Add SETA stages 3/4/5 baseline candidate

SETA strict functional baseline: 3,126 candidate tasks and 27 archived tasks, reconstructed from the original TaskTrove source.

This variant omits static-check-only package pinning, instruction clarifications, manual verifier/reference quality rewrites and the standard instruction suffix. Execution repairs remain. It restores the 43 tasks previously excluded solely by static checks.

The archive preserves 19 Apptainer-incompatible tasks and 8 historical build/setup/reference/no-op exclusions with their reasons. The README contains the conversion history and stages 3, 4 and 5 only, with a notice that those historical outcomes do not certify the rebuilt payloads.

| Validation | Status |
| --- | --- |
| Patcher tests | 41 passed |
| Content audit | Original instructions preserved for 3,125 tasks; one SSH compatibility wording change |
| Fresh stages 3, 4 and 5 | Ten-task pilot submitted; full validation pending |

This is a candidate PR for review. It is not ready to merge as a validated training dataset. Fresh results and any resulting functional repairs/exclusions will be added here.

camel-ai__SETA-Env_filtered_stages_3_4_5/README.md

@@ -0,0 +1,25 @@
+# SETA filtered by stages 3, 4 and 5
+
+**Candidate awaiting fresh validation.** This version is rebuilt from the original TaskTrove tasks without static-check-only repairs. It contains 3,126 candidate tasks and archives 27 tasks using historical compatibility/build/reference/no-op exclusions. The historical results below describe the earlier repaired release; they do not certify these rebuilt payloads.
+
+The original SETA-Env release contains 4,567 tasks. The conversion in [TaskTrove PR #4](https://huggingface.co/datasets/open-thoughts/TaskTrove/discussions/4) excludes 1,414 tasks whose environments relied on overly specialized Dockerfiles or could not be converted reliably from Docker to Apptainer. All 3,153 retained tasks were audited on an Apptainer cluster: a no-op scored 0, the packaged gold solution scored 1, and every verifier emitted its score through reward.txt.
+
+We then tried to reproduce the results with validation pipeline here: 
+
+## Stage 3: Verifying build
+
+- **19 excluded:** Rocky Linux builds required file ownership our Apptainer builder couldn’t preserve. Fakeroot completed installation but produced incorrect ownership.
+- **5 additional exclusions:** 2 read-only NVIDIA-tool conflicts and 3 very slow R-package setups.
+- **Total: 24 excluded; 3,129 passed stage 3.**
+
+## Stage 4: Reference solution reward 1
+
+- **6 failures fixed:** 2 memory increases, 1 fix to how the verifier matched filenames, 1 fix to how the PHP monitor and verifier sent requests, 1 reference script fixed to start reliably and generate logs faster, and 1 setup-timeout increase. All six then scored 1.
+- **1 task excluded:** installing desktop packages during task setup took too long, so the reference solution could not run successfully.
+- **Total for stage 4: 7 tasks changed: 6 patched and 1 excluded.**
+
+## Stage 5: No-op reward 0
+
+- **32 additional failures fixed:** 31 setup-timeout increases and 1 verifier fix so a missing backup script fails the tests normally instead of stopping them from running. All 32 then scored 0.
+- **2 additional tasks excluded:** R-package installation during task setup too slow.
+- **Total for stage 5: 34 tasks changed: 32 patched and 2 excluded.**

camel-ai__SETA-Env_filtered_stages_3_4_5/archive.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2944ca630b3f68c6223c4393cd5699e1316e0341530739723aefdfe02379a08
+size 216328

camel-ai__SETA-Env_filtered_stages_3_4_5/evidence/content-audit.json

@@ -0,0 +1,29 @@
+{
+  "candidate_tasks": 3126,
+  "original_instructions_preserved": 3125,
+  "compatibility_instruction_edits": [
+    "ask_ubuntu__evolve__1060__d1"
+  ],
+  "manual_quality_payloads_reverted": [
+    "ask_ubuntu__synth__1",
+    "ask_ubuntu__synth__138",
+    "ask_ubuntu__synth__264",
+    "ask_ubuntu__synth__396",
+    "ask_ubuntu__synth__56",
+    "stack_overflow__synth__9282967",
+    "unix_linux_se__synth__41803"
+  ],
+  "pilot_tasks": [
+    "ask_ubuntu__synth__1",
+    "ask_ubuntu__synth__56",
+    "ask_ubuntu__synth__138",
+    "ask_ubuntu__synth__396",
+    "ask_ubuntu__synth__264",
+    "stack_overflow__synth__9282967",
+    "unix_linux_se__synth__41803",
+    "ask_ubuntu__synth__1060",
+    "nl2bash__synth__001",
+    "kaggle_notebook__synth__niharika41298_netflix-visualizations-recommendation-eda"
+  ],
+  "status": "content checks passed; runtime validation pending"
+}

camel-ai__SETA-Env_filtered_stages_3_4_5/evidence/validation-status.json

@@ -0,0 +1,15 @@
+{
+  "status": "candidate_pending_fresh_validation",
+  "candidate_tasks": 3126,
+  "historical_exclusions": 27,
+  "stages": [
+    3,
+    4,
+    5
+  ],
+  "pilot_job_id": "1022891",
+  "patcher_tests": {
+    "passed": 41
+  },
+  "content_audit": "evidence/content-audit.json"
+}

camel-ai__SETA-Env_filtered_stages_3_4_5/tasks.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a55a555fa0d91e6d6bb3f7b8a1425f854a05fdae9949f8c17b3f215d131ef803
+size 732782655