SOFT_ERROR // BOOT

> soft error

> booting…

> loading manifesto…

> 0 trackers attached

> ready. _

Field notebook // case studies & useful failures

Lab Notes

Anonymised client case studies and the engineering behind them — how each brief started, what we found, what we built, and what actually changed. Names and numbers withheld; the method is the point.

CASE 0004

A university research group Research · Data

Making a research dataset legible

A research group had brilliant data trapped in forty confusing tables. We turned the export into something a newcomer could understand by lunchtime.

A 40-table export became one documented, queryable API; onboarding a new researcher went from days to an afternoon

CASE 0003

A B2B SaaS support team SaaS · Applied AI

The triage layer that knew when to stop

A support team wanted AI to answer tickets. The valuable part was teaching it when to refuse and hand a human the wheel.

Routine tickets resolved automatically behind a hard human-handoff rule; no confidently-wrong answers shipped to customers

CASE 0002

A European logistics operator Logistics

Nine seconds to under one

An operations dashboard was so slow that staff kept a spreadsheet instead. We made the real thing faster than the workaround.

p95 first-meaningful-load fell from ~9s to under a second; the "is it broken?" emails stopped

CASE 0001

A UK consumer-payments startup Fintech

Nine screens to three

A payments startup was losing people at sign-up. We rebuilt onboarding around the one thing the user actually came to do.

Onboarding completion climbed and the "stuck at verification" tickets dropped to a trickle