CrossCodeEval: complete pass@16 results and combined trace statistics

crosscodeeval-csharp-v5/pass_counts.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f0c566d17837c7a9628bbe7f9097a32ef46ef24330db4993bd84edaf3823ee7
+size 213144

crosscodeeval-java-v4/pass_counts.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee9e0a1a183762facbd2604098c14c606e40afa16b538e6c58588b323e26a0a1
+size 310701

crosscodeeval-python-v3/pass_counts.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e6d1a167d1c8e73957f34e2eb1ae22b803c5641fd8c02937721dbd8a914d297
+size 75110

crosscodeeval-typescript-v3/pass_counts.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e33626110de802f7ed9a683805eeb0492ed3d6ce480675934ee37622d67a4a28
+size 480717

runs/2026-10-07-crosscodeeval-pass16.json

@@ -0,0 +1,534 @@
+{
+  "run": "2026-10-07-crosscodeeval-pass16",
+  "kind": "pass_counts_and_trace_metrics",
+  "summary": {
+    "qwen3-30b-instruct-2507": {
+      "tasks": 6710,
+      "attempts": 107360,
+      "pass@1": 0.027971311475409835,
+      "pass@2": 0.04189269746646796,
+      "pass@4": 0.05787564894122272,
+      "pass@8": 0.07538018034292251,
+      "pass@16": 0.09552906110283159,
+      "trace_metrics": {
+        "trajectories": 107360,
+        "graded": 107360,
+        "solved": 3003,
+        "turns": {
+          "n": 107359,
+          "mean": 2.6162874095325033,
+          "median": 2,
+          "p90": 2,
+          "p95": 4,
+          "max": 433
+        },
+        "input_tokens": {
+          "n": 107359,
+          "mean": 11270.863020333647,
+          "median": 3865,
+          "p90": 7488,
+          "p95": 9877,
+          "max": 6431947
+        },
+        "input_tokens_total": 1210028583,
+        "output_tokens": {
+          "n": 107359,
+          "mean": 786.0006427034529,
+          "median": 393,
+          "p90": 616,
+          "p95": 890,
+          "max": 134532
+        },
+        "output_tokens_total": 84384243,
+        "peak_context_tokens": {
+          "n": 107338,
+          "mean": 2840.7634481730606,
+          "median": 2247,
+          "p90": 3575,
+          "p95": 4483,
+          "max": 72569
+        },
+        "peak_context_fraction": {
+          "n": 107338,
+          "mean": 0.08669322046426577,
+          "median": 0.068572998046875,
+          "p90": 0.109100341796875,
+          "p95": 0.136810302734375,
+          "max": 2.214630126953125
+        },
+        "trajectories_above_90_percent_context": 713,
+        "latency_seconds": {
+          "model_call": {
+            "n": 284844,
+            "mean": 2.6475805339975143,
+            "max": 149.47315096855164,
+            "median": null,
+            "p90": null,
+            "p95": null,
+            "note": "Per-call samples were not retained; pooled quantiles cannot be reconstructed."
