CalibForge splits: calibforge_clean 2897, calibforge_repair 1768

README.md

@@ -35,8 +35,8 @@ a fresh sandbox scored 1:
 |---|---|---:|---:|---:|
 | `tmax_clean` | TMAX, judged clean (tier SEED, verdict CLEAN) **and** a replay-verified solution | **5,763** | 5,763 | 499 |
 | `tmax_repair` | every other judged TMAX task | **5,765** | 2,723 | 726 |
-| `calibforge_clean` | CalibForge, judged clean **and** a replay-verified solution | **2,895** | 2,895 | 239 |
-| `calibforge_repair` | every other judged CalibForge task | **1,770** | 1,162 | 159 |
+| `calibforge_clean` | CalibForge, judged clean **and** a replay-verified solution | **2,897** | 2,897 | 239 |
+| `calibforge_repair` | every other judged CalibForge task | **1,768** | 1,162 | 159 |
 
 ```python
 from datasets import load_dataset
@@ -57,9 +57,9 @@ The same three steps and the same rule for both corpora.
    attempted up to twice more at reasoning effort max, stopping at the first attempt that scored 1.
    - TMAX: DeepSeek V4.1 Flash, effort high for the first attempt, max for the retries.
    - CalibForge: **two models.** The first attempt of 544 tasks ran on DeepSeek V4.1 Flash at effort high with
-     top_p 1.0, and of 4,108 tasks on Kimi K3 at effort low with top_p 0.95 (the only top_p its provider accepts).
-     Every retry ran on Kimi K3 at effort max with top_p 0.95. Of the 4,057 replay-verified CalibForge
-     solutions, 418 come from DeepSeek and 3,639 from
+     top_p 1.0, and of 4,110 tasks on Kimi K3 at effort low with top_p 0.95 (the only top_p its provider accepts).
+     Every retry ran on Kimi K3 at effort max with top_p 0.95. Of the 4,059 replay-verified CalibForge
+     solutions, 418 come from DeepSeek and 3,641 from
      Kimi K3; `solution_model` names the one on each row.
 3. **Replayed.** The task's first reward-1 attempt counts only if its recorded action sequence, replayed once in a
    fresh sandbox of the attempt's own size, is graded 1.0 again. That one replay decides; the reward the model
@@ -80,7 +80,7 @@ The same three steps and the same rule for both corpora.
 | SEED/CLEAN, expected zero: one attempt scored 0 and was not retried | 0 | 22 |
 | SEED/CLEAN, an attempt scored 1 but its replay graded 0 | 28 | 29 |
 | SEED/CLEAN, an attempt scored 1 but its replay ended in an error and was not graded | 0 | 6 |
-| SEED/CLEAN, no counted attempt | 1 | 6 |
+| SEED/CLEAN, no counted attempt | 1 | 4 |
 
 A repair row carries a solution when one of its attempts was replay-verified, and none otherwise. Notes on the
 CalibForge column:

data/calibforge_clean.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04bc1668a34278cdb1c5e7d25cd146487cd931649ce926dda5083641ca6bf91f
-size 10414780
+oid sha256:75659d33f864a1a64c7a48533d6fc85d7a8703d70b820c6872525ba041268cf8
+size 10427434

data/calibforge_repair.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63eb8bf14f242258f3f1b316798a0b5f4b530f2fc854e43300fef6fc5436ac52
-size 4919903
+oid sha256:dc3b5542fa54bfaf6cdce3094a38ed9bd1e3eb677ec805c4a8fb76bf9cceb2c0
+size 4917451

data/tasks-calibforge_clean.tar

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:231e5e3b34985370b217ab0d5925c376f8597d4688156cab0bf71ec246426330
-size 1077923840
+oid sha256:bd8480b34e7e46f16981c3cb4676e78c3431c588010e0e45bdb407ad755bcc9d
+size 1081436160

data/tasks-calibforge_repair.tar

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63f1c7b87224beebb8d7d395126116b79386a4903e83796cd970ec5c2d8581f2
-size 652482560
+oid sha256:082d06a67e87c997c5dfdc02cd9ad6ab1a8d5e7da0a948e4c252bd795c301f60
+size 648970240

metadata/split_assignment_calibforge.tsv

@@ -5138,7 +5138,7 @@ system-administration_20260602_105139_009	calibforge_repair	repair_verdict: FLAG
