C1: close cross-step wiring gap between Locate and Post journey-delta steps

Validation's cycle-3 adversarial audit found that reverting ONLY the "Post
the journey-cost delta PR comment" step's --metrics-dir back to the old
broken hardcoded path left the full suite green — nothing verified the Post
step stayed wired to the "Locate this run's journey metrics" step's actual
output directory. Same class of gap Cycle 1 found: a step's own mechanism
tested in isolation, the wiring between two steps not.

Adds TestJourneyDeltaLocateAndPostStepsShareMetricsDir, which extracts both
steps' real run: blocks and derives each step's directory value from the
actual script text (not a hardcoded assumption), then asserts they match.
Verified it catches the exact adversarial mutation validation performed.

Also folds in the two minors from the same review: fixes a stale "backfill
CLI" doc-comment reference in shallow_boot_window_record.go left over from
the AC-4 TRIM (should read "runbook"), and adds
TestRenderJourneyDeltaCommentShowsTokenClassBreakdown, which checks the PR
comment's Cache Read Δ / Cache Creation Δ table cells land in the correct
(non-swapped) columns — verified it catches a reversion to the old
Turns/Tokens-total/Cost table shape.

Author: Spike Test

internal/ensigncycle/shallow_boot_window_record.go

@@ -1,5 +1,5 @@
 // ABOUTME: Boot-window telemetry extraction shared by the live shallow-boot
-// ABOUTME: scenario test and the release-ledger backfill CLI (AC-1/AC-4).
+// ABOUTME: scenario test and the release-ledger backfill runbook (AC-1/AC-4).
 package ensigncycle
 
 import (
@@ -95,8 +95,8 @@ func preGreetPeakCacheCreation(turns []journeymetrics.ClaudeTurn, greet int) int
 // silent alias-resolution change (e.g. "sonnet" moving to a new snapshot)
 // rather than the FO's own contract.
 // Exported (unlike the live-scenario test glue around it) so the AC-4 release-
-// ledger backfill CLI can apply the SAME extraction logic to archived streams,
-// standalone, rather than duplicating it.
+// ledger backfill runbook's throwaway `go run` script can apply the SAME
+// extraction logic to archived streams, standalone, rather than duplicating it.
 func BuildShallowBootWindowRecord(turns []journeymetrics.ClaudeTurn, model string, claudeCodeVersion string, resolvedModel string) (journeymetrics.Record, error) {
 	if len(turns) == 0 {
 		return journeymetrics.Record{}, fmt.Errorf("stream carried no assistant turns — nothing to record")

internal/release/journey_workflow_test.go

@@ -782,6 +782,45 @@ func TestJourneyDeltaLocateMetricsStepFindsNestedJourneyMetricsFiles(t *testing.
 	}
 }
 
+// TestJourneyDeltaLocateAndPostStepsShareMetricsDir closes the cross-step
+// wiring gap validation's cycle-3 adversarial audit found: the "Locate this
+// run's journey metrics" step's own mechanism is tested in isolation above,
+// but nothing previously verified that the "Post the journey-cost delta PR
+// comment" step's --metrics-dir argument stays wired to the Locate step's
+// actual output directory. A one-line regression reverting ONLY the Post
+// step's --metrics-dir back to the old hardcoded broken path left the full
+// suite green, because each step's script was only ever checked in isolation.
+// This derives BOTH directories from the REAL extracted step scripts (not a
+// hardcoded assumption on either side) and asserts they're the same value.
+func TestJourneyDeltaLocateAndPostStepsShareMetricsDir(t *testing.T) {
+	live := readWorkflow(t, "runtime-live-e2e.yml")
+	locateScript := extractStepRun(t, live, "Locate this run's journey metrics")
+	postScript := extractStepRun(t, live, "Post the journey-cost delta PR comment")
+
+	locateDir := firstQuotedArg(t, locateScript, `mkdir -p "`)
+	postDir := firstQuotedArg(t, postScript, `--metrics-dir "`)
+	if locateDir != postDir {
+		t.Fatalf("Locate step writes to %q but Post step's --metrics-dir reads from %q — the two steps are no longer wired together", locateDir, postDir)
+	}
+}
+
+// firstQuotedArg extracts the double-quoted value immediately following the
+// given prefix (e.g. `mkdir -p "`) in script, failing the test if the prefix
+// or its closing quote is not found.
+func firstQuotedArg(t *testing.T, script, prefix string) string {
+	t.Helper()
+	start := strings.Index(script, prefix)
+	if start < 0 {
+		t.Fatalf("script does not contain %q:\n%s", prefix, script)
+	}
+	start += len(prefix)
+	end := strings.Index(script[start:], `"`)
+	if end < 0 {
+		t.Fatalf("unterminated quoted argument after %q:\n%s", prefix, script)
+	}
+	return script[start : start+end]
+}
+
 // extractStepRun pulls a named step's `run: |` block out of a workflow document,
 // dedented to a runnable shell script, so tests exercise the EXACT script CI
 // runs rather than a hand-copied duplicate.