+          },
+          "tool_call": null,
+          "agent": {
+            "n": 107360,
+            "mean": 11.691557514307005,
+            "median": 6.294525,
+            "p90": 9.719629,
+            "p95": 13.298929,
+            "max": 2464.816905
+          },
+          "verifier": {
+            "n": 107360,
+            "mean": 1.288768823733234,
+            "median": 1.236909,
+            "p90": 1.375125,
+            "p95": 1.411272,
+            "max": 117.897129
+          },
+          "trial": {
+            "n": 107360,
+            "mean": 28.064082660068927,
+            "median": 13.718699,
+            "p90": 34.028572,
+            "p95": 89.080016,
+            "max": 2501.000071
+          }
+        },
+        "tools": {
+          "bash_command": {
+            "calls": 167067,
+            "calls_per_trajectory": 1.5561382265275707,
+            "failed": null,
+            "failure_rate": null
+          },
+          "mark_task_complete": {
+            "calls": 215763,
+            "calls_per_trajectory": 2.0097149776453054,
+            "failed": null,
+            "failure_rate": null
+          }
+        },
+        "tool_steps": 252089,
+        "tool_steps_without_output": 0,
+        "tool_steps_with_error_text": 7146,
+        "tool_steps_with_error_text_rate": 0.028347131370270024,
+        "terminations": {
+          "session_ended": 4,
+          "task_complete": 107272,
+          "task_timeout": 84
+        },
+        "command_distribution": {
+          "                        DELETE": 7.291978094897803e-07,
+          "                    matcher.group(\"artifactId\"),\n": 4.3088961469850656e-07,
+          "                    matcher.group(\"classifier\"),\n": 4.3088961469850656e-07,
+          "                    matcher.group(\"groupId\"),\n": 4.3088961469850656e-07,
+          "                    matcher.group(\"type\")),\n": 4.3088961469850656e-07,
+          "                    matcher.group(\"version\"),\n": 4.3088961469850656e-07,
+          "                .collect(Collectors.joining(\"\\n\"));": 1.5799285872278573e-06,
+          "                .collect(Collectors.joining(\"\\n\"));\n": 1.5799285872278573e-06,
+          "                .map(entry -> \"\\u2022 \" + entry.getDisplayValue() + \" (\" + entry.getRank() + \")\")\n": 3.1598571744557146e-06,
+          "                Collectors.toCollection(LinkedHashSet::new)))));\n": 4.3088961469850656e-07,
+          "                matcher -> new Dependency(\n": 4.3088961469850656e-07,
+          "                throw new UnsupportedOperationException(Unsupported": 7.291978094897803e-07,
+          "            Collectors.mapping(\n": 4.3088961469850656e-07,
+          "            ex.printStackTrace();\n": 7.291978094897803e-07,
+          "            matcher -> matcher.group(\"module\"),\n": 4.3088961469850656e-07,
+          "            preparedStatement.close();\n": 7.291978094897803e-07,
+          "            preparedStatement.executeUpdate();\n": 7.291978094897803e-07,
+          "            }\n": 7.291978094897803e-07,
+          "            } else {\n": 7.291978094897803e-07,
+          "        .collect(Collectors.groupingBy(\n": 4.3088961469850656e-07,
+          "        .filter(Matcher::matches)\n": 4.3088961469850656e-07,
+          "        .map(line -> DEPENDENCY_COPY_ITEM_PATTERN.matcher(line))\n": 4.3088961469850656e-07,
+          "        }\n": 7.291978094897803e-07,
+          "        } catch (SQLException ex) {\n": 7.291978094897803e-07,
+          "    if (newVotes.length === 0) break;\n": 1.0532857248185714e-06,
+          "    page++;\n": 1.0532857248185714e-06,
+          "    resultsSize += newVotes.length;\n": 1.0532857248185714e-06,
+          "    votes = votes.concat(newVotes);\n": 1.0532857248185714e-06,
+          "    }\n": 7.291978094897803e-07,
+          "  return votes;\n": 1.0532857248185714e-06,
