Skip to content
Menu
The market research case

Market research · system map

Market research, under the hood

Explore how a research question travels from a product team to the public ledger and back as a decision. Play the tour, pick one of the 6 journeys, or select any part to see what it does.

Illustration: the platform stays unnamed

Journeys6 journeys through the system

PEOPLEwho asks and decidesSCRIPTSwritten with Claude, reviewedPUBLIC LEDGERthe full recordRAW RECORDSevery event in the periodCHECK, ANALYSE, DECIDEProduct teamReviewerClaudeAIExtraction scriptVersion controlPublic ledgertrades · ordersbalance changesLedger totalsRaw extractClean tableReconciliationAnalysisAIResearch note

Overview

Illustration. One question travels through 12 parts and 12 connections, from the product team to the ledger and back as a decision.

An illustration of the research workflow described in the case. The platform stays unnamed, and the questions are examples.

The method in numbers

stages, from the public ledger to a decision
5
rules every figure follows
4
example questions, each run from script to decision
3
to answer a question from the source
A day

Process facts from the case page. The case is an illustration, so no platform figures are shown.

How each number was counted
5 stages, from the public ledger to a decision
The method on the case page: source, extract, clean and check, analyse, decide, 12 October 2026.
4 rules every figure follows
The method on the case page: full record, scripts versioned and rerun, tied to ledger totals, traceable figures, 12 October 2026.
3 example questions, each run from script to decision
The explorer on the case page, where the questions, markets and findings are illustrations, 12 October 2026.
A day to answer a question from the source
The case page: questions that used to wait in a queue for a data request, 12 October 2026.

The journeys, step by step

The same 6 journeys the map plays, written out. Each step names the parts it connects.

01A question becomes a scriptScripts · 4 steps
  1. 1The product team asks a question, for example which markets carry the trading and how deep their order books are.Product team → Claude
  2. 2Claude drafts the extraction script from the question.Claude → Extraction script
  3. 3A person reviews the script before it runs.Reviewer → Extraction script
  4. 4The script is kept under version control, so it can be rerun from the start when the question changes.Extraction script → Version control
02Reading the full recordPublic ledger · 3 steps
  1. 1The script reads the public ledger through its public interfaces.Extraction script → Public ledger
  2. 2Every trade, order and balance change in the period comes back. Nothing is sampled, so small markets and rare cases are counted too.Public ledger → Raw extract
  3. 3Raw records become a clean table: one row per event, with consistent types, units and market labels.Raw extract → Clean table
03Checked against the ledgerCheck, analyse, decide · 3 steps
  1. 1Before any analysis, the clean table’s row counts and totals are taken.Clean table → Reconciliation
  2. 2They are compared with the ledger’s own figures.Ledger totals → Reconciliation
  3. 3A mismatch stops the work until it is explained. Only matched tables go on to analysis.Clean table → Reconciliation; Ledger totals → Reconciliation
04From finding to decisionCheck, analyse, decide · 3 steps
  1. 1The matched table is analysed: liquidity, trading activity and how participants behave, market by market.Reconciliation → Analysis
  2. 2AI helps explore the data and explain what the numbers show. The finding goes into a research note.Analysis → Research note
  3. 3Each note ends in a recommendation the product team can act on, with every figure traceable back to raw records.Research note → Product team
05A follow-up questionScripts · 4 steps
  1. 1A follow-up changes the question.Product team → Claude
  2. 2The script changes with it.Claude → Extraction script
  3. 3The new version is kept alongside the old ones.Extraction script → Version control
  4. 4It runs again from the first record, so a follow-up costs minutes rather than a new data request.Extraction script → Public ledger
06Where AI is usedAI · 5 steps
  1. 1Claude writes and maintains the extraction scripts, work that would normally need a data engineering team.Claude → Extraction script
  2. 2A person reviews every script before it runs.Reviewer → Extraction script
  3. 3Nothing is trusted on its own: every table must match the ledger’s own totals first.Clean table → Reconciliation; Ledger totals → Reconciliation
  4. 4AI helps explore the data and explain what the numbers show.Analysis → Research note
  5. 5The product team makes the call, with every figure traceable to raw records.Research note → Product team

Want a system like this, documented like this?

Tell us about the process or the product. We'll reply within two business days with a first view of what we'd build.

Start a project