Report both mean and median validation times where task records support them

Derive retained-task ScaleSWE and TermiGen statistics from saved per-task reports; preserve the distinction between validation and image-build time.

TermiGen_environment-reliable/README.md

@@ -39,11 +39,11 @@ Uses [historical checks and reruns](https://huggingface.co/datasets/FWeindel/val
 - **Implementation:** [Marin Harbor `7faf878c14b6`](https://github.com/marin-community/harbor/tree/7faf878c14b6d72737579ec936ebf5f74ba3194d) + [our runtime patches](https://github.com/MLI-lab/OT-next-data/tree/main/harbor_patches).
 - **Resources:** Helma, x86_64; **48 CPUs, 32 GiB/job**, 2 concurrent tasks.
 
-| Check | Time per task | Tasks measured |
-| --- | ---: | ---: |
-| Rule-based checks | Not recorded | n/a |
-| Environment start, inspection and stop | **60.62 s mean** | 3,336 |
-| Reference solution | Skipped: no reference solutions | n/a |
-| No-op baseline | Not recorded | n/a |
-
-Environment timing excludes image builds and verification.
+| Check | Mean per task | Median per task | Tasks measured |
+| --- | ---: | ---: | ---: |
+| Rule-based checks | **0.40 s** | **0.35 s** | 3,336 |
+| Build and setup | **60.62 s** | **71.32 s** | 3,336 |
+| Reference solution | Skipped: no reference solutions | n/a | n/a |
+| No-op baseline | Not recorded | n/a | n/a |
+
+Retained tasks only. Rule-based time sums the recorded checks; build/setup measures environment start, inspection and stop, excluding image builds and verification.

crosscodeeval_environment-reliable/README.md

@@ -39,11 +39,11 @@ Results are from the [historical validation](https://huggingface.co/datasets/FWe
 - **Implementation:** [Marin Harbor `7faf878c14b6`](https://github.com/marin-community/harbor/tree/7faf878c14b6d72737579ec936ebf5f74ba3194d) + [our runtime patches](https://github.com/MLI-lab/OT-next-data/tree/main/harbor_patches).
 - **Resources:** Helma, x86_64; **48 CPUs, 128 GiB/job**, 16 concurrent tasks, no GPUs.
 
-| Check | Time per task | Tasks measured |
-| --- | ---: | ---: |
-| Rule-based checks | **3.36 s mean** | 6,710 |
-| Build and setup | **16.50 s mean** | 6,710 |
-| Reference solution | **14.39 s mean** | 6,710 |
-| No-op baseline | **15.35 s mean** | 6,710 |
-
-Means cover check execution, environment start/inspection/stop, and reference/no-op trials respectively. Image builds are excluded.
+| Check | Mean per task | Median per task | Tasks measured |
+| --- | ---: | ---: | ---: |
+| Rule-based checks | **3.36 s** | **0.54 s** | 6,710 |
+| Build and setup | **16.50 s** | **13.36 s** | 6,710 |
+| Reference solution | **14.39 s** | **11.20 s** | 6,710 |
+| No-op baseline | **15.35 s** | **11.50 s** | 6,710 |
+
+Means and medians cover check execution, environment start/inspection/stop, and reference/no-op trials respectively. Image builds are excluded.

scaleswe_environment-reliable/README.md

@@ -40,11 +40,11 @@ Counts below describe retained repairs; repair groups overlap. Results combine h
 - **Implementation:** [Marin Harbor `7faf878c14b6`](https://github.com/marin-community/harbor/tree/7faf878c14b6d72737579ec936ebf5f74ba3194d) + [our runtime patches](https://github.com/MLI-lab/OT-next-data/tree/main/harbor_patches).
 - **Resources:** Helma, x86_64; cached image builds: **4 CPUs, 8 GiB**.
 
-| Check | Time per task | Tasks measured |
-| --- | ---: | ---: |
-| Rule-based checks | Not recorded | n/a |
-| Build and setup | Not recorded | n/a |
-| Reference solution | **20.50 s mean** | 17,202 |
-| No-op baseline | **19.77 s mean** | 17,202 |
-
-Latest recorded trial per original task, including subsequently archived tasks.
+| Check | Mean per task | Median per task | Tasks measured |
+| --- | ---: | ---: | ---: |
+| Rule-based checks | **0.30 s** | **0.29 s** | 16,786 |
+| Build and setup | **10.04 s** | **9.38 s** | 16,786 |
+| Reference solution | **16.53 s** | **10.70 s** | 16,786 |
+| No-op baseline | **15.71 s** | **9.90 s** | 16,786 |
+
+Latest recorded checks for the **16,786 retained tasks**. Build/setup measures environment start, inspection and stop; image builds are excluded. Reference/no-op timings cover the recorded trials.