v11 result: 2/40

results/v11/RESULT.md

@@ -0,0 +1,25 @@
+# five-desks-v11 result (third unchanged replicate of v4)
+
+Run `921070b07ab55aa91fc3d16d`, submitted 2026-10-19 18:29 EDT; evaluation done 21:48 EDT.
+
+**Private score: 2/40** (industrial 1, office 1; delta -2 against the 4/40 reference).
+
+| Draw | Run | Steps | Updating steps | Score |
+|---|---|---|---|---|
+| v4 (2026-10-12) | b915134c53bebcc68ce512f9 | 68 | 29 | 5/40 (industrial 2, office 2, software 1) |
+| v9 (2026-10-17) | 8eceb03347c6f626626db1c4 | 82 | 45 | 3/40 (industrial 2, office 1) |
+| v10 (2026-10-18) | ab4d729e0946348182d0ab1c | 76 | ≥31 (51 logged) | 2/40 (industrial 1, office 1) |
+| v11 (2026-10-19) | 921070b07ab55aa91fc3d16d | 83 | pending (Trackio not responding) | 2/40 (industrial 1, office 1) |
+
+## Reading
+
+- Four draws of one request: 5, 3, 2, 2 (mean 3.0), level with the field mean (2.80 over 131 runs, board #294). v4's
+  5/40 was a good draw, not a property of the request.
+- All four draws had at most 50 updating steps where known, the band that scored 3.17 in the field study (#300). These
+  runs entered that study's fitted data (v4, v9, v10) or come after it (v11, which joins the pre-registered recheck set).
+- No new per-domain best; leaderboard unchanged at 0.175, joint second behind ruv (0.225).
+
+## Training
+
+Trackio's metric API timed out for every request between 21:52 and 22:15 EDT, so the per-task table and updating-step
+count will be added when it answers.