Skip to content

Foray 382 · Essai 03 · US federal IT portfolio

Why do nine projects share one missed measure?

The FDA investment for Mission Accomplishments and Regulatory Compliance Services, 009-000005317, pays for nine project records. A performance report shows the same “Not Met” beside each. Before the portfolio meeting, reopen both saved graphs. Can they still tell us which record owns the result?

Two snapshots, one recorded outcome

Historical exportWhole graph: nodes / directed edgesFDA projectsLinked measuresProject–measure pairs
Health2,796 / 3,6779545
Complete18,728 / 28,771919

Ask the investment owner about measure 3481. It records the number of imported products screened per year and has the recorded result “Not Met”. Pairing nine projects with it makes nine rows. Each row reaches the same measure through the same investment.

Inspect the identities that survived the reload

Both exports returned project IDs 2150, 2151, 2152, 2154, 2155, 2156, 2157, 2160 and 2167. Health also links measures 3479, 3480, 3482 and 3483; their outcome fields are empty. Those four records explain the extra 36 pairs. Why their coverage differs remains a question for the data steward.

Ask the live graph

Paste the comparison into Aura Query. Each export has its own copy of the investment, tagged with its snapshot. The query counts unique source IDs and ID pairs. The fresh result should reproduce both rows above and give ["3481"] as the lone missed measure in each.

MATCH (i:F382Historical:Investment {Identifier:'009-000005317'})
WHERE i.replayCase IN ['health', 'complete']
MATCH (i)-[:pays_for]->(p:F382Historical:Project),
      (i)-[:judged_by]->(m:F382Historical:Metric)
WHERE p.replayCase = i.replayCase AND m.replayCase = i.replayCase
RETURN i.replayCase AS snapshot, count(DISTINCT p.Id) AS projects,
       count(DISTINCT m.ID) AS measures,
       count(DISTINCT [p.Id, m.ID]) AS pairs,
       collect(DISTINCT CASE WHEN m.Metric_results = 'Not Met'
                             THEN m.ID END) AS missed
ORDER BY snapshot;

Reopen the portfolio

Follow the workshop connection guide. To load these two source files alone, run python3 replay.py federal. Once populated, run python3 replay.py federal --check-only. That checks both database counts, all nine project IDs, every linked measure outcome and both pair counts. We ran the loader twice on the same Neo4j 5.26 database and then ran the read-only checks. All identities and counts were reproduced.

Source IDs restart between the exports. The loader keeps a separate key space for each and uses a stable record number for each relationship. This preserves even the parallel edges in the source. It also prevents a repeat run from growing the graph. These snapshots are historical research-and-development material from Lawrence’s model of the 2013–2018 IT portfolio, prepared in 2020. They describe the coverage and recorded outcomes of those exports.

Download the replay · Inspect the original FDA evidence in Foray 324 · Historical source shelf and research-use terms.