Controlled fixture conformance for the current beta ruleset.

Benchmark corpus version: 2026-06-26-r3p1.

This is not a production accuracy benchmark. It is a small controlled fixture conformance benchmark for static-rule regression coverage.

Controlled corpus result: 18 benchmark cases, 31 expected blockers detected, 0 false positives, 0 false negatives, 0 forbidden evidence hits, precision 1.00 against target 0.98, and recall 1.00 against target 0.95.

Benchmark coverage includes legacy Content API usage, Merchant API migration shape, ProductInput/dataSource mapping, product identifier risk, price and numeric handling, customBatch replacement risk, auth/retry/concurrency review signals, and clean-code controls.

Per-rule conformance is runner-derived for 19 measured rules: 31 expected instances, 31 detected, 0 missing, 0 unexpected, and 1.00 precision and recall per measured rule in this corpus.

Per-language coverage is case membership, not language-specific production accuracy: JavaScript has 8 cases; TypeScript and Python have 2 each; Kotlin and Google Apps Script have 1 each; C#, Go, Java, PHP, Ruby, and JSON each appear in one shared mixed-language fixture. Unsupported or skipped files remain NOT_VERIFIED.

Tested-scope formula: tested_scope_percent equals completed_static_rule_groups divided by applicable_rule_groups.

A rule group counts as completed only when scanner health passes, submitted files are analyzable, and the rule can evaluate the declared input type.

Skipped files, unsupported languages, dynamic runtime behavior, Merchant Center ownership, OAuth permissions, staging traffic, and unsubmitted repositories are reported as NOT_VERIFIED instead of counted as ready.

Perfect scores in this controlled fixture benchmark do not guarantee a production cutover. They only show that the current static ruleset matched the controlled test corpus for submitted-source scenarios.

Run free scan - no signup required See a complete sample finding View pricing