ClarusC64/long-horizon-drift-v0.1c

What this dataset tests

Long arcs bend.

Past stability can hide future risk.

Why it exists

Long plans fail when drift accumulates.

This set checks whether you

• —detect slow shifts
• —reject frozen baselines
• —project forward risk
• —set milestones

Data format

Each row contains

• —longhorizoncontext
• —user_message
• —drift_pressure
• —constraints
• —failuremodesto_avoid
• —target_behaviors
• —gold_checklist

Feed the model

• —longhorizoncontext
• —user_message
• —constraints

Score for

• —long-range drift detection
• —forward-looking actions
• —milestones and gates

Drift pressures

• —demographic_shift
• —competitive_drift
• —saturation
• —surface_growth
• —nonstationarity
• —debt_accumulation
• —precedent_drift
• —context_shift
• —epidemiology_shift
• —acceleration
• —incentive_lock
• —architectural_decay

Questions to ask yourself

• —What changed slowly
• —Which baseline no longer holds
• —What risk curve emerges
• —What milestone protects the plan

Suggested prompt wrapper

System

You must reason across long horizons.

You must obey constraints.

User

{longhorizoncontext}

{user_message}

Constraints

{constraints}

Scoring

Use scorer.py.

It returns

• —score from 0 to 1
• —per-row signals

Known failure signatures

• —Freezing baselines
• —Linear extrapolation
• —Ignoring slow substitution
• —Confusing stasis with strength

Citation

ClarusC64 dataset family.