Arid 2.0 keeps the detector and rebuilds the plumbing around it

A baseline file is a database keyed on whatever identity your linter hands out, which is why keying it on path and line is a slow-motion bug. Insert twenty lines at the top of a file and the physical coordinates of a duplicate move while the duplicate does not ^[11]; a pipeline told to fail on new findings now has one. Arid 2.0's answer is `arid-finding-v1:sha256:...`, computed over the normalized duplicate content and held stable across path, physical line number, occurrence ordering and multiplicity, structural metadata, output format, and worker mode ^[12]. The author's own framing is that this is dull in a CLI report and useful when a CI system or a coding agent has to decide whether it has seen a finding before ^[14].

The scoping design is where the trap is. Analysing only `src/payments/` when its duplicate lives in `src/customers/` removes half the evidence, so a path argument that narrows the scan is not the same feature as a path argument that narrows the report ^[15]. The post says v2 separates the two, and the copy available to me breaks off mid-sentence at exactly that point ^[16], so read the mechanism as stated intent rather than documented behaviour.

The maintainer's cost accounting is more interesting than his headline multiple. If v2 sits at roughly 219x Pylint on the isolated duplicate-code job ^[9] and carries at most a few percent over the qualified 1.2 build ^[10], then 1.2 was somewhere near 230x ^[18], and the entire cost of v2's contracts, baselines and workflow machinery is about two hundredths of one percent of the Pylint run it displaces ^[19]. That is the arithmetic underneath his line that "at some point optimization becomes an excellent way to avoid working on things users actually need" ^[17].

What none of this changes is reach. The detector still matches exact Python source after configurable Python-aware normalization and still reports DUP001 ^[3], with comments, docstrings, imports and function signatures available for exclusion from duplicate identity ^[5], and it explicitly does not chase semantic equivalence or fuzzy AST clones ^[4]. A copy edited past the normalizer stays invisible, by design.

The benchmark deserves the same plain reading. Every figure comes from the tool's author, using the pinned corpora and Hyperfine methodology that qualified 1.2, against Pylint 4.0.6 running serially ^[7], with the comparison narrowed to Pylint's duplicate-code functionality rather than a whole Pylint run ^[8]. That narrowing is the honest way to do it, and it also fixes the scope of the offer: one check moves out of Pylint, and the rest of the Pylint install stays wherever it already was.