The Unwritten
Job Description.

What the machine assumes.

All scenarios, people and events are invented. Four neutral role labels are used (Lead, Practitioner, Coordinator, Support); no protected characteristics are assigned or inferred. A seed hashes to a reproducible ordering of daily tasks; its value has no predictive meaning. Each task has a category, weight, essentiality, original owner and default credit recipient.

Illustrative indices are editorial. Four fictional roles each have a small capacity budget; tasks carry integer load units. On each observed day, reassignment changes role load or attribution, with an explicit before-and-after indicator trace. Each recorded allocation revision also adds short-term model friction, never a claim of real-world causal magnitude. Overload and unassigned essential tasks raise the invented failure-risk index. Formal visibility improves the invented pay-visibility index but raises monitoring pressure. No index is a validated instrument.

Each ownership or credit reassignment is recorded as one illustrative administrative transaction. The indices include a one-point short-term capacity and trust cost, one-point pay-allocation traceability increase, and one-point handoff risk increase. Other effects depend on load, visibility, unassigned work and attribution. These editorial coefficients are not estimates of real organizations.

The structural hearing compares five counterfactual changes to the same observed ten tasks. Rotation redistributes existing load; demand reduction removes select negotiable tasks; compensation increases formal recognition without creating capacity; authority reduces selected coordination or repair weights; documentation increases traceability but not total demand. Compare maximum load as well as total work so a apparent reform cannot hide displaced burden.

Research informed the questions, not the formulas. A causal narrative in this toy cannot validate an organizational intervention. No workplace or demographic inference is produced. This site transmits no case text, runs without a server or analytics, and retains nothing after refresh; exporting or printing is an explicit user choice.