predict

name: predict
description: Forecast future states or outcomes based on current data and trends. Use when estimating future values, projecting trajectories, forecasting outcomes, or anticipating system behavior.
argument-hint: "[target] [horizon] [conditions]"
disable-model-invocation: false
user-invocable: true
allowed-tools: Read, Grep
context: fork
agent: explore
layer: UNDERSTAND

Intent

Forecast future states or outcomes for a target based on current state, historical patterns, and assumed conditions. This capability consolidates all forecasting tasks (risk, impact, time, etc.) into a single parameterized operation.

Success criteria:
- Prediction for requested target and horizon provided
- Probability or confidence assigned to prediction
- Alternative outcomes considered
- Assumptions explicitly stated

Compatible schemas:
- docs/schemas/prediction_output.yaml

Inputs

Procedure

1) Define prediction target: Clarify what outcome is being predicted
- Specify the metric, state, or event to forecast
- Establish the prediction horizon
- Note any boundary conditions

2) Gather historical data: Collect relevant past observations
- Identify patterns and trends
- Note data quality and coverage
- Look for relevant precedents

3) Establish conditions: Document assumptions about the future
- Note what must remain constant
- Identify key variables that could change
- Consider external factors

4) Generate prediction: Forecast the most likely outcome
- Apply trend analysis or modeling
- Calculate probability of primary prediction
- Identify alternative outcomes

5) Consider alternatives: Evaluate other possible outcomes
- List plausible alternative scenarios
- Assign rough probabilities to alternatives
- Note what would cause each alternative

6) Ground prediction: Document evidence and reasoning
- Reference data supporting the prediction
- Note the reasoning chain
- Explicitly state all assumptions

Output Contract

Return a structured object:

prediction:
  outcome: any  # Predicted value, state, or event
  probability: number  # 0.0-1.0 likelihood of this outcome
  horizon: string  # When this prediction applies
alternatives:
  - outcome: any  # Alternative outcome
    probability: number  # Likelihood
    trigger: string  # What would cause this
trajectory:  # Optional: predicted path to outcome
  - timestamp: string
    state: any
reasoning: string  # Explanation of prediction logic
confidence: 0..1  # Confidence in prediction methodology
evidence_anchors: ["file:line", "data:source"]
assumptions: []  # Critical assumptions
invalidation_conditions: []  # What would invalidate this prediction

Field Definitions

Examples

Example 1: Predict Sprint Completion

Input:

target: "sprint story completion"
horizon: "end of sprint (5 days)"
conditions:
  team_size: 4
  no_major_blockers: true

Output:

prediction:
  outcome: "7 of 10 stories completed (70%)"
  probability: 0.65
  horizon: "end of sprint (5 days)"
alternatives:
  - outcome: "10 of 10 stories completed (100%)"
    probability: 0.15
    trigger: "No unexpected blockers, all estimates accurate"
  - outcome: "5 of 10 stories completed (50%)"
    probability: 0.20
    trigger: "Major blocker or team member unavailable"
trajectory:
  - timestamp: "Day 2"
    state: "3 stories completed"
  - timestamp: "Day 4"
    state: "5-6 stories completed"
  - timestamp: "Day 5"
    state: "7 stories completed"
reasoning: "Based on team velocity of 1.5 stories/day and current progress, projecting linear completion. Historical sprints show 70% completion rate."
confidence: 0.7
evidence_anchors:
  - "tool:jira:sprint-history"
  - "tool:jira:current-burndown"
assumptions:
  - "Team velocity remains consistent"
  - "Story estimates are accurate"
  - "No unplanned work introduced"
invalidation_conditions:
  - "Team member becomes unavailable"
  - "Critical production incident"
  - "Scope change to sprint"

Example 2: Predict System Failure

Input:

target: "disk space exhaustion"
horizon: "next 30 days"
conditions:
  growth_rate: "current trend"

Output:

prediction:
  outcome: "Disk will reach 95% capacity"
  probability: 0.85
  horizon: "18-22 days"
alternatives:
  - outcome: "Disk reaches 100% (service outage)"
    probability: 0.60
    trigger: "No cleanup or expansion action taken"
  - outcome: "Capacity remains stable"
    probability: 0.10
    trigger: "Log rotation reduces growth rate"
trajectory:
  - timestamp: "Day 7"
    state: "85% capacity"
  - timestamp: "Day 14"
    state: "90% capacity"
  - timestamp: "Day 20"
    state: "95% capacity"
reasoning: "Current growth rate of 0.5% per day on 80% utilized 500GB disk. Linear projection shows critical threshold in ~20 days."
confidence: 0.8
evidence_anchors:
  - "command:df -h /data"
  - "tool:monitoring:disk-trend-7d"
assumptions:
  - "Growth rate continues at current pace"
  - "No bulk data imports or exports"
  - "Log retention policy unchanged"
invalidation_conditions:
  - "Growth rate changes significantly"
  - "Disk expanded or data archived"
  - "Application behavior changes"

Verification

• [ ] Prediction includes specific outcome and probability
• [ ] Horizon is clearly specified
• [ ] At least one alternative outcome considered
• [ ] Assumptions are explicitly documented
• [ ] Evidence supports the prediction reasoning

Verification tools: Read (to verify historical data references)

Safety Constraints

• mutation: false
• requires_checkpoint: false
• requires_approval: false
• risk: low

Capability-specific rules:
- Always provide probability, never claim certainty about future
- Document assumptions that could invalidate prediction
- Consider alternative outcomes, especially failure modes
- Do not predict beyond available data horizon without noting extrapolation
- Flag when prediction confidence is too low to be actionable

Composition Patterns

Commonly follows:
- measure - Measurements provide basis for predictions
- observe - Current state observations inform predictions
- discover - Discovered patterns enable predictions

Commonly precedes:
- plan - Predictions inform planning decisions
- compare - Predicted outcomes can be compared
- simulate - Predictions guide simulation scenarios

Anti-patterns:
- Never use predict for current state (use measure or observe)
- Avoid predict when historical data is insufficient

Workflow references:
- See workflow_catalog.yaml#digital_twin_sync_loop for forecasting in digital twins