Foray 323 · Essai 01
Which handover lesson belongs on the agenda?
Merebridge Station is handing over its new passenger-information displays. Searching earlier station records for the exact asset name finds two lessons. Neither answers the handover lead’s question. Two lessons filed under different names do.
Friday’s agenda
- Ask for a separate evidence image against each installed display ID. At Alder Quay, the reviewer could not tell which display a cabinet photograph belonged to. The team indexed a destination-change image under each ID. The later minute AQ-HO-17 records acceptance and supports lesson AQ-17.
- Ask for the station ID beside each supplier serial number. At Fen Vale, two entries could only be found under the supplier’s numbering. Closeout record FV-CR-09 shows what the station librarian accepted and supports lesson FV-09.
These are questions to take to the receiving team. The records establish what worked at the two earlier stations. Merebridge still has to agree its own evidence requirements.
The seven-record register
Read what each search brings back
Start with the exact asset name. Then include aliases and see which records appear. Open any record to inspect the wording a reviewer would have to defend.
| Read the register | Records returned |
|---|---|
| Exact recorded name | EJ-04 and HC-06 |
| Preferred and alternative names | AQ-08, AQ-17, EJ-04, FV-09, FV-12 and HC-06 |
| Relevant, current and supported | AQ-17 and FV-09 |
Why the other records matter
One word, two assets. Fen Vale’s FV-12 just says “screen”. In this register it could mean a passenger display or a privacy partition. The reviewer needs an asset ID. A wider search helps find the entry but cannot supply its missing identity.
One asset, two instructions. AQ-08 once accepted one photograph for a pair of displays. The controlled index links AQ-08 to its replacement, AQ-17. The later source minute accepts the revised evidence pack. Both entries use the same asset label. Following that relationship changes what the next project learns.
One unanswered question. Harbour Cross’s unsigned HC-06 note recalls a request for daylight photographs. The missing minute is reason to ask for corroboration. It cannot establish whether the request was ever accepted.
How the answer was produced
The seven extracts, dates and acceptance decisions were invented for this exercise. A case-insensitive match on the recorded asset name returns two records. A second search follows the preferred and alternative names in a small SKOS vocabulary and returns six. The selection rule then checks whether the term identifies a single asset, whether the record was superseded, whether the installation setting matches, and whether the source provides accepted evidence.
The rule is applied to the dated register as at 21 November 2026. For this closed example we treat the list of superseding entries as complete. A real team would need to establish who owns that register and when it was checked. Two results agreeing with two deliberately authored cases is a check on the implementation. It gives no estimate of accuracy on a real portfolio.
The first run adapts the ontology inspection, label lookup, text search, RDF and SPARQL work in steps 19 and 22–25 of Michael K. Bergman’s Cooking with Python and KBpedia. In his 2020 trial, Lawrence found KBpedia’s luggage concept only after changing how the search handled case. Here the additional problem is deciding whether a successful match is relevant enough to act on. Bergman’s original step 23 search notebook · All 65 source steps and their project applications.
Reproduce the run
Save the four linked files in one folder. With Python 3.12, install rdflib==7.6.0 and pyparsing==3.3.3. Run python run.py --check to execute the SPARQL query, classify the records and compare the result with the saved run. The check also grants corroboration to HC-06 and removes the vocabulary match for FV-09 in memory. The first moves onto the agenda and the second needs clarification. Run python run.py to regenerate the saved result.