Tasks

Forecast which person or entity the participant will contact next, or whether they will contact a given entity within a fixed window.

Example question

Who will the participant initiate contact with first tomorrow morning (08:00–12:00)?

Answer space

A short, person-specific list of frequent contacts plus an explicit “other” / “no contact” option.
["contact_A","contact_B","contact_C","other","no_contact"]

Resolution

Resolved from communications metadata (first outbound message/call to a contact within the window). Adjudicated by the federated client; only the outcome label is reported.

Baseline

R1 population contact base rates; R2 the participant's own recent contact frequency/recency.

Difficulty

Heavy-tailed contact distributions make R2 strong. Skill must come from context that shifts the next contact away from routine.

Forecast whether a calendar or planned event actually occurs, is cancelled, or moves.

Example question

Will the 14:00 meeting on the participant's calendar tomorrow occur as scheduled, move, or cancel?

Answer space

Categorical: occurs as scheduled / moves / cancels.
["occurs","moves","cancels"]

Resolution

Resolved from calendar state and communications metadata at the resolution time. Status is adjudicated against an explicit rubric (e.g., a >30 min shift counts as “moves”).

Baseline

R1 population cancellation/move rates; R2 the participant's historical event-realization rates.

Difficulty

Base rates are informative; genuine skill requires reading signals that a specific event is at risk.

Forecast whether the participant replies to a given message and, if so, the response-latency band.

Example question

Given this received message, will the participant reply within 1h, within 24h, or not within 24h?

Answer space

Ordered bands: reply <1h / reply 1–24h / no reply <24h.
["reply_lt_1h","reply_1_24h","no_reply_24h"]

Resolution

Resolved from outbound reply timestamps relative to the triggering message. Bands are fixed in advance in the protocol.

Baseline

R1 population reply-latency distributions; R2 the participant's per-contact reply history.

Difficulty

Strong per-contact priors (R2) set a high bar; skill comes from situational context (workload, location, time of day).

Forecast which active project or topic will receive the most attention in the next window.

Example question

Which active project will receive the most of the participant's working attention tomorrow?

Answer space

A person-specific set of active projects/topics plus “other.”
["project_1","project_2","family","admin","other"]

Resolution

Resolved from the participant's own end-of-window labelling and/or activity evidence, against a pre-registered attribution rule. Only the resolved label leaves the client.

Baseline

R1 population topic priors (weak); R2 the participant's recent attention distribution.

Difficulty

Attention is volatile and partly intention-driven. This is where richer evidence (L2–L3) is hypothesised to help most.

Forecast deviations from the participant's established routine.

Example question

Will the participant's tomorrow deviate from their typical weekday routine on a pre-specified dimension (e.g., start time, commute, core block)?

Answer space

Binary or categorical deviation on a pre-registered dimension.
["no_deviation","minor_deviation","major_deviation"]

Resolution

Resolved by comparing the realized day to a routine model on the pre-specified dimension, against a fixed threshold rubric.

Baseline

R1 population deviation rates; R2 the participant's own deviation base rate (a deliberately strong baseline).

Difficulty

By construction R2 is hard to beat. Skill requires anticipating the specific causes of deviation, not its average frequency.