Same month, different case

Sort the rows by date and mark the ones that share a month. Add a column for what each matter is about, in the source’s own words, not the report’s.

Ask what each kind of link would need: a source that says the matters are connected, the same alleged conduct, or the same case in court. Similar timing, the same unit and the same place are not on that list.

Separate what happened (officers, a court order, documents) from what it was called. Then check who is speaking in each source.

Look at the distance in Source A. Then ask what a place of arrest could mean other than “this is where the crime happened”.

If you scored six different reports and every score came out identical, what would you conclude about the scorer?

Write out Query 1’s seven words. Tick each one off against the text of Source B, then C, then D. Does any single source contain all seven?

1. Timeline.

Three events fall in September, across two years. Two of them are in September 2024 and are different matters.

2. Claim table.

3. The word “raid”. Both sources agree that officers of the Team came to the council on the morning of 17 September 2025 with a court order and the council handed over specific documents. “Raid” is the word of two opposition parties, quoted in Source B. The council rejects it in Source C, and the Team spokesperson in B describes executing a court order for documents. Claim 3 should say what happened and note that the label is disputed. Printing “raided” as fact adopts one side’s word.

4. Where someone is arrested. Source A says the four work at the municipality’s head office in Stonebridge, about 90 km from the provincial capital, and were arrested in Alder Cross. A place of arrest tells you where officers found the suspects. It does not tell you which case they belong to, and the sources do not say why the arrest happened there. The report matched a place name instead of a case.

5. The numbers and the search.

(a) A scorer that gives six different reports the same number is not measuring how they differ. The evidence confidence of 0.30 is then the average of one repeated value. It cannot tell a strong case from a weak one, and the pack suggests it is a default rather than a measurement.

(b) Query 1 has seven words: alder, cross, council, tender, corruption, police, raid. Source B has six of them but not “police”. Source C has “council”, “tender” and “raid” (and “Alder Cross” in its heading) but not “corruption” or “police”. Source D has “police” and the council but not “tender”, “corruption” or “raid”. No single source has all seven, so a search that needs every word finds nothing. News also describes one event in different words: “visited”, “court order”, “documents”, “search warrant”. An empty search shows that those words did not match. It does not show that the event did not happen.

6. Sample verdict. The pack supports a real arrest of four Stonebridge officials by the Northmark Serious Corruption Team on Wednesday 15 April (reported 16 April), in a case traced to incidents reported in September 2024. It supports a court-ordered document request at Alder Cross City Council on 17 September 2025, which the council says was not a raid, and a separate bank-collapse search warrant at the council in September 2024. It does not support the report’s links: same month, same unit and a shared place of arrest are not evidence of one investigation, and no source says the matters are connected. The 0.30 scores are identical and measure nothing. I cannot tell whether a link exists, only that this pack does not show one. All matters are allegations.

Scoring (10 points, pass at 7).

• Timeline, 1 point: for four correct events with dates and matters.
• Claim table, 3 points: 3 for seven correct rows with sources, 2 for five or six, 1 for three or four, 0 for fewer.
• The word “raid”, 1 point: for separating what happened from what it was called.
• Where someone is arrested, 1 point: for saying a place of arrest does not show which case a person belongs to.
• Numbers and search, 2 points: 1 for saying identical scores measure nothing, 1 for explaining the empty search and saying an empty search is not an empty world.
• Verdict, 2 points: 1 for what the pack supports and what the report got wrong, 1 for stating limits (allegations, no link shown is not no link).

Debrief

Six habits this exercise is built to teach:

• The same month is not the same case. Three events fell in September and were three different matters.
• The same unit is not the same investigation. A police unit runs many cases at the same time.
• Where someone was arrested is not what they are accused of. A place of arrest is a fact about the arrest.
• A label is a claim. “Raid” is a word someone chose. Say who chose it.
• An empty search is not an empty world. Try the event in different words before you write “no source found”.
• A score that never changes measures nothing. If every item scores the same, the number is a default.

For the checks this builds on, try Exercise 004: One arrest, wrong map. For the real audit and rebuild behind this exercise, read A polished report is not a checked report and its follow-up, Checking the check.