[takeokunn]

Releases v1.6.2, a patch containing one change: the lint-fennel-janet-depth
cost gate no longer asserts a wall-clock ratio.

Why

cost_tests::a_rules_per_call_cost_does_not_grow_with_the_file asserted
ratio < 1.8 on ns-per-check-call at two file sizes. The rules it measures
cost 16–75 ns per call in a release build, so the assertion is decided by
machine load as much as by the code. Measured on the same unmodified tree:

Condition	fennel-redundant-do ratio
dev, isolated, --test-threads 1	1.05
release, idle	1.27
downstream nix sandbox build	2.04 — failed

Two mechanisms, neither of them a regression:

• at that magnitude the ratio of two wall-clock measurements is mostly
  scheduler noise;
• doubling the file doubles the tree, so the traversal's working set stops
  fitting in cache. Constant work per call, more time per call — which
  spends most of the 1.8 budget before any contention.

The rule it accused is correctly ordered: fennel_redundant_do/rule.rs
calls the cheap domain check first and the O(file) guard second.

What replaces it

• each_rule_is_dispatched_once_per_head_occurrence — asserts exact
  invocation counts, which the head index and dialect scope fix before any
  check body runs. Identical idle and loaded.
• ignored_bench_per_call_cost_does_not_grow_with_the_file — the ratio,
  #[ignore]d, printing rather than asserting.

The O(file)-per-call shape is still gated by
every_rule_stays_far_cheaper_than_a_materializing_one, which compares
against a deliberately mis-ordered control rule and has ~9,000× of headroom
against its 10× floor. That is the test that actually catches a guard moved
above its domain check.

This is the pattern six sibling packages already use — see
lint-introspection/src/lib.rs, which documents the same retreat and why.

Verification

Check	Result
cargo fmt -- --check	pass
cargo clippy --all-targets --all-features -- -D warnings	pass
cargo test --release -p paredit-feature-lint-fennel-janet-depth, idle	88 passed, 1 ignored
Same, under 32 spinners on 16 cores (3.3× slowdown)	88 passed
nix flake check	running locally at time of opening; CI is the gate

Control: under that load the benchmark showed
fennel-nested-associative-operator drift 1.03 → 1.39 on unchanged code,
confirming the ratio tracks load rather than the rules.

Release checklist

• Cargo.toml (both package and workspace.package) at 1.6.2
• Cargo.lock refreshed
• v1.6.2 in the install examples in docs/src/getting-started.md and
  docs/src/reference/compatibility.md
• Tag v1.6.2 on the resulting main tip after squash-merge
• Paste release notes into the draft release.yml creates, then publish