+          "  } while (page < maxPage);\n": 1.0532857248185714e-06,
+          " && !key) {\n                        key = true;\n                        if (currentKey != null && currentValue != null) {\n                            cookieStore.add(currentKey.toString(), currentValue.toString());\n                        }\n                        currentKey = null;\n                        currentValue = null;\n                    } else if (key) {\n                        if (currentKey != null) currentKey.append(character);\n                        else currentKey = new StringBuilder().append(character);\n                    } else {\n                        if (currentValue != null) currentValue.append(character);\n                        else currentValue = new StringBuilder().append(character);\n                    }\n                }\n                if (currentKey != null && currentValue != null) {\n                    cookieStore.add(currentKey.toString(), currentValue.toString());\n                }\n            });\n        }\n    });": 1.5799285872278573e-06,
+          " + ag.getEndereco() + n;": 2.369892880841786e-06,
+          " + ag.getNome() + ": 2.369892880841786e-06,
+          " + conta.getAgencia().getNumAgencia() + ": 1.5799285872278573e-06,
+          " + conta.getNumConta() + ": 2.369892880841786e-06,
+          " + conta.getSaldo() + ": 1.5799285872278573e-06,
+          " + conta.getTipoDeConta() + ": 1.5799285872278573e-06,
+          " + e.getStackTrace())": 1.5799285872278573e-06,
+          " + response.getBody().length() + rn, StandardCharsets.UTF_8);\nwrite(rn, StandardCharsets.UTF_8);\nwrite(response.getBody(), StandardCharsets.UTF_8);\nflush();\nclose();": 1.3542245033381632e-06,
+          " + url.toString();\n    }\n}\n": 3.1598571744557146e-06,
+          " 4;\n    }\n\n    public Response respond(String userInput) {\n        List<Map<String, String>> fullPrompt = getFullPrompt(userInput);\n        return openAIModel.generate(fullPrompt);\n    }\n\n    public void addGoal(String goal) {\n        goals.add(goal);\n    }\n\n    public String getName() {\n        return name;\n    }\n\n    public String getDescription() {\n        return description;\n    }\n\n    public List<String> getGoals() {\n        return goals;\n    }\n}\n": 3.1598571744557146e-06,
+          ") : item;\n    if (seen.has(key)) {\n      field.report(messages.array.distinct, \"array.distinct\", field, options);\n      return;\n    }\n    seen.add(key);\n  }\n": 3.1598571744557146e-06,
+          "))\n        })\n    }\n": 4.739785761683572e-06,
+          ")) }\n      : { success: true, value: result }\n  }": 4.739785761683572e-06,
+          ");": 4.989248170193233e-07,
+          ")}\n            headers.append(\"Set-Cookie\", cookieHeader)\n        })\n    }": 2.369892880841786e-06,
+          ")}\n            headers.set(\"Set-Cookie\", cookieHeader)\n        })\n    }\n": 2.369892880841786e-06,
+          ")}\n            headers.set(Set-Cookie, cookieHeader)\n        })\n    }\n": 2.369892880841786e-06,
+          ")}\n  }": 4.739785761683572e-06,
+          ")}  }": 3.1598571744557146e-06,
+          ")}  };\n": 1.8959143046734288e-06,
+          ")}  };\n}\n": 3.1598571744557146e-06,
+          "7 + event.getMessage()));": 2.7295185694946742e-06,
+          "8\u2192": 2.7295185694946742e-06,
+          ":wq": 1.3542245033381632e-06,
+          "=": 5.266428624092857e-07,
+          "C-o": 1.5799285872278573e-06,
+          "C-x": 3.4758428919012863e-06,
+          "CPF:": 6.504039350754679e-06,
+          "Conta:": 3.3441821762989643e-06,
+          "Dias": 1.5799285872278573e-06,
+          "EOF": 1.6771549618264947e-06,
+          "EOL": 2.1544480734925328e-07,
+          "ESC": 1.3542245033381632e-06,
+          "Exception": 3.1598571744557146e-06,
+          "Merge": 3.1598571744557146e-06,
+          "Saldo:": 2.369892880841786e-06,
