diff --git a/CHANGELOG.md b/CHANGELOG.md
index 069bae9..590facc 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,10 @@ Please choose versions by [Semantic Versioning](http://semver.org/).
 * MINOR version when you add functionality in a backwards-compatible manner, and
 * PATCH version when you make backwards-compatible bug fixes.
 
+## Unreleased
+
+- fix(rules): `go-errors/no-fmt-errorf.yml` rewritten as a structural rule. The original `pattern: fmt.Errorf($$$ARGS)` was parsed by tree-sitter Go grammar as a `type_conversion_expression` (because `Type(arg)` is a valid Go type cast at pattern-compile time), so the rule matched no real call sites — silently emitting zero findings since the YAML shipped. Replaced with `kind: call_expression` + structural `selector_expression` match on `fmt.Errorf`. Verified against scenario 004's fixture (`pkg/scenarios-test-fixture/violations.go` on bborbe/maintainer#2): the `Boom` function's `fmt.Errorf` now fires. Scenario 004's `findings_count` floor lifted ≥4 → ≥5.
+
 ## v0.15.0
 
 - feat(pipeline): dispatcher refactor for `/coding:pr-review` + `/coding:code-review` — Step 4 replaced with `ast-grep-runner` (mechanical funnel) → per-Owner LLM-tier adjudication → citation validation. Decouples LLM-call count from PR file count; small PR-size now equals small LLM-call count for the same rule coverage. Migrated 13 rule-enforcer agents to the dispatcher contract.
diff --git a/rules/go/no-fmt-errorf.yml b/rules/go/no-fmt-errorf.yml
index be0c946..73d2ece 100644
--- a/rules/go/no-fmt-errorf.yml
+++ b/rules/go/no-fmt-errorf.yml
@@ -6,7 +6,34 @@ message: |
   Use errors.Wrapf (wrapping) or errors.Errorf (new error) from github.com/bborbe/errors instead.
   See docs/go-error-wrapping-guide.md (RULE go-errors/no-fmt-errorf).
 rule:
-  pattern: fmt.Errorf($$$ARGS)
+  # ast-grep 0.43 + Go tree-sitter grammar parses bare
+  # `fmt.Errorf($A)` / `fmt.Errorf($$$ARGS)` as a
+  # `type_conversion_expression`, because `Type(arg)` is a valid Go
+  # type cast at pattern-compile time (no surrounding expression
+  # context disambiguates). Real call sites like
+  # `return fmt.Errorf(...)` parse as `call_expression` because the
+  # surrounding context fixes the parse. Net effect: the unanchored
+  # pattern matches NOTHING in real code — verified empirically.
+  # `kind: call_expression` alone does NOT fix it (kind filters
+  # output; pattern still mis-parses). Use a fully structural rule
+  # instead: match `call_expression` whose function field is a
+  # `selector_expression` with operand `fmt` and field `Errorf`.
+  # Surfaced 2026-06-03 during scenario 004's walk — the YAML had
+  # been silently emitting zero findings since the rule shipped.
+  # See [[ast-grep - Structural Code Search]] (Gotchas) for the family.
+  kind: call_expression
+  has:
+    field: function
+    kind: selector_expression
+    all:
+      - has:
+          field: operand
+          kind: identifier
+          regex: '^fmt$'
+      - has:
+          field: field
+          kind: field_identifier
+          regex: '^Errorf$'
 ignores:
   - "main.go"
   - "**/main.go"
diff --git a/scenarios/004-findings-exist-path.md b/scenarios/004-findings-exist-path.md
index 047caa9..8525089 100644
--- a/scenarios/004-findings-exist-path.md
+++ b/scenarios/004-findings-exist-path.md
@@ -16,7 +16,7 @@ The stable fixture is [bborbe/maintainer#2](https://github.com/bborbe/maintainer
 - `go-concurrency/no-raw-go-func` (bare `go func(){...}()`)
 - `go-errors/no-fmt-errorf` (`fmt.Errorf` in production code)
 
-The PR stays open in perpetuity — the title says so. Walking 004 = re-pointing the dispatcher at the same SHA and verifying the funnel still surfaces all 5 violations.
+All 5 violations now fire after the no-fmt-errorf YAML's structural rewrite (was silently parsing as `type_conversion_expression` until the 2026-06-03 fix). The PR stays open in perpetuity — the title says so. Walking 004 = re-pointing the dispatcher at the same SHA and verifying the funnel still surfaces all 5 violations.
 
 ## Setup
 
@@ -39,7 +39,7 @@ The PR stays open in perpetuity — the title says so. Walking 004 = re-pointing
 
 ## Expected
 
-- [ ] `cat /tmp/scen004-findings-count` returns ≥ `4` — at least 4 of the 5 deliberate violations surface. The 5th violation (`fmt.Errorf` flagged by `go-errors/no-fmt-errorf`) does not currently fire because the YAML pattern `fmt.Errorf($$$ARGS)` is parsed by tree-sitter Go grammar as a `type_conversion_expression` rather than a `call_expression`, so the pattern matches no real call sites. Tracked as a separate rule bug; the scenario's contract accepts the current 4-of-5 baseline so it can promote to `active` independent of that fix. When the YAML is corrected, lift this floor to `≥ 5`.
+- [ ] `cat /tmp/scen004-findings-count` returns ≥ `5` — all 5 deliberate violations surface. The fmt.Errorf violation now fires after `rules/go/no-fmt-errorf.yml` was rewritten as a structural rule (`kind: call_expression` + selector match on `fmt.Errorf`) — the original `pattern: fmt.Errorf($$$ARGS)` was parsed as `type_conversion_expression` and matched nothing.
 - [ ] `wc -l < /tmp/scen004-rules` returns ≥ `3` — multiple distinct rule_ids surfaced (negative control: if the fixture parses to zero, the funnel didn't run)
 - [ ] `wc -l < /tmp/scen004-owners` returns ≥ `1` — at least one Owner has findings to adjudicate
 - [ ] Every line in `/tmp/scen004-agent-presence` ends with `AGENT_PRESENT` — Step 4b can resolve every Owner's agent file (regression risk: a renamed agent file silently no-ops that owner's findings; this catches it)