Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
The harness reported “duplicates not canonicalized” because it counted how many results came back, not whether any results contained duplicate facts. It requested five results per query and treated an average count of five as evidence against canonicalization. Worse, the run had written only about 256 of 1,000 planned facts, so the planted duplicate records were never available to test. The duplicate verdict was therefore unsupported—not proof that canonicalization failed, or that it worked.
What did the harness actually measure?
In a scale probe described by Pranab Sarkar, the harness ran duplicate queries with top_k=5 and calculated an average of 5.0 returned items. The code used len(items): it measured the number of rows in each response, not whether the returned text included duplicate facts. Every query requested a five-item page, so a full page could produce an average of 5.0 regardless of duplicate behavior. Sarkar’s explanation of the query and metric
The described logic gathered results for queries named Q-dup-*, averaged each query’s returned value, and marked canonicalization false when the average reached two. Although a code comment described checking whether multiple results contained duplicated text, the reported logic did not inspect result text. A count of returned rows cannot answer a question about repeated content.
One example makes the mismatch clear: a query for a preferred “nord” color scheme returned five loosely related editor preferences. None was the target fact, and none duplicated another result. Five rows indicated that the requested page was full; it did not establish what the rows contained.
#1 Best Overall
Why couldn’t this run test the planted duplicates?
The probe attempted to write 1,000 facts but reportedly completed only about 256 successful writes. Its planted duplicate records were IDs 901–1000, beyond the portion that had been written when the run stopped. The transcript described repeated backpressure retries and a timeout around fact 357. Thus the duplicate examples the harness was supposed to assess were absent from the test store. The reported write counts and test-data placement
These are two separate flaws: the metric confused page size with duplicate content, and the intended duplicate inputs had not been stored. Together they mean the run did not establish whether duplicate canonicalization succeeds or fails. The proper result for that claim is unknown.
Rank #2
- Features Over 160 Latin Songs
- Arranged for C Instruments
- Standard Notation
- 48 Pages
Was the write stall a real problem?
Yes, according to Sarkar’s account. He attributes the stall to Python constructors that created an engine without starting its background worker pool. Without the compactor, the in-memory delta tier filled at 256. The article says an engine change dated June 27, 2026 fixed the issue and that the plugin’s minimum engine version at publication was 0.12.1. These are the author’s explanations and timeline; they are not independently confirmed here. Sarkar’s account of the stall and engine change
Free tools Windows power users keep installed
One-click scans. No signup required.
The write problem should not be turned into evidence that canonicalization failed. It explains why the duplicate test inputs were missing; the invalid page-count metric explains why the harness produced a negative verdict anyway.
Rank #3
What does the separate 16/20 score tell you?
The article also reports precision-at-five of 16/20 for a separate retrieval check over facts that were successfully written. Sarkar notes that this used word-overlap scoring on a templated corpus and had not been audited hit by hit. It is an author-reported result with that limitation, not validation of the duplicate test or a general performance benchmark. The separate retrieval result and its qualification
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to evaluate a benchmark verdict
When a harness says a behavior failed, check that the metric measures that behavior and that the test cases actually reached the system. Reproducibility alone is not enough: a harness can repeatedly return five rows and reproducibly calculate 5.0 while measuring only page fullness.
- Inspect the assertion: For duplicate canonicalization, compare result identities or normalized content for repeated facts. A result count measures quantity, not duplication.
- Confirm the executed corpus: Report successful writes, failed or pending writes, and whether the specific records targeted by each query were present.
- Trace the verdict to evidence: Link the conclusion to the raw response that supports it, rather than relying only on a summary number.
- Use “unknown” when prerequisites are missing: If a run never wrote the cases needed to test a claim, it cannot support either a positive or negative verdict.
Sarkar puts the standard plainly: “A verdict should cite the raw response that backs it, and a verdict computed from a run that never wrote its test inputs should say unknown.” The author’s statement on evidence and verdicts
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