+          "Skip": 8.689607229753214e-06,
+          "String": 4.533467972257381e-06,
+          "Taxa": 1.5799285872278573e-06,
+          "Valor": 3.1598571744557146e-06,
+          "[assignment]": 1.5143085500586491e-06,
+          "[no_commands]": 0.0025974025974025974,
+          "[unknown]": 0.004908990875697409,
+          "a + player.getName() + ": 2.7295185694946742e-06,
+          "adb": 4.739785761683572e-06,
+          "awk": 4.607435722962914e-06,
+          "base64": 1.692780629172704e-07,
+          "bash": 1.7406895149040074e-07,
+          "break": 4.560107608626286e-06,
+          "cPlayer": 1.1939746037764807e-05,
+          "cat": 0.046385863754286834,
+          "cd": 0.007175432611753033,
+          "chmod": 1.4712562505402267e-06,
+          "cp": 2.038648462480105e-05,
+          "curl": 5.3656911531333515e-06,
+          "currentClass": 3.1598571744557145e-07,
+          "dd": 3.0609478551048396e-06,
+          "deleted": 2.369892880841786e-06,
+          "double": 4.551625368764433e-05,
+          "e + distance + m": 1.3647592847473371e-06,
+          "echo": 0.21406217099956362,
+          "ed": 2.1724018074383036e-06,
+          "exit": 1.0608091942815612e-05,
+          "fetch": 5.266428624092857e-07,
+          "find": 0.0017318503370308245,
+          "grep": 0.0004912695394780914,
+          "h": 1.0532857248185714e-06,
+          "hasDuplicateHue": 1.0532857248185714e-06,
+          "head": 0.00014870228432184597,
+          "hexdump": 8.942992003176551e-08,
+          "i": 1.3542245033381632e-06,
+          "int": 5.721004568488241e-06,
+          "java": 1.5799285872278573e-06,
+          "linha": 2.085149069522744e-05,
+          "ls": 0.01619231052112797,
+          "mkdir": 0.0002947435179175777,
+          "mv": 8.953492679604566e-05,
+          "n": 9.479571523367144e-06,
+          "nano": 1.468736796498762e-05,
+          "newQuery": 4.739785761683572e-06,
+          "nodo1": 2.369892880841786e-06,
+          "nodo2": 2.369892880841786e-06,
+          "nodo3": 2.369892880841786e-06,
+          "owner": 7.425664359970929e-06,
+          "printf": 0.0012112613252845217,
+          "pwd": 2.944343971929525e-05,
+          "python": 1.0703502398209337e-05,
+          "r)": 3.1598571744557146e-06,
+          "rT": 4.739785761683572e-06,
+          "readlink": 1.708030905111197e-07,
+          "return": 5.622881268677078e-05,
+          "rm": 2.0012428771552857e-06,
+          "sed": 5.8525138694014606e-05,
+          "set": 3.38395376097374e-06,
+          "setup": 1.8959143046734288e-06,
+          "sleep": 9.479571523367144e-07,
+          "sort": 1.3202270713437093e-05,
+          "strace": 8.540154525555985e-08,
+          "student": 1.610907579134286e-06,
+          "tail": 0.00025213846232612487,
+          "tee": 0.00014118800176096701,
+          "this.leaderboardversion": 1.5799285872278573e-06,
+          "throw": 6.723100371182371e-08,
+          "touch": 3.22419012831557e-05,
+          "tr": 7.48387225528985e-07,
+          "updated": 2.369892880841786e-06,
+          "var": 5.576218543157143e-07,
+          "vim": 3.6422308415492373e-06,
+          "wget": 6.259053914497793e-07,
+          "which": 3.4004784822806465e-06,
+          "write_solution": 0.7033376584757732,
+          "write_solution.sh": 1.4712562505402267e-06,
+          "xargs": 0.0004095698831842976,
+          "xxd": 4.989248170193233e-07,
+          "yyyy": 1.5799285872278573e-06,
+          "\u00daltimo": 1.5799285872278573e-06
+        },
+        "output_cap_events": 0,
+        "malformed_tool_calls": 25144,
+        "trajectories_with_malformed_tool_calls": 1436,
+        "errors": {
+          "model_server": {
+            "trajectories": 0,
+            "types": {}
+          },
+          "environment": {
+            "trajectories": 0,
+            "types": {}
+          },
+          "verifier": {
+            "trajectories": 0,
