diff --git a/evals/agent-blind-spots/runs/2026-05-29/spike-driver/summary.md b/evals/agent-blind-spots/runs/2026-05-29/spike-driver/summary.md
new file mode 100644
index 0000000..9c2bc9e
--- /dev/null
+++ b/evals/agent-blind-spots/runs/2026-05-29/spike-driver/summary.md
@@ -0,0 +1,24 @@
+# Spike driver run — summary
+
+- Run date: 2026-05-29
+- Cells attempted: 4
+- Cells with transcript: 4
+- Cells judged: 4
+- Cells errored: 0
+
+## Cell matrix
+
+| Fixture | Agent | Tier | Catch | Evidence | Delta | Status |
+|---|---|---|---|---|---|---|
+| pr1-fix-clv | claude | 0 | catch | 0 | same | ok |
+| pr1-fix-clv | claude | 1 | catch | 1c | improvement | ok |
+| pr1-fix-clv | codex | 0 | miss | 0 | same | ok |
+| pr1-fix-clv | codex | 1 | miss | 0 | same | ok |
+
+## Judge self-consistency (two passes on the same transcript)
+
+- **catch**: 100% (4/4 double-judged cells)
+- **tier**: 100% (4/4 double-judged cells)
+- **delta**: 75% (3/4 double-judged cells)
+
+**Stability bar:** ≥80% per axis. Below that, the judge can't replace human grading at N=6.
diff --git a/evals/agent-blind-spots/runs/2026-05-29/spike-driver/verdicts.csv b/evals/agent-blind-spots/runs/2026-05-29/spike-driver/verdicts.csv
new file mode 100644
index 0000000..d1187b1
--- /dev/null
+++ b/evals/agent-blind-spots/runs/2026-05-29/spike-driver/verdicts.csv
@@ -0,0 +1,5 @@
+fixture,agent,tier,catch,evidence_tier,delta,catch_2,tier_2,delta_2,returncode,error
+pr1-fix-clv,claude,0,catch,0,same,catch,0,improvement,0,
+pr1-fix-clv,claude,1,catch,1c,improvement,catch,1c,improvement,0,
+pr1-fix-clv,codex,0,miss,0,same,miss,0,same,2,
+pr1-fix-clv,codex,1,miss,0,same,miss,0,same,2,