diff --git a/api/internal/querylang/parser/parser.go b/api/internal/querylang/parser/parser.go
index 01201e7..d1daf4e 100644
--- a/api/internal/querylang/parser/parser.go
+++ b/api/internal/querylang/parser/parser.go
@@ -41,11 +41,36 @@ func (p *parser) advance() {
 }
 
 func (p *parser) parseQuery() (*ast.Query, error) {
-	base, err := p.parseBoolExpr()
-	if err != nil {
-		return nil, err
+	q := &ast.Query{}
+
+	// base_search is normally mandatory (see the grammar comment on
+	// ast.Query), but every pipe-stage keyword is also a valid bare
+	// Ident, so a query with no leading filter -- e.g. `stats count by
+	// host`, documented as valid in
+	// /docs/query-language-reference.md's "stats" section -- used to
+	// get silently misparsed as a base_search of four ANDed free-text
+	// terms ("stats", "count", "by", "host"), matching nothing rather
+	// than erroring or aggregating. Recognizing a leading pipe-stage
+	// keyword up front and skipping straight to stage-parsing (q.Base
+	// stays its zero value; compileBoolExpr already treats zero Terms
+	// as "match everything", so no planner/executor change is needed)
+	// fixes that without a leading "|". atPipeStageKeyword's comparator
+	// lookahead keeps a genuine field named e.g. "where" (`where=foo`)
+	// parsing as a structured filter, same as today.
+	if p.atPipeStageKeyword() {
+		stage, err := p.parsePipeStage()
+		if err != nil {
+			return nil, err
+		}
+		q.Pipes = append(q.Pipes, stage)
+	} else {
+		base, err := p.parseBoolExpr()
+		if err != nil {
+			return nil, err
+		}
+		q.Base = base
 	}
-	q := &ast.Query{Base: base}
+
 	for p.cur.Kind == lexer.Pipe {
 		p.advance()
 		stage, err := p.parsePipeStage()
@@ -60,6 +85,24 @@ func (p *parser) parseQuery() (*ast.Query, error) {
 	return q, nil
 }
 
+// atPipeStageKeyword reports whether the parser is sitting on one of the
+// six pipe-stage keywords used as the query's very first token, with no
+// preceding base filter. The comparator lookahead disambiguates from a
+// structured comparison on a field that happens to share one of these
+// names (e.g. `where=foo`), which must keep parsing as a filter, not a
+// stage.
+func (p *parser) atPipeStageKeyword() bool {
+	if p.cur.Kind != lexer.Ident || isComparatorStart(p.next.Kind) {
+		return false
+	}
+	switch p.cur.Value {
+	case "where", "stats", "sort", "fields", "head", "tail":
+		return true
+	default:
+		return false
+	}
+}
+
 func (p *parser) parseBoolExpr() (ast.BoolExpr, error) {
 	var expr ast.BoolExpr
 	term, err := p.parseTerm()
diff --git a/api/internal/querylang/parser/parser_test.go b/api/internal/querylang/parser/parser_test.go
index 4f78ddd..f92cbab 100644
--- a/api/internal/querylang/parser/parser_test.go
+++ b/api/internal/querylang/parser/parser_test.go
@@ -266,6 +266,77 @@ func TestParseMultipleAggregations(t *testing.T) {
 	}
 }
 
+func TestParseLeadingStatsStageWithNoBaseFilter(t *testing.T) {
+	// Documented as valid in /docs/query-language-reference.md's "stats"
+	// section -- used to get silently misparsed as four ANDed free-text
+	// base terms ("stats", "count", "by", "host"), matching nothing.
+	q, err := Parse(`stats count by host`)
+	if err != nil {
+		t.Fatalf("Parse() error = %v", err)
+	}
+	if len(q.Base.Terms) != 0 {
+		t.Fatalf("expected an empty base (match-everything), got %+v", q.Base.Terms)
+	}
+	if len(q.Pipes) != 1 {
+		t.Fatalf("expected 1 pipe stage, got %d: %+v", len(q.Pipes), q.Pipes)
+	}
+	stats, ok := q.Pipes[0].(ast.StatsStage)
+	if !ok {
+		t.Fatalf("expected StatsStage, got %T", q.Pipes[0])
+	}
+	if len(stats.Aggs) != 1 || stats.Aggs[0].Func != "count" || len(stats.By) != 1 || stats.By[0] != "host" {
+		t.Fatalf("unexpected stats stage: %+v", stats)
+	}
+}
+
+func TestParseLeadingStatsStageWithParenAggAndNoBaseFilter(t *testing.T) {
+	// The specific case that first surfaced this: count() (parens, zero
+	// args) as the very first token of the query.
+	q, err := Parse(`stats count(), avg(latency_ms) as avg_latency by host, service`)
+	if err != nil {
+		t.Fatalf("Parse() error = %v", err)
+	}
+	stats := q.Pipes[0].(ast.StatsStage)
+	if len(stats.Aggs) != 2 || stats.Aggs[0].Func != "count" || stats.Aggs[1].Func != "avg" {
+		t.Fatalf("unexpected stats aggs: %+v", stats.Aggs)
+	}
+}
+
+func TestParseLeadingWhereStageWithNoBaseFilter(t *testing.T) {
+	q, err := Parse(`where status>=500`)
+	if err != nil {
+		t.Fatalf("Parse() error = %v", err)
+	}
+	if len(q.Base.Terms) != 0 {
+		t.Fatalf("expected an empty base, got %+v", q.Base.Terms)
+	}
+	where, ok := q.Pipes[0].(ast.WhereStage)
+	if !ok {
+		t.Fatalf("expected WhereStage, got %T", q.Pipes[0])
+	}
+	cmp := where.Expr.Terms[0].(ast.Comparison)
+	if cmp.Field != "status" || cmp.Op != ">=" || cmp.Value != "500" {
+		t.Fatalf("unexpected where comparison: %+v", cmp)
+	}
+}
+
+func TestParseFieldNamedLikeAPipeStageKeywordStillFilters(t *testing.T) {
+	// atPipeStageKeyword's comparator lookahead must not misfire on a
+	// genuine structured filter for a field that happens to share one
+	// of the six reserved names.
+	q, err := Parse(`where=foo`)
+	if err != nil {
+		t.Fatalf("Parse() error = %v", err)
+	}
+	if len(q.Pipes) != 0 {
+		t.Fatalf("expected no pipe stages, got %+v", q.Pipes)
+	}
+	cmp, ok := q.Base.Terms[0].(ast.Comparison)
+	if !ok || cmp.Field != "where" || cmp.Value != "foo" {
+		t.Fatalf("expected Comparison(where=foo), got %+v", q.Base.Terms[0])
+	}
+}
+
 // --- error cases ---
 
 func TestParseErrorEmptyQuery(t *testing.T) {