Seven programme case studies, one pattern: work handed over as execution that turned out to be a bigger problem than the ticket said.
| Case study | What changed | Result |
|---|---|---|
| Redline: contract review that shows its work | Reviews a freelance contract before it's signed, cites the exact clause behind every flag, and drafts the counter-offer | 5 of 6 risky clauses caught |
| The £6M Thread Pull | Turned a small circuit migration into an audit of Vodafone's whole third-party link inventory | Over £6M recurring opex |
| The Macro | Watched an engineer write per-site scripts by hand, then automated the parts that never changed | 20–25 min → under 5 min |
| Huawei out of the Core | Moved core traffic onto Vodafone's own Ciena network and ended the dark-fibre contract | £3.6M of the £6M |
| Vendor incentives | Payment tied to validated completion at each site, agreed directly with the vendor's director | Per-site sign-off |
| Neither Two Weeks Nor Five Months | Countered an unrealistic two-week deadline with a scoped 2.5-month plan, instead of a flat no | 2.5 months |
| The Light Test | Added a physical end-to-end light test to the migration process, after a circuit that existed in no database failed twice | New standard step |
| Poka-Yoke | Paired struggling engineers with an expert and mistake-proofed the process, instead of managing them out | Retained, not replaced |
| This website, built with Claude Code | A portfolio where every claim traces back to a written source, designed, built and deployed by me with Claude Code | Live, self-deployed |
| A progress tracker that flags the slips | One private record of my child's schoolwork, tracked against the UK Year 2 curriculum, with practice papers that flag skills needing review | In use at home |
For programme roles
- Redline: contract review that shows its work
Run like a programme: 10 decision records, 12 tickets each shipped as working, tested code, and a live test against planted risky clauses.
Discovery angle: An AI reviewer is only useful if you can check it. The product is built around a property a reviewer can verify, not around how much it finds.
- The £6M Thread Pull
Thousands of circuits, a team of 9, three-way verification and a rollback plan on every change.
Discovery angle: Opportunity spotted inside a delivery task, sized, and argued as recurring opex rather than one-off capex.
- The Macro
Around 90–120 engineer hours a year recovered, and still in use when the programme ended.
Discovery angle: Discovery by watching real work, then separating what varies from what doesn't. The same abstraction behind most automation and AI tooling.
- Huawei out of the Core
High-consequence core change under government scrutiny, de-risked with a proven verification method. Active SC clearance.
Discovery angle: A compliance mandate treated as a commercial decision. Meeting it without buying replacement capacity.
- Vendor incentives
Vendor governance: mandatory site templates and validation before payment, still in use today.
Discovery angle: Solved with incentives rather than escalation, so the vendor's commercial interest and our delivery needs pointed the same way.
- Neither Two Weeks Nor Five Months
Delivered in 2.5 months against a 4-5 month baseline by parallelising work, pre-booking critical resources, and triaging by risk, with zero escalations.
Discovery angle: A counter-proposal, not a refusal: reframed 'yes or no' into 'what's actually achievable,' backed by a real plan.
- The Light Test
Root-cause discipline: two failed remote fixes, then a physical trace between exchange buildings, then a new standard step added so it couldn't happen again.
Discovery angle: The audit behind this programme was verified three ways, and none of them could see a circuit missing from every record. Fixing your own process when reality disagrees with it is rarer than getting it right the first time.
- Poka-Yoke
Paired underperformers with an expert, audited their output for two weeks to confirm real progress, and rewrote instructions so steps couldn't be skipped.
Discovery angle: Poka-yoke, mistake-proofing a process so an error becomes structurally hard to make, is the same instinct behind good systems design: don't rely on people to be perfect, design so mistakes are hard to make.
- This website, built with Claude Code
Run like a small programme: a written plan for SEO, a recorded design system, a decision log, and frequent small deploys.
Discovery angle: A portfolio is only as credible as its weakest claim. The product is built around a rule that every number and story traces back to my own experience record.
- A progress tracker that flags the slips
Built in phases from data model to reports, with a written product brief and fixed curriculum terminology. 24 commits over four weeks.
Discovery angle: Progress that only moves forward hides slips. The product adds a separate, self-clearing flag so a skill that's slipping shows up without rewriting the record.