare bounded tests of shift coverage, queues, routing, staffing, priorities, and constrained assignments.
1 search-reviewed record · 0 failed outcomes retainedReview this task area →Guide 02
See how AI handles discrepancies, logs, inventories, cash records, reports, and other checkable operational data.
1 search-reviewed record · 0 failed outcomes retainedReview this task area →Latest search-reviewed evidence records
Every visible score is tied to a bounded task, exact prompts, two saved outputs, and five checks. Released research-corpus records stay outside this search-facing view until an additional evidence review admits them.
Completed evidence testSynthetic fixture · structured JSON · text-only evaluation
The first reconciliation silently cleared three exceptions. One correction exposed the partial payments, missing reference, and exact totals.
Completed evidence testSynthetic fixture · workbook · sandbox artifact run
We tested whether AI could create a valid 24/7 employee shift schedule, preserved the exact prompt, and checked the Excel file against five fixed rules.
Browse by the job
Set the task and pass/fail rules before prompting.
Save the exact words, tool, model disclosure, and date.
Check the output independently; do not trust confident prose.
Keep the mistakes, correction, score, and final artifact together.
Evidence taxonomy
A synthetic fixture can produce a workbook, a structured JSON record, or a text transcript. Every case labels input origin, artifact type, execution mode, and run count separately so a downloadable file is never mistaken for real-world input.
Transparency note
WeTriedAI currently runs no ads, analytics, affiliate links, or tracking cookies. If that changes, the relevant pages and disclosures will change first.
Our editorial policy →