+            "types": {}
+          },
+          "other": {
+            "trajectories": 0,
+            "types": {}
+          }
+        },
+        "trajectory_files_found": 107359
+      },
+      "pass@3": 0.05105306578667235,
+      "pass@5": 0.06333516210565392,
+      "pass@6": 0.0679098658420566,
+      "pass@7": 0.07186940481277292,
+      "pass@9": 0.07855124383812909,
+      "pass@10": 0.08145727223596076,
+      "pass@11": 0.08415129952015198,
+      "pass@12": 0.08667226216406544,
+      "pass@13": 0.0890496593570364,
+      "pass@14": 0.09130650769995032,
+      "pass@15": 0.09346125186289121
+    },
+    "qwen3-30b-thinking": {
+      "tasks": 6710,
+      "attempts": 107360,
+      "pass@1": 0.02687220566318927,
+      "pass@2": 0.0404284649776453,
+      "pass@4": 0.05652691570724358,
+      "pass@8": 0.07446319204888505,
+      "pass@16": 0.09344262295081968,
+      "trace_metrics": {
+        "trajectories": 107360,
+        "graded": 107360,
+        "solved": 2885,
+        "turns": {
+          "n": 107360,
+          "mean": 2.2323491058122205,
+          "median": 2,
+          "p90": 2,
+          "p95": 3,
+          "max": 158
+        },
+        "input_tokens": {
+          "n": 107360,
+          "mean": 5324.964381520119,
+          "median": 3723,
+          "p90": 7102,
+          "p95": 9203,
+          "max": 2091916
+        },
+        "input_tokens_total": 571688176,
+        "output_tokens": {
+          "n": 107360,
+          "mean": 2052.357004470939,
+          "median": 1504,
+          "p90": 2906,
+          "p95": 3648,
+          "max": 171157
+        },
+        "output_tokens_total": 220341048,
+        "peak_context_tokens": {
+          "n": 107331,
+          "mean": 3477.7956322031846,
+          "median": 2917,
+          "p90": 4658,
+          "p95": 5515,
+          "max": 337979
+        },
+        "peak_context_fraction": {
+          "n": 107331,
+          "mean": 0.10613389990854445,
+          "median": 0.089019775390625,
+          "p90": 0.14215087890625,
+          "p95": 0.168304443359375,
+          "max": 10.314300537109375
+        },
+        "trajectories_above_90_percent_context": 751,
+        "latency_seconds": {
+          "model_call": {
+            "n": 239646,
+            "mean": 10.0809203531748,
+            "max": 684.4818048477173,
+            "median": null,
+            "p90": null,
+            "p95": null,
+            "note": "Per-call samples were not retained; pooled quantiles cannot be reconstructed."
+          },
+          "tool_call": null,
+          "agent": {
+            "n": 107360,
+            "mean": 27.920856770892325,
+            "median": 19.771693,
+            "p90": 38.249375,
+            "p95": 48.429497,
+            "max": 3310.508183
+          },
+          "verifier": {
+            "n": 107359,
+            "mean": 1.29676886014214,
+            "median": 1.230452,
+            "p90": 1.379742,
+            "p95": 1.430097,
+            "max": 118.029042
+          },
+          "trial": {
+            "n": 107360,
+            "mean": 46.75835246828427,
+            "median": 29.411649,
+            "p90": 70.003692,
+            "p95": 130.925438,
+            "max": 3318.148226
+          }
+        },
+        "tools": {
+          "bash_command": {
+            "calls": 129254,
+            "calls_per_trajectory": 1.2039307004470938,
+            "failed": null,
+            "failure_rate": null
+          },
+          "mark_task_complete": {
+            "calls": 218961,
+            "calls_per_trajectory": 2.03950260804769,
+            "failed": null,
+            "failure_rate": null
+          }
+        },
+        "tool_steps": 220846,
+        "tool_steps_without_output": 0,
+        "tool_steps_with_error_text": 2681,
+        "tool_steps_with_error_text_rate": 0.012139681044709888,
+        "terminations": {