internal/release/journeydelta_test.go

@@ -170,6 +170,54 @@ func TestRenderJourneyDeltaCommentIncludesExactDeltasAndMarker(t *testing.T) {
 	}
 }
 
+// TestRenderJourneyDeltaCommentShowsTokenClassBreakdown proves the comment
+// renders Cache Read Δ and Cache Creation Δ as their OWN columns, in the
+// correct (non-swapped) position — not collapsed into the Tokens Δ (total)
+// figure. cache_read and cache_creation differ ~12x in cost and meaning for a
+// boot metric, so folding them into one number hides the signal this Minor
+// exists to surface. Uses distinguishable, easily-confused-if-swapped values
+// (CacheRead=111, CacheCreation=222) and checks the actual table row's cell
+// positions, not just substring presence anywhere in the body (which would
+// not catch a column swap).
+func TestRenderJourneyDeltaCommentShowsTokenClassBreakdown(t *testing.T) {
+	deltas := []JourneyDelta{
+		{
+			ScenarioID: "shallow-boot-window", Runtime: "claude", Model: "claude-sonnet-4-6",
+			HasBaseline: true, BaselineRunURL: "https://github.com/spacedock-dev/spacedock/actions/runs/27931963802",
+			TurnsDelta: 2, TokensDelta: journeymetrics.TokenTotals{CacheRead: 111, CacheCreation: 222, Total: 1034}, CostDeltaUSD: 0.15,
+		},
+	}
+	body := RenderJourneyDeltaComment(deltas)
+
+	var row string
+	for _, line := range strings.Split(body, "\n") {
+		if strings.HasPrefix(line, "| shallow-boot-window") {
+			row = line
+			break
+		}
+	}
+	if row == "" {
+		t.Fatalf("comment body missing the data row:\n%s", body)
+	}
+	cells := strings.Split(row, "|")
+	// | (empty) | Scenario | Runtime | Model | Turns Δ | Cache Read Δ | Cache Creation Δ | Tokens Δ (total) | Cost Δ (USD) | Baseline | (empty) |
+	if len(cells) != 11 {
+		t.Fatalf("data row has %d cells, want 11 (header shape changed?): %q", len(cells), row)
+	}
+	cacheReadCell := strings.TrimSpace(cells[5])
+	cacheCreationCell := strings.TrimSpace(cells[6])
+	tokensTotalCell := strings.TrimSpace(cells[7])
+	if cacheReadCell != "+111" {
+		t.Fatalf("Cache Read Δ cell = %q, want +111 (got %q for Cache Creation Δ) — columns may be swapped", cacheReadCell, cacheCreationCell)
+	}
+	if cacheCreationCell != "+222" {
+		t.Fatalf("Cache Creation Δ cell = %q, want +222 (got %q for Cache Read Δ) — columns may be swapped", cacheCreationCell, cacheReadCell)
+	}
+	if tokensTotalCell != "+1034" {
+		t.Fatalf("Tokens Δ (total) cell = %q, want +1034", tokensTotalCell)
+	}
+}
+
 // TestRenderJourneyDeltaCommentRendersNoBaselineAsNewNotSelfDelta proves a
 // scenario/model with no matching baseline observation renders "n/a (new)" in
 // its delta cells rather than a self-delta against an implicit zero baseline