Skip to content

Exporting

Everything you do in Scelo is reproducible. Two kinds of output: runnable code and the board-pack PDF.

Code export

export · code is available on every stage (Soft, Tools, Hard) and as EXPORT · WHOLE PIPELINE on the macro view. It turns your actions — the cleaning ops you applied, the date reformats, the derived columns, the models you picked, the runs — into a script.

Choose the language:

  • Python — pandas + the actuarial stack (chainladder, …).
  • R — tidyverse + ChainLadder, …
  • C++ — a scaffold with the steps as comments / TODOs.
  • Prompt — a natural-language description of the whole flow, for handing to another tool.

The export reads from an activity log that records each step (cleaning, date reformat, per-column clean, derived column, model run, data augmentation, …), so the script mirrors exactly what you did.

What a cleaning + reformat looks like in Python

# Cleaning ops applied via the banner:
#   • trim whitespace
#   • normalise missing markers
for c in df.select_dtypes(include='object').columns:
    df[c] = df[c].astype(str).str.strip()
df = df.dropna(axis=1, how='all')
df = df.drop_duplicates()
# Reformat date column(s) to American (MM/DD/YYYY):
df["joined_date"] = pd.to_datetime(df["joined_date"], errors="coerce").dt.strftime("%m/%d/%Y")

Dataset export

export ▾ in Soft Data writes the current (cleaned) dataset back out as CSV or Parquet.

The board-pack PDF

In Hard Data, the Board Pack node (⤢) or the report · pdf toolbar button opens a printable report — executive summary, estimates (forest plot), trajectory, and a per-model breakdown — with a download pdf button. See Hard Data.