+          "context_limit": 1,
+          "session_ended": 29,
+          "task_complete": 107213,
+          "task_timeout": 117
+        },
+        "command_distribution": {
+          "    | Value Contains Null: ": 2.357045208127092e-06,
+          "    | Value Type: ": 2.357045208127092e-06,
+          ");": 9.428180832508368e-06,
+          "9Particle": 4.714090416254184e-06,
+          ":wq": 1.990393731307322e-06,
+          "C-o": 1.1493401395819723e-05,
+          "C-x": 1.1493401395819723e-05,
+          "Exception": 3.1427269441694556e-06,
+          "String": 5.268689288754676e-06,
+          "[no_commands]": 0.0016216471031914393,
+          "[unknown]": 0.00398421158584876,
+          "^O": 5.668132524305625e-06,
+          "^X": 7.553768690807299e-06,
+          "apt": 5.499772152296547e-06,
+          "break": 3.1427269441694556e-06,
+          "cat": 0.0033628010640209837,
+          "cd": 0.004080864787670896,
+          "double": 4.6669495120916424e-05,
+          "echo": 0.28151211856513014,
+          "exit": 0.0007355159571960589,
+          "find": 1.0390045745723867e-05,
+          "grep": 3.1161974310043894e-05,
+          "hasDuplicateHue": 1.5713634720847278e-06,
+          "huffmanMetaCodes": 9.428180832508368e-07,
+          "huffmanXSize": 1.990393731307322e-06,
+          "huffmanYSize": 1.990393731307322e-06,
+          "i": 9.428180832508368e-07,
+          "int": 2.9332118145581587e-06,
+          "linha": 1.178522604063546e-06,
+          "ls": 0.002538584625983866,
+          "mkdir": 9.35675522014088e-05,
+          "mv": 4.9340813023460444e-05,
+          "n": 3.523023313884079e-05,
+          "nano": 0.00018754597173491246,
+          "nextPrefix": 2.357045208127092e-06,
+          "printf": 9.349612658904132e-05,
+          "pwd": 4.173092420865014e-05,
+          "return": 2.2149046997849478e-05,
+          "sed": 5.3779914832099805e-05,
+          "solution.txt": 1.3468829760726238e-06,
+          "tail": 8.571073484098516e-07,
+          "task_complete": 0.003645563255236569,
+          "task_complete:": 0.000479265858985842,
+          "task_completen": 1.414227124876255e-05,
+          "tee": 6.034035732805355e-05,
+          "touch": 1.830638444978708e-05,
+          "true": 0.0004572667703766558,
+          "vi": 2.8284542497525105e-06,
+          "vim": 1.1104301869398743e-05,
+          "which": 1.18750182390403e-05,
+          "write_solution": 0.6966988534284532,
+          "y": 9.428180832508368e-06
+        },
+        "output_cap_events": 0,
+        "malformed_tool_calls": 9456,
+        "trajectories_with_malformed_tool_calls": 1630,
+        "errors": {
+          "model_server": {
+            "trajectories": 0,
+            "types": {}
+          },
+          "environment": {
+            "trajectories": 0,
+            "types": {}
+          },
+          "verifier": {
+            "trajectories": 0,
+            "types": {}
+          },
+          "other": {
+            "trajectories": 0,
+            "types": {}
+          }
+        },
+        "trajectory_files_found": 107360
+      },
+      "pass@3": 0.049585373642750695,
+      "pass@5": 0.06213431377365804,
+      "pass@6": 0.06684295585189773,
+      "pass@7": 0.07090064927620815,
+      "pass@9": 0.07763596500369974,
+      "pass@10": 0.08049430078118602,
+      "pass@11": 0.08309433053285512,
+      "pass@12": 0.08547911105288154,
+      "pass@13": 0.08768229721098575,
+      "pass@14": 0.08973050173869847,
+      "pass@15": 0.09164493293591654
+    }
+  },
+  "stage7": "runs/2026-10-07-crosscodeeval-pass16/stage-7-combined.json.gz",
+  "coverage": "runs/2026-10-07-crosscodeeval-pass16/coverage.json",
+  "notes": [
+    "Includes original collection and accepted recovery attempts.",
+    "One context-limit failure counts as reward 0 under the agreed evaluation policy; its absent verifier score and original error remain recorded."
