docs(benchmarks): verify crystal, elm, and ponyc targets

[mstykow]

Summary

• Verify three new compare-outputs repository targets not already in docs/BENCHMARKS.md: crystal-lang/crystal, elm/compiler, and ponylang/ponyc.
• Record end-state timings and Provenant-vs-ScanCode advantages; regenerate the benchmark chart (255 runs).
• Drive-by: reject changelog copyright-year bump prose and related holder/author junk exposed while triaging Crystal (start-anchored so real notices with embedded “update copyright year” instructions are kept).
• compare-outputs triage surface: flat ScanCode/Provenant-favored review queues, review_queue_summary, stdout triage pointer, and run-manifest.json duration fields — so agents inspect artifacts instead of writing ad-hoc scripts.

Scope and exclusions

• Included: benchmark entries for the three targets; changelog/contribution-guide junk filtering in copyright refine + final refine path; compare-outputs review queues + duration parsing; focused unit coverage; verify-benchmark-target skill pointer updates.
• Explicit exclusions: no Crystal shard.yml package parser (tracked in Support parsing: Crystal shard.yml #1312; ScanCode also emits 0 packages on this snapshot); no broader literary-quote author cleanup on Elm's Quotes.hs.

How to verify

• Spot-check the three new docs/BENCHMARKS.md entries for present-tense end-state wording and alphabetical placement under Julia / Nix / Haskell / other ecosystems.
• Optional: rerun one cached compare, e.g. cargo run --manifest-path xtask/Cargo.toml --bin compare-outputs -- --repo-url https://github.com/crystal-lang/crystal.git --repo-ref 82b5094b716b1e9a3f4fe6c601fd48272fc35d01 --profile common, and confirm (1) changelog copyright-year bullets stay absent and (2) comparison/samples/scancode_favored_review_queue.json plus run-manifest.json *.duration_secs are present.

Follow-up work

• Created: Crystal shard.yml package extraction — Support parsing: Crystal shard.yml #1312.

Expected-output fixture changes

• Files changed: none (junk filtering and compare artifacts covered by unit tests; no golden/fixture updates).

Compare runs (local):

• Crystal: .provenant/compare-runs/20260719T171119Z-crystal-9555
• Elm: .provenant/compare-runs/20260719T171137Z-compiler-10473
• Ponyc: .provenant/compare-runs/20260719T171145Z-ponyc-11639

Generated with Cursor.

Made with Cursor

[greptile-apps]

Greptile Summary

This PR updates benchmark documentation and compare-output triage artifacts. The main changes are:

• Adds Crystal, Elm, and Ponyc benchmark entries and refreshes the chart.
• Adds flat ScanCode-favored and Provenant-favored review queues to compare artifacts.
• Records optional scanner duration fields in compare run manifests.
• Narrows changelog copyright-year junk filtering and adds focused tests.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
src/copyright/refiner/copyrights_junk_patterns.rs	Adds narrower changelog copyright-year junk filters with coverage for real notices that contain update-year prose.
src/copyright/refiner/holders_junk_patterns.rs	Adds targeted filters for changelog and PR holder fragments.
src/copyright/refiner/authors_junk_patterns.rs	Adds filters for contribution-guide and month prose mistaken as authors.
src/compare.rs	Adds diagnostic review queue artifact generation and summary pointers.
xtask/src/bin/compare_outputs.rs	Adds optional duration parsing for compare run manifests.
xtask/src/manifests.rs	Extends scanner manifest records with optional duration fields.
docs/BENCHMARKS.md	Adds new benchmark rows and updates aggregate counts.

_{Reviews (4): Last reviewed commit: "fix(copyright): keep junk filter out of ..." | Re-trigger Greptile}

[mstykow]

@greptileai review