 system-administration_20260603_181444_001	calibforge_repair	repair_unsolved: no rollout scored 1 in 3 counted rollouts	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	3	0.0	0	moonshotai/Kimi-K3	low	0.95	moonshotai/Kimi-K3	max	0.95	moonshotai/Kimi-K3	max	0.95
 system-administration_20260603_181444_002	calibforge_repair	repair_verdict: FLAGGED/INSTR-VERIFIER-MISMATCH; with its replay-verified solution	0	FLAGGED	INSTR-VERIFIER-MISMATCH	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95						
 system-administration_20260603_181444_003	calibforge_clean	tier SEED, verdict CLEAN, and a replay-verified solution	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	2	1.0	1	moonshotai/Kimi-K3	low	0.95	moonshotai/Kimi-K3	max	0.95			
-system-administration_20260603_181444_004	calibforge_repair	repair_not_rolled: no counted rollout (kill budget spent on sandbox/infra failures)	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	0		0									
+system-administration_20260603_181444_004	calibforge_clean	tier SEED, verdict CLEAN, and a replay-verified solution	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95						
 system-administration_20260603_181444_005	calibforge_repair	repair_verdict: FLAGGED/INSTR-VERIFIER-MISMATCH; with its replay-verified solution	0	FLAGGED	INSTR-VERIFIER-MISMATCH	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95						
 system-administration_20260603_181444_006	calibforge_repair	repair_verdict: FLAGGED/INSTR-VERIFIER-MISMATCH; with its replay-verified solution	0	FLAGGED	INSTR-VERIFIER-MISMATCH	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95						
 system-administration_20260603_181444_007	calibforge_repair	repair_verdict: FLAGGED/INSTR-VERIFIER-MISMATCH; with its replay-verified solution	0	FLAGGED	INSTR-VERIFIER-MISMATCH	v13_harbor_row:cfr_h3	remaining	contrastive_solver	2	1.0	1	moonshotai/Kimi-K3	low	0.95	moonshotai/Kimi-K3	max	0.95			
@@ -5205,7 +5205,7 @@ system-administration_20260604_193356_005	calibforge_clean	tier SEED, verdict CL
 system-administration_20260604_193356_006	calibforge_clean	tier SEED, verdict CLEAN, and a replay-verified solution	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	deepseek-flash	high	1.0						
 system-administration_20260604_193356_007	calibforge_clean	tier SEED, verdict CLEAN, and a replay-verified solution	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95						
 system-administration_20260604_193356_008	calibforge_repair	repair_verdict: FLAGGED/INSTR-VERIFIER-MISMATCH; with its replay-verified solution	0	FLAGGED	INSTR-VERIFIER-MISMATCH	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95						
-system-administration_20260604_193356_009	calibforge_repair	repair_not_rolled: no counted rollout (kill budget spent on sandbox/infra failures)	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	0		0									
+system-administration_20260604_193356_009	calibforge_clean	tier SEED, verdict CLEAN, and a replay-verified solution	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95						
 system-administration_20260604_194036_001	calibforge_repair	repair_verdict: FLAGGED/UNSTABLE-REWARD	0	FLAGGED	UNSTABLE-REWARD	v13_harbor_row:cfr_h3	remaining	contrastive_solver	3	0.0	0	moonshotai/Kimi-K3	low	0.95	moonshotai/Kimi-K3	max	0.95	moonshotai/Kimi-K3	max	0.95
 system-administration_20260604_194036_002	calibforge_repair	repair_verdict: FLAGGED/INSTR-VERIFIER-MISMATCH; with its replay-verified solution	0	FLAGGED	INSTR-VERIFIER-MISMATCH	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95						
 system-administration_20260604_194036_003	calibforge_clean	tier SEED, verdict CLEAN, and a replay-verified solution	0	SEED	CLEAN	v13_harbor_row:cfr_h3	remaining	contrastive_solver	1	1.0	1	moonshotai/Kimi-K3	low	0.95