+  ],
+  "pass_at_k": "1 - comb(16 - solved, k) / comb(16, k)",
+  "reported_k": [
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    7,
+    8,
+    9,
+    10,
+    11,
+    12,
+    13,
+    14,
+    15,
+    16
+  ]
+}

runs/2026-10-07-crosscodeeval-pass16/add_all_k.py

@@ -0,0 +1,56 @@
+"""Add every pass@k from 1 through 16 to the existing teacher-results PR."""
+from pathlib import Path
+import sys,json,math,hashlib,shutil
+import pyarrow as pa
+import pyarrow.parquet as pq
+from huggingface_hub import HfApi,CommitOperationAdd,hf_hub_download
+root=Path(__file__).resolve().parent;release=root/'release';run='2026-10-07-crosscodeeval-pass16';out=release/'runs'/run
+sys.path.insert(0,'/home/hpc/y500bb/y500bb12/OT_next_data')
+from validation.checks.reward_metrics import pass_at_k
+from validation.reporting.plot_results import rates
+state=json.loads((root/'hf-publication.json').read_text());changed=[];groups={}
+for p in sorted(release.glob('crosscodeeval-*/pass_counts.parquet')):
+ rows=pq.read_table(p).to_pylist()
+ for row in rows:
+  if row['run']!=f'runs/{run}.json':continue
+  n,c=row['attempts'],row['solved'];assert n==16
+  for k in range(1,17):
+   value=pass_at_k(n,c,k);expected=1-(math.comb(n-c,k)/math.comb(n,k) if n-c>=k else 0)
+   assert abs(value-expected)<1e-12
+   row[f'pass@{k}']=value
+  assert all(row[f'pass@{k}']<=row[f'pass@{k+1}'] for k in range(1,16))
+ pq.write_table(pa.Table.from_pylist(rows),p);changed.append(p)
+summary=json.loads((out/'summary.json').read_text())
+for model in summary:
+ grouped=rates(out/f'{model}-stage-6.json',list(range(1,17)));groups[model]=grouped
+ summary[model].update({f'pass@{k}':v for k,v in grouped['All tasks'].items()})
+(out/'summary.json').write_text(json.dumps(summary,indent=2)+'\n');changed.append(out/'summary.json')
+p=release/'runs'/f'{run}.json';record=json.loads(p.read_text());record['summary']=summary;record['reported_k']=list(range(1,17));p.write_text(json.dumps(record,indent=2)+'\n');changed.append(p)
+p=out/'pass-at-k.json';p.write_text(json.dumps(groups,indent=2)+'\n');changed.append(p)
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+labels=['csharp','java','python','typescript','All tasks'];fig,axes=plt.subplots(1,2,figsize=(13,4.8),sharey=True,layout='constrained')
+for ax,(model,grouped) in zip(axes,groups.items()):
+ for label in labels:
+  values=grouped[label];ax.plot(list(values),[100*v for v in values.values()],marker='o',markersize=3,label=label,linewidth=2 if label=='All tasks' else 1.3)
+ ax.set(title='Qwen Instruct' if 'instruct' in model else 'Qwen Thinking',xlabel='k (attempts)',xticks=range(1,17),xlim=(1,16),ylim=(0,20));ax.grid(alpha=.2);ax.spines[['top','right']].set_visible(False);ax.legend(frameon=False)
+axes[0].set_ylabel('Pass@k (%)')
+for ext in ['png','svg']:
+ p=out/f'pass-rates-all-k.{ext}';fig.savefig(p,dpi=180);changed.append(p)
+plt.close(fig)
+# Keep the original preparation script consistent for per-task and summary exports.
+p=root/'prepare.py';s=p.read_text().replace('for k in [1,2,4,8,16]','for k in range(1,17)');p.write_text(s);shutil.copyfile(p,out/p.name);changed.append(out/p.name)
+shutil.copyfile(Path(__file__),out/Path(__file__).name);changed.append(out/Path(__file__).name)
+p=root/'pull-request.md';body=p.read_text().replace('pass@1/2/4/8/16','pass@k for every k from 1 through 16').replace('Plots show pass rates by language,','Plots show pass rates for every k from 1 through 16, by language,');p.write_text(body)
+api=HfApi();commit=api.create_commit(repo_id=state['repo'],repo_type='dataset',revision=state['revision'],parent_commit=state['commit'],operations=[CommitOperationAdd(path_in_repo=str(p.relative_to(release)),path_or_fileobj=str(p)) for p in changed],commit_message='Report every pass@k from 1 through 16')
+state['commit']=commit.oid;state['status']='all-k-uploaded';(root/'hf-publication.json').write_text(json.dumps(state,indent=2)+'\n')
+for p in changed:
+ remote=hf_hub_download(state['repo'],str(p.relative_to(release)),repo_type='dataset',revision=commit.oid,cache_dir=str(root/'verify-cache'))
+ assert hashlib.sha256(Path(remote).read_bytes()).digest()==hashlib.sha256(p.read_bytes()).digest()
+# The opening discussion comment contains the PR description.
+d=api.get_discussion_details(state['repo'],6,repo_type='dataset')
+comments=[e for e in d.events if e.type=='comment']
+original=next(e for e in comments if 'Adds per-task teacher results' in e.content)
+api.edit_discussion_comment(repo_id=state['repo'],discussion_num=6,comment_id=original.id,new_content=body,repo_type='dataset')
+state.update(status='verified',reported_k=list(range(1,17)),files=sum(p.is_file() for p in release.rglob('*')));(root/'hf-publication.json').write_text(json.dumps(state,indent=2)+'\n');print(state['pr_url'],flush=True)

runs/2026-10-07-crosscodeeval-pass16/attempts.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1dbfe1c672a15f9ad53285e73e5da60fd6bedb52518dd14341617ceccfd231ba
+size 1746949

runs/2026-10-07-crosscodeeval-pass16/check_hf.py

@@ -0,0 +1,30 @@
+from pathlib import Path
+import json,hashlib,tarfile,io,sys
+sys.path.insert(0,"/home/hpc/y500bb/y500bb12/OT_next_data")
+from validation.contract import files_digest
+from huggingface_hub import HfApi,hf_hub_download
+import pyarrow.parquet as pq
+root=Path(__file__).resolve().parent;api=HfApi();repo='FWeindel/validated-tasks';head=api.repo_info(repo,repo_type='dataset').sha
+expected={(r['model'],r['task_id']):r for r in json.loads((root.parent/'pass16-retry-20261007/combined-task-results.json').read_text())}
+sources={};mismatch=[];mapping={};normalizations=[]
+def files(row):
+ with tarfile.open(fileobj=io.BytesIO(row['task_binary'])) as tar:return {m.name:tar.extractfile(m).read() for m in tar if m.isfile()}
+
+for lang,v in [('csharp',5),('java',4),('python',3),('typescript',3)]:
+ folder=f'crosscodeeval-{lang}-v{v}';path=hf_hub_download(repo,folder+'/tasks.parquet',repo_type='dataset',revision=head,cache_dir=str(root/'hf-cache'));rows=pq.read_table(path).to_pylist();sources[folder]={'path':path,'tasks':len(rows)}
+ original={r['path']:r for r in pq.read_table(root.parent/f'parquets-pinned/laion__exp_rpt_{folder}.parquet').to_pylist()}
+ for r in rows:
+  a=files(original[r['path']]);b=files(r);sha=files_digest(a.items());assert files_digest(b.items())==r['content_sha256']
+  differences=[n for n in a.keys()|b.keys() if a.get(n)!=b.get(n)]
+  if differences:
+   assert differences==['instruction.md'],(r['path'],differences)
+   suffix=b['instruction.md'].decode()[len(a['instruction.md'].decode().rstrip()):].strip()
+   assert b['instruction.md'].decode().startswith(a['instruction.md'].decode().rstrip())
+   assert suffix=='You have 900 seconds to complete this task. Do not cheat by using online solutions or hints specific to this task.',(r['path'],suffix)
+   normalizations.append(r['path'])
+  for model in ['qwen3-30b-