Scelo without the IDE¶

The brain layer ships as two libraries, scelo for Python and scelo for R, for actuaries who would rather write three lines of code than click through a workstation. They are the workstation's own logic, not a re-implementation: the same import typing, the same cleaning rules and thresholds, the same life tables and commutation functions, the same reserving engine, the same W(M, T, R) forecast down to the random stream, and a client for the bundled swarm.

PythonR

import scelo as sc

df  = sc.load("claims.csv")        # typed the way the IDE types it
df  = sc.clean(df)                 # the cleaning banner's safe ops, audited
res = sc.reserve(df)               # chain ladder, Mack, BF, ODP bootstrap
sc.report(res, to="pack.html")     # a board pack: tables that carry their basis and hash

library(scelo)

df  <- sc_load("claims.csv")
df  <- sc_clean(df)
res <- sc_reserve(df)
sc_report(res, to = "pack.html")

The three rules both libraries follow¶

A data frame goes in first, and the columns are inferred. sc.triangle(df) finds the origin, development and amount columns by name (accident_year, AY, origin all mean origin); pass names only when it guesses wrong, and the error tells you exactly what to pass.
Every answer is a Table. Still a real pandas.DataFrame / data.frame, but it carries its title, its basis and its notes, and prints them under the numbers. hard adds a content hash, a timestamp, the version and the audit trail: the one-way pipeline rule in practice — a number never travels without what produced it.
Everything is one sc. (Python) or sc_ (R) away. Type the prefix and let completion show the map, or print it: sc.cheatsheet() / sc_cheatsheet().

The manual¶

Chapter	What it covers
Install & set up	pip / R install, the extras, the IDE's runtimes, the swarm
Quickstart	a full session in sixty seconds, both languages
The Table	the result type: title, basis, notes, provenance
Soft data	load, profile, describe, tab, the 18 cleaning ops, combine
Life & mortality	life tables, commutation, factors, premiums, graduation, Lee–Carter, Kaplan–Meier, BasicTerm, SCR, CSM
Reserving	triangles, chain ladder, Mack, BF, Cape Cod, ODP bootstrap
Finance	interest, annuities, bonds, curves, Smith–Wilson, Nelson–Siegel, Hull–White
Risk	VaR / TVaR, aggregate losses, fitting, credibility, SCR aggregation
Pricing & fairness	GLMs, relativities, lift and Gini, fairness metrics and audits
Climate	reanalysis ensembles, return periods, parametric triggers, AAL
Forecast & the swarm	the W(M, T, R) engine, scenario parsing, council, society, augment
Workspace diagnostics	the bottleneck and active-subspace readouts from the Global Workspace paper
Hard data & reports	hard, verify, board packs, snapshots, the audit ledger
Charts	the `plot_*` family and its palette
The command line	`scelo profile file.csv` and friends
Python notes · R notes	what is idiomatic on each side
Function reference	every exported name, A → Z

What is in them¶

Stage	Python	R
Soft data	`load` `profile` `describe` `tab` `suggest` `clean` `combine`	`sc_load` `sc_profile` `sc_describe` `sc_tab` `sc_suggest` `sc_clean` `sc_combine`
Life	`life_table` `commutation` `factors` `premium` `ae` `model_points` `graduate` `lee_carter` `kaplan_meier` `exposure` `basicterm` `scr_life` `csm` `lifelib_run`	the same with `sc_`
Reserving	`triangle` `ata` `chain_ladder` `mack` `bf` `cape_cod` `bootstrap` `tail` `reserve`	the same with `sc_`
Finance	`discount_curve` `smith_wilson` `nelson_siegel` `nss` `hull_white` `pv` `irr` `annuity_certain` `duration`	the same with `sc_`
Risk	`var` `tvar` `aggregate_loss` `fit` `credibility` `aggregate_scr` `risk_margin`	the same with `sc_`
Pricing & fairness	`glm` `relativities` `freq_sev` `loss_ratio` `lift` `gini` `fairness` `fairness_audit`	the same with `sc_`
Climate	`ensemble` `return_period` `parametric_trigger` `aal` `anomaly`	the same with `sc_`
Forecast & swarm	`wmtr` `sensitivity` `council` `society` `augment`	the same with `sc_`
Hard data	`hard` `report` `export` `audit` `verify` `snapshot`	the same with `sc_`
Charts	`plot_rates` `plot_relativities` `plot_projection` `plot_scr` `plot_csm` `plot_bars` `plot_lines` `plot_triangle` (matplotlib)	the same with `sc_plot_` (base graphics)

Twins, tested against each other¶

Three implementations are kept in lock-step: the TypeScript inside Scelo IDE, the Python package, and the R package. The R suite reads a golden fixture computed by the Python package — life tables, commutation columns, Mack on the published RAA triangle, Smith–Wilson, Panjer, GLM coefficients, the cleaning plan for the dirty sample — and matches it to 1e-9. Both packages check the W(M, T, R) engine against the same fixture generated by the IDE's TypeScript, down to the identical Mulberry32 random stream. Pick the language your team writes; the answers do not change.