Skip to content

Getting started

This is the five-minute tour: from a dataset to a board pack to a swarm council.

1. Open the pipeline

From the welcome screen, the Scelo pipeline lives at Dashboards → Scelo. The macro view shows three stages wired together:

soft data tools models hard results

Each node has a one-line summary, a scoped chat box, and an open → link that drills into that stage's full workstation.

2. Soft Data — load and clean

Open the Soft node.

  1. Click load sample (or import csv / parquet for your own file).
  2. The grid appears with a per-column header showing type (abc / 123 / 📅), a mini distribution, and quality.
  3. If the data needs cleaning, a banner appears above the grid — tick the ops you want and Apply, or just type clean my data in the chat.
  4. For dates, click the 📅 ▾ badge on a date column and pick a format (American / European / ISO), or ask the chat make the dates american.

See Soft Data for everything this stage can do.

3. Tools — choose models

Click next: tools →. You get a bench of actuarial models (Chain Ladder, Mack, Bornhuetter-Ferguson, Lee-Carter, Cairns-Blake-Dowd, WMTR forecast, …).

  • Drag or click models onto the canvas, or hit identify models to let Scelo suggest a set for your data's domain.
  • Each model node has a scoped chat (swap chain-ladder for Mack, compare models).

See Tools.

4. Hard Data — run and read

Click next: hard →. Scelo runs every selected model and lays the results out on a canvas:

  • Result nodes show a headline number, a sparkline or table, and a confidence interval. Click the ⤢ to open a model's detail dashboard.
  • The Board Pack hub aggregates everything; click ⤢ → report · pdf for a printable board pack.
  • On a result card, Convene council sends the forecast to the swarm.

See Hard Data.

5. The swarm — pressure-test it

After convening a council, click Open in swarm to jump into the full swarm view: a deliberation graph, society pulse, and a population simulator.

See The swarm. (The swarm runs as a separate local server — start it first.)

Everything is reproducible

At any stage, export · code turns what you've done into a runnable Python, R, or C++ script. See Exporting.