Methods · The working approach

Start with a question.
Make it useful.

The tools change between a website, a financial model and a machine-learning study. The habits carry across: understand the task, look closely at the evidence and make the result something another person can use and question.

01

Find the decision behind the request.

The work starts with a practical task. A site map might support a shortlist; a document tool might make a review easier. Being clear about that job helps define a useful first version.

What it leaves behind

A specific question, a user and a way to judge the result.

See the clinic site explorer
02

Get close to the source material.

Source review looks for coverage gaps, inconsistent definitions and examples that do not fit. That might mean reading a report, inspecting a workbook or checking what a sensor actually recorded. These details often change the approach.

What it leaves behind

A source register, a data check or a small working example.

See the healthcare data work
03

Make the judgement calls visible.

Some inputs are measured; others are estimates or choices. Keeping them distinct makes it possible to explore how an assumption changes the answer. A model is easier to use when someone can see where their judgement belongs.

What it leaves behind

Named assumptions, a baseline and useful comparisons.

See the scenario modelling
04

Build a way to question the result.

The checks fit the work: a comparison with an existing workbook, an evaluation on held-out sessions or a reviewer inspecting the source behind an extracted field. A convincing demo is the start of that process.

What it leaves behind

Checks and review steps that match the claim being made.

See the document review workflow
05

Leave something useful behind.

The output needs to fit the next step. Sometimes that is an editable application, sometimes a spreadsheet or a clearly sourced report. Documented decisions and known limits make the work easier to pick up again.

What it leaves behind

A usable result with enough context to maintain and improve it.

See the publishing workflow

A few useful distinctions

Small words can
change the claim.

Clear descriptions distinguish what a system does, what a study found and what remains on the development roadmap.

Built, tested and validated

A feature can be implemented before its performance is well understood. Tests establish something about the cases they cover. Validation needs a clear basis and setting. Naming the stage helps a reader assess a working prototype on its own terms.

Ichne’s current research stage
Recorded values and modelled estimates

A staffing estimate is different from a recorded headcount. A scenario is different from a forecast. Keeping the source and assumptions visible helps prevent a useful approximation from becoming a fact by repetition.

How the workforce model handles this
A promising result and a transferable one

A result on a public dataset or a short backtest can help choose the next experiment. It does not establish performance in a new workplace or market. The evaluation setting stays close to the claim, including results that challenge the approach.

The local vision evaluation study

Have a question in mind?

Start with what the work needs to do.

A little context about the task is a good place to start.