Perfect data kills projects

I've helped launch infrastructure disclosure portals across Uganda, Mozambique, Thailand, and Honduras. Here's what I've learned about what actually works.

1. People don't want perfect data. They want something

The pattern repeats everywhere. In Kaduna State Nigeria, we published 900+ projects with incomplete data fields. Usage didn't wait for perfection. It started the day we went live.

2. The hardest part isn't building the portal

Sustainability beats launch quality every time. Ecuador's portal has 30,000+ projects because they baked updates into the Ministry's quarterly reporting cycle. It's not glamorous. It works.

3. Standards enable speed, not slow it down

OC4IDS gave us a headstart. Mozambique's team reused Uganda's field mappings. What took Uganda 8 weeks took Mozambique 3 because the schema was already proven.

4. Users don't read methodology. They trust logos

Institutional credibility is inherited, not earned per-project. We still publish the methodology. But we stopped expecting people to read it before trusting the numbers.

5. Contractors notice before citizens do

The supply side watches itself. That's not a bug. It's a feature. When contractors know their performance is public, behavior shifts before activists even see the data.

6. Search matters more than visualization

Utility beats aesthetics. We spent 2 weeks building an interactive budget treemap. It got 47 views in 3 months. The filterable project table got 4,000+ searches in the same period.

Users search for specific projects. They do not browse dashboards. Build search and filter before visualization.

7. Incomplete beats invisible

Iteration happens in public. Thailand's portal improved because users could see what was missing and tell us. Waiting for completeness meant waiting forever.

The pattern: Imperfect launch, sustained attention

Perfect planning never survives contact with real users. The portals that matter most weren't the ones with flawless schemas. They were the ones that launched imperfectly, listened hard, and kept showing up.

If you're building infrastructure transparency systems, this is my offer: Stop waiting for perfect. Start listening in public. The gap between your data and the ideal spec is less important than the gap between silence and something.

Want to compare notes? I've seen what works across four continents. Let's talk about your context, not my theories.

Playbook

Decision Table

Option	When to Use	Tradeoff
Launch at 70% completeness	Data exists but has gaps in non-critical fields	Immediate usage, visible quality gaps
Wait for full coverage	Legal or audit requirements mandate completeness	Delayed impact, possible scope creep
Phased rollout by entity	Some agencies are ready, others are not	Partial visibility, manageable scope

Execution Checklist

• Define the minimum viable field set for first publication.
• Set a fixed launch date independent of data quality milestones.
• Build search and filter before visualization and dashboards.
• Document the update process before launch, not after.

Failure Modes

• Launch is delayed repeatedly while chasing 100% field completeness.
• Visualization is prioritized over basic search and filter.
• Nobody documents the update process until the original team leaves.

Continue Reading

LLMs Can Read 100,000 Contracts. Here's Why That's Not Enough.

LLMs accelerate pattern detection in procurement data. They also accelerate false confidence. A practitioner's field guide to what works, what breaks, and what to build instead.

Why Transparency Portals Die

The most successful infrastructure disclosure system I helped build stopped updating eighteen months after launch. Technology worked. Data standard was correct. Two years later, the portal loads. Nobody maintains it. After a decade building these systems across Africa and Southeast Asia, here's why disclosure portals die - and what survival requires.

Five Ways Portals Die

I can point to procurement disclosure portals across three continents that launched to press releases and ministerial speeches, then went dark within three years. The graveyard is bigger than the survivors. After a decade building these systems, here are the five ways they die—and what survivors share.