Skip to content

Foray 324 · Purpose

What should a departmental IT reviewer investigate?

End

At the end of a portfolio review, a departmental lead has time to call one person. Use relationships between an agency, an investment, the projects it funds, its contracts and its measures to decide what to ask and who could resolve it. A useful answer names the record behind the question. An apparent problem can disappear once the reviewer counts the right thing.

Ways

Start in the Health extract or choose a question from the departmental scenarios. State the decision, the unit being counted, the source file and the evidence that would change the reviewer’s next move. Follow one record by hand. Then choose a method from Tomaž Bratanič’s notebooks and read his account of why it works. Keep both the hand-checked answer and the method’s answer. Change one feature, such as including unnamed records or counting a repeated link only once. Show the reader what happens. Choose a departmental review.

Keep or collapse repeated relationships

Bratanič’s multigraph example compares projections that retain each relationship with projections that combine them. For FDA, connect the project and measure records through their common investment and compare the resulting row count with distinct measure IDs. For Interior, change the unit again from contract to supplier and see which review question survives. Article, 30 March 2020 · Multigraph notebook.

Ask a question in ordinary language

Bratanič supplies a model with the graph’s labels, properties and relationship directions. For Education, ask which projects the student loan database investment pays for. Compare the returned project IDs with a hand trace. Run the same question with the wrong relationship direction and check whether an empty result prompts a schema check. Article, 2 May 2023 · Schema-to-Cypher notebook.

Test a grouping across periods

Bratanič uses the previous period’s communities to initialise the next period’s groups. A reviewer could ask whether investments linked by suppliers stay together when the time window changes. First establish comparable dates and supplier identities. The Health and complete exports overlap. Treating them as successive observations would manufacture a time series. If dated relationships cannot support two comparable windows, record that limit. Article, 25 November 2019 · Community detection notebook.

For each further essai, look for an unused method in the original repository that would sharpen a particular departmental decision. A second department can also challenge a result that appeared straightforward in the first. Publish the question, reviewed method, data slice, comparison, result, limits and the next question a reviewer would ask. Bratanič’s repository · Bratanič’s articles.

Means

Use Lawrence’s 2020 US IT portfolio model and its independently accessible Library sources. He credits Projecting Success / Project Hack and Oxford Business School for the US government IT records, described as covering 2013–2018 and restricted to research and development. His preparation omitted incomplete rows. The full export, Health extract and FDA files preserve different selections. Missing edges, unnamed records and unreported results affect what a reviewer can count. A CIO evaluation concerns an investment. Interpreting it as customer satisfaction or a present-day rating requires evidence outside this collection. Illustrated walkthrough and source corrections · Public Library source shelf.

Foray 323 asks which project knowledge travels between settings. Its portfolio essai returns to Lawrence’s earlier question about a proposed project intervention. Here we use a different body of records to ask which investment merits a follow-up and whether the graph gives enough evidence to justify one. Read the neighbouring portfolio question.