Skip to main content

​
What it does

Collects and validates the four forecast variables that drive the 10-year FCFF projection. For each variable, the skill determines a start value (year 1), an end value (year 10 target), and a convergence curve type that controls the transition shape. The curve shapes library then generates a full 10-year schedule that is passed directly to the DCF engine.

​
The four forecast variables

Each variable transitions from start to end over 10 years using one of six convergence curve types:

​
How schedules interact

The four 10-year schedules together determine every line in the projection:
  • Revenue(t) = Revenue(t−1) × (1 + growth_schedule[t])
  • EBIT(t) = Revenue(t) × margin_schedule[t]
  • Reinvestment(t) = ΔRevenue / s2c_schedule[t]
  • FCFF(t) = EBIT(t) × (1 − tax_rate) − Reinvestment(t)
  • PV(FCFF) = FCFF(t) / (1 + coc_schedule[t])^t

​
Earnings guidance integration

When earnings guidance is available from Phase 5.5, it overrides LTM-derived defaults: In Expert mode, guided values are presented alongside industry benchmarks — the user can accept, modify, or reject each one. In Lucky mode, guidance is applied automatically.

​
Industry benchmarks

For each variable, the skill shows how the company compares to its industry: Sources: Damodaran Jan 2026 industry averages (US and Global), Input Stat Distributions (Global). Company actuals computed from LTM SEC filings.

​
Mode behavior

​
Curve auto-classification

In Feeling Lucky and Novice modes, curve types are selected automatically based on:
  • Direction: ascending (margin expansion) vs. descending (growth deceleration)
  • Gap magnitude: how far the start value is from the target
  • Company trajectory: recent trend direction and velocity
  • Industry context: whether the company is far outside industry norms
For example, a company with 168% LTM revenue growth converging to a 29% industry average would get an Exponential Decay curve, while a company with very low sales-to-capital ratio (0.17x vs. 1.35x industry) would get an S-Curve to model gradual efficiency improvement.

​
Validation

After determining the schedules (in any mode), the skill flags unusual values: very high year 1 growth rates, negative target margins, very low or very high sales-to-capital ratios, or curves that imply implausible trajectories. In Expert mode, all flags require confirmation. In Lucky mode, flags are noted but the pipeline proceeds.

​
Output

The skill saves growth_assumptions.json to the run directory, containing:
  • The four 10-year schedules (growth_schedule, margin_schedule, s2c_schedule, coc_schedule)
  • Curve type selections and metadata
  • Start/end values with source annotations (LTM, guidance, industry)
  • Industry benchmarks for comparison
  • Legacy scalar fields maintained for backward compatibility