The research loop
- 01QuestionA narrow question that evidence can answer
- 02MapThe real workflow, data, and people
- 03PrototypeThe smallest system that can answer it
- 04EvaluateTest sets, rubrics, and shadow runs
- 05BuildTurn it into a product capability
- 06PublishFrameworks and findings, shared openly
- LogEvidence logEvery step is recorded. Each loop ends in a decision: continue, build, or stop.
- Then back to step 01 with the next question.
Every pattern we publish or build into Marketix starts as a narrow question and moves through the same six-step loop. Most questions stop early, and that is by design: a fast, honest no is as valuable as a yes.
What each part does
Question
We write a single question that evidence can answer, and decide in advance what result would change our minds.
Map
We document the real workflow: inputs, outputs, people, data, and where decisions happen. Unstable workflows stop here.
Prototype
We build the smallest system that can answer the question, with read-only tools wherever possible and full logging.
Evaluate
We test against realistic cases with rubrics, compare with a baseline, and run in shadow mode when real inputs matter.
Build
Patterns that hold up across several loops become Marketix product capabilities, sometimes developed with partner teams under a written agreement.
Publish
We share frameworks and findings, including negative results, so the wider field can learn from them.
Evidence log
Every question, build, test, and result is recorded. Decisions are made from the log, not from memory or enthusiasm.
See the architecture in the product
Marketix is built on these diagrams: specialist agents, governed connectors, and a hub that plans, routes, and records every run. Sign up for early access.
