Earnings Prediction

Goal: Predict stock direction and magnitude before market reacts, using point-in-time data only.

Triggers

Called by earnings-orchestrator (not user-invocable).

Input

accession_no - 8-K earnings filing accession number
ticker - Company ticker symbol
quarter_label - Fiscal quarter (e.g., Q1_FY2024)

Output

result.json structure:

{
  "direction": "up",
  "magnitude": "extreme",
  "confidence_pct": 85,
  "primary_reason": "Beat EPS by 12%, raised FY guidance"
}

Field	Type	Values
direction	string	`up` / `down`
magnitude	string	`tiny` (0-1%) / `small` (1-3%) / `medium` (3-5%) / `large` (5-8%) / `extreme` (8%+)
confidence_pct	integer	0-100
primary_reason	string	Brief explanation

Workflow (2 Steps)

Step 1: Load Context

Read prediction/context.json for the event and treat pit_datetime as the hard cutoff for all fetched data.

Step 2: Write Prediction Output

Produce prediction/result.json with the required fields (direction, magnitude, confidence_pct, primary_reason) using only PIT-safe context.

Scripts

No mandatory script calls. This skill is bundle/context-driven and writes the prediction result file.

Hooks

PostToolUse: .claude/hooks/build-thinking-on-complete.sh - Builds thinking files on completion

Data Guardrails

See .claude/filters/rules.json for:

Forbidden patterns (lookahead bias blockers)
PIT date fields per data source

Folder Structure

earnings-analysis/Companies/{TICKER}/
├── cumulative/
│   ├── guidance.csv              # Full history (orchestrator only)
│   └── news.csv                  # Full history (orchestrator only)
└── events/
    └── {quarter_label}/
        ├── prediction/
        │   ├── context.json      # PIT = 8-K filing_datetime
        │   └── result.json       # Prediction output
        └── attribution/
            ├── context.json      # PIT = 10-Q filing_datetime
            └── result.json       # Attribution output

Output Files

Context: earnings-analysis/Companies/{TICKER}/events/{quarter_label}/prediction/context.json Result: earnings-analysis/Companies/{TICKER}/events/{quarter_label}/prediction/result.json

Invariants (Must Always Hold)

All data queries must be PIT-filtered
Never access return data (daily_stock, hourly_stock)
Consensus must come from pre-filing sources only

Version 1.0 | 2026-02-04 | Fresh rebuild from template

faisalanjum-earnings-prediction