Capstone: map a real Pakistani government AI deployment against the seven prior lessons

Three ways to see it

1. 1

   Step one: pick a system. Choose one named Pakistani public-sector AI deployment that is at least mentioned in news, an annual report, or a procurement notice. Examples that work well: NADRA biometric duplicate detection. FBR audit case selection. PTA spam call detection. BISP eligibility scoring. PEC examination paper grading. Punjab Safe Cities facial analysis. Lahore Waste Management traffic routing. State Bank suspicious transaction monitoring. Higher Education Commission plagiarism scanner. Pick the one closest to your professional context so you can also approach insiders for clarification.

2. 2

   Step two: gather public evidence. Spend 60 to 90 minutes assembling everything publicly available. Press releases, news articles, parliamentary questions, RTI responses, vendor case studies, conference talks, the operating ministry's annual report, the federal budget line item if there is one, any procurement notice on PPRA. Write a half-page situation summary: what the system does, who built it, when it went live, who it affects, and what is publicly claimed about its accuracy and safety. This is your baseline. Most Pakistani government AI deployments will have a thinner public record than you expect, and the thinness itself is the first finding.

3. 3

   Step three: score against the seven prior lessons. Use a 0/1/2 scale per lesson. Lesson 1 transparency definitions: does the public record clearly state model, decision, and operational transparency posture? Lesson 2 model card: is one published or referenced? Lesson 3 bias audit: is there any subgroup-disaggregated performance reporting in Pakistani axes? Lesson 4 RTI: has anyone tested the disclosure pipeline, and are RTI responses on record? Lesson 5 procurement: does the procurement notice show the ten technical criteria? Lesson 6 data lineage: is the training data source named? Lesson 7 incident disclosure: has any incident been publicly disclosed? Maximum 14. Most current deployments will score under 6.

The why-tree

Try this with Claude

Capstone challenge — three actions to actually finish this. (1) Today, pick the system and write its name and one-paragraph description; commit to it in writing to a colleague. (2) Within seven days, complete the situation summary, the score table, and the gap-fix triplets. (3) Within 14 days, finish the one-page bilingual recommendation and walk one trusted civil servant through it; the conversation is the test of whether the work was useful.