[codex] add post-report benchmark artifacts

evaluations/conversation_logs/README.md

@@ -0,0 +1,8 @@
+# Conversation Logs
+
+This directory preserves the Codex conversation around the data harmonization benchmark work so reviewers can inspect the user's instructions and Codex responses.
+
+- `rollout-2026-06-30T13-49-41-019f1a4b-79ea-7070-bee5-609ceccdba08.jsonl`: complete raw Codex session JSONL copied from `/h/jmc/.codex/sessions/2026/06/30/` at artifact creation time. It includes tool calls and tool outputs as recorded by Codex.
+- `conversation_019f1a4b-79ea-7070-bee5-609ceccdba08.md`: readable Markdown extraction of user and assistant messages from the raw JSONL.
+
+The raw transcript was scanned for obvious credential patterns before check-in. Token-shaped values were redacted in the committed copy. It includes Codex system/developer context because that is part of the raw session record.

evaluations/post_report_results/README.md

@@ -0,0 +1,19 @@
+# Post-Report Completed Results
+
+Created: 2026-08-11T17:43:41+00:00
+
+This directory captures benchmark-related outputs that were found after the overall evaluation report branch had been checked in.
+
+## Contents
+
+- `parallel_batch_20260701/`: local `.runs/fold-*` result artifacts for the parallel leave-one-out batch described in GitHub issue #42. These include manifests, instructions, action logs, generated harmonizer code, mapping JSON, curator bundles, and notes.
+- `parallel_batch_20260701/large_generated_csv_manifest.csv`: checksums, sizes, and line counts for generated harmonized CSVs that were too large to commit directly to GitHub.
+- `github_actions/run_28483156122_eval_holdout_7/`: downloaded artifact from the successful `Run Harmonization Eval` GitHub Actions run for holdout 7.
+
+## Trace Caveat
+
+The `.runs` folders do not contain complete raw sub-agent tool-call transcripts. They contain `AGENT_ACTION_LOG.md` files and final outputs. Future benchmark harness runs should persist the complete raw sub-agent JSONL/tool-call transcript for each sub-agent, not only action logs and output files.
+
+## Large CSV Policy
+
+Generated harmonized CSVs were not committed because several are too large for normal GitHub storage, including files over 100 MB. They are represented by size, line count, and SHA-256 digest in `parallel_batch_20260701/large_generated_csv_manifest.csv`.

evaluations/post_report_results/github_actions/run_28483156122_eval_holdout_7/artifact/eval-holdout-7/AGENT_INSTRUCTIONS.md

@@ -0,0 +1,10 @@
+# Run environment — leave-one-cluster-out
+
+Harmonize the held-out dataset(s) below using ONLY:
+- the skills in `skills/`,
+- the exemplars in `data/processed/.../sm_data_harmonization_mapping.json` and the code patterns in `data/gold/expert_code/harmonize_sm/` (both have the held-out cluster removed),
+- the shared raw inputs under `~/ess-dive_wfsfa_soil_datasets/` and the cached metadata under `data/external/ess-dive_meta/`.
+
+Do NOT look up the held-out dataset's existing harmonized output, expert code, or mapping entry from any other location. The held-out datasets are:
+
+- `ess-dive-38e901ec3d7bd24-20230504T211548257225`

evaluations/post_report_results/github_actions/run_28483156122_eval_holdout_7/artifact/eval-holdout-7/MANIFEST.json

@@ -0,0 +1,35 @@
+{
+  "name": "holdout-7",
+  "holdout_indices": [
+    7
+  ],
+  "holdout_identifiers": [
+    "ess-dive-38e901ec3d7bd24-20230504T211548257225"
+  ],
+  "exemplar_indices": [
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    8,
+    9,
+    10,
+    15,
+    16,
+    17,
+    18,
+    23,
+    24,
+    25,
+    26,
+    27
+  ],
+  "n_exemplars": 27,
+  "sources": {
+    "package": "data/gold/expert_code/harmonize_sm",
+    "mapping": "data/gold/sm_data_harmonization_mapping.json",
+    "skills": "skills"
+  }
+}

evaluations/post_report_results/github_actions/run_28483156122_eval_holdout_7/artifact/eval-holdout-7/_inspect.py

@@ -0,0 +1,19 @@
+import pandas as pd, numpy as np
+D="/home/runner/ess-dive_wfsfa_soil_datasets/ess-dive-38e901ec3d7bd24-20230504T211548257225/"
+df=pd.read_csv(D+"BM_Merged_T_VWC_0616_1018.csv")
+print("VWC file shape",df.shape)
+print("cols",df.columns.tolist())
+print("depths(cm)",sorted(df['Depth (cm)'].dropna().unique()))
+print("first/last",df['date.time'].iloc[0],"/",df['date.time'].iloc[-1])
+print("vwc min/max",df['Volumetric Water Content'].min(),df['Volumetric Water Content'].max())
+print(df.groupby('Depth (cm)').size())
+# interval check at one depth
+g=df[df['Depth (cm)']==10].copy()
+g['dt']=pd.to_datetime(g['date.time'],format="%m/%d/%y %H:%M",errors='coerce')
+print("median interval min (depth10)", g['dt'].diff().dt.total_seconds().median()/60)
+print("nat count", g['dt'].isna().sum())
+print("\n--- CO2/loc file ---")
+co=pd.read_csv(D+"BM_EGM_Well_CO2.csv")
+print("shape",co.shape,"cols",co.columns.tolist())
+print("Locations",co['Location'].unique())
+print("lat/lon",co.iloc[0,1],co.iloc[0,2])