The command line¶

Installing the Python package also installs scelo, a small command line for the times a terminal is closer than a notebook: profile a file someone just sent you, clean it, run the reserving battery, print a life table. Twelve subcommands, each a thin wrapper over the library function of the same name — same inference, same output, same notes.

$ scelo --help
usage: scelo [-h]
             {profile,describe,suggest,quick,clean,reserve,life,wmtr,samples,sample,cheatsheet,version} ...

Scelo for the terminal: soft data → tools → hard data.

Looking at a file¶

scelo profile  claims.csv     # type, missing, unique, five-number summary, fences, top values
scelo describe claims.csv     # the statistician's view: sd, quantiles, shape, ranked by CV
scelo suggest  claims.csv     # the cleaning plan, with evidence and safe flags
scelo quick    claims.csv     # profile + the plan in one go — the first thing to run on a new file

Cleaning¶

scelo clean claims.csv                     # safe ops only, prints what changed
scelo clean claims.csv --all -o clean.csv  # every op, written to a new file

The input file is never modified: cleaned output goes to stdout as a summary, and to disk only where -o/--out says.

Models¶

scelo reserve claims.csv                             # chain ladder · Mack · BF · ODP bootstrap
scelo reserve claims.csv --origin uw_year --value paid   # when inference needs overriding
scelo life                                           # the illustrative life table
scelo life --i 0.04                                  # commutation functions at 4 %
scelo life mortality.csv                             # a life table from your own age + qx file
scelo wmtr "pension scheme with a weakening sponsor covenant" --paths 500

reserve prints the same four-method summary the library's reserve() returns, notes included:

$ scelo reserve claims.csv
                         latest      ultimate          ibnr ...
chain-ladder          1874437.0  3.407400e+06  1.532963e+06 ...
mack                  1874437.0  3.407400e+06  1.532963e+06 ...
bornhuetter-ferguson  1874437.0  3.407400e+06  1.532963e+06 ...
bootstrap             1874437.0  3.448399e+06  1.573962e+06 ...
— Reserve summary
  basis: paid · origin_year × dev
  · Triangle: 7 origins × 7 lags. Mack SE ±1.96 → [854,731, 2,211,194]; bootstrap p95 2,316,627.

Samples and the map¶

scelo samples                 # list the six bundled datasets
scelo sample dirty -o dirty.csv   # write one out to practise on
scelo cheatsheet              # the whole library on one screen
scelo version

The bundled samples (the same ones sc.sample() / sc_sample() return in code):

Key	What it is
`claims`	79-row incomplete P&C claims triangle, origins 2018–2024, with policy, line, province, age, sex, paid, incurred, settled
`dirty`	53-row customer ledger with every real-world mess: currency strings, %-numbers, sentinel ages, mixed booleans and dates, mojibake, missing markers, duplicates
`climate`	30 daily records for one grid cell (Pretoria, Jan 2024): 2-m temperature and precipitation under ERA5 / MERRA-2 / JRA-3Q
`wmtr-scenarios`	12 scenario rows for the W(M, T, R) forecast: alphas, relational weights, shock, horizon
`lifelib-mp`	100-row in-force term model-point file shaped like lifelib's `basic_term_sample`
`workspace-demo`	2,000-policy synthetic annuity book: three low-variance real drivers through nonlinear channels, ten nuisance columns

R users

The command line ships with the Python package only. From R, the same one-liners are sc_profile(sc_load("claims.csv")) and friends — every subcommand has an sc_ twin.