Onfolio Loading the public task catalogue…
Development preview
Evaluation suites for professional work.
We are developing task suites and resettable environments for evaluating AI agents on ecommerce, search, editorial, and acquisition workflows.
Each task combines a brief, source records, scoped tools, and grading criteria. Agents investigate the evidence, prepare changes, and submit work for evaluation.
Cedar & LoomDemo environment
Investigate a retention incident
Reconcile store and campaign records, identify configuration defects, and prepare a repair plan.
Store ordersPurchase eventsCustomer permissionsCampaign reportsSupport ticketsFlow settings
Findings with source references
Proposed configuration changes
Validation and release handoff
Synthetic merchant snapshot
Method
Practitioner-led task design.
Commercial assignments will be designed by domain practitioners, with explicit success criteria and acceptable alternatives.
The environment supplies the records, tools, and starting state. Automated checks assess calculations, configuration, and policy constraints. Independent expert review assesses evidence and judgment.
- Task authorship
- Practitioners define the assignment and material failure cases. AI assists with specification and implementation.
- Source records
- Tasks include linked records, operating policies, and defined action boundaries.
- Evaluation
- Score observable outcomes and preserved constraints. Record expert assessments separately.
Public demo
Cedar & Loom
A synthetic homewares merchant with ten related assignments covering tracking, audiences, lifecycle messaging, offers, and release review.
Demo status The current tasks and records are AI-designed fixtures for testing the tools and grading workflow. Expert-authored commercial tasks, independent reviews, and live model evaluations are not yet available.
Original incident snapshot, frozen at 20 October 2026. Money fields ending in “cents” are integer USD cents. Release-review tasks use derived snapshots in the download.
This browser preview lets you inspect briefs and records. The downloadable pack contains public inputs and local Python tools for preparing submissions. Grading is run separately; private graders and reference answers are not included.
Collections
Domains and current status.
Ecommerce has a working synthetic demo. The remaining collections are planned; commercial task design will be led by practitioners in each domain.
- Demo available
Ecommerce
Tracking, lifecycle workflows, customer audiences, and campaign reconciliation.
- Planned
SEO
Technical diagnosis, search evidence, and supported priorities.
- Planned
Generative engine optimisation
Source accuracy, entity consistency, and reproducible visibility measurement.
- Planned
Proofreading
Accurate corrections, preserved meaning, and sound editorial judgment.
- Planned
Acquisition due diligence
Reconciled financials, supported adjustments, and decision-ready findings.
Enquiries
Research and pilot enquiries.
For evaluation requirements, scoped pilots, or practitioner contributions, contact Dom Wells.