Why
Follow-up to #593/#594 (branch fix/retire-sonnet-model-default-delete-contradicted-benchmark-claim, trajectory issue #37). The plugin repo carries a full flawed-era benchmark narrative that our newer campaign contradicts, and it's intertwined: deleting docs/contributing/BENCHMARK.md alone breaks links in 9 files. Decision: retire it from the plugin repo entirely and let the separate benchmarks repo be canonical for methodology + receipts.
Scope
Acceptance
Notes
Keep the tests/l7-benchmark/ runner only if there's a reason to keep an L7 layer in-repo; current decision is full move to the benchmarks repo. Confirm at PR time.
Why
Follow-up to #593/#594 (branch
fix/retire-sonnet-model-default-delete-contradicted-benchmark-claim, trajectory issue #37). The plugin repo carries a full flawed-era benchmark narrative that our newer campaign contradicts, and it's intertwined: deletingdocs/contributing/BENCHMARK.mdalone breaks links in 9 files. Decision: retire it from the plugin repo entirely and let the separate benchmarks repo be canonical for methodology + receipts.Scope
docs/contributing/BENCHMARK.md.tests/l7-benchmark/(the whole tree — README.md, RESULTS.md, tasks/, run-l7*.sh, lib/). Its canonical home is the benchmarks repo.RESULTS.md) into the benchmarks repo first, so nothing measured is lost (honesty bar: receipts are preserved, not discarded).BENCHMARK.md/tests/l7-benchmark:docs/reference/REFERENCE.md:17, and any others surfaced bytests/l1-lint/link-check.sh.CHANGELOG.md(v0.7.0-rc.3 'Measured' section, line ~149: '~61% token reduction … 8/8 … beats raw 6/8 · $10.31 … See docs/contributing/BENCHMARK.md') — it repeats numbers the project has retracted and points at a deleted file.docs/index summaries that mentioncontributing/BENCHMARK(e.g.docs/README table).Acceptance
bash tests/l1-lint/link-check.shpasses (no dangling links).Notes
Keep the
tests/l7-benchmark/runner only if there's a reason to keep an L7 layer in-repo; current decision is full move to the benchmarks repo. Confirm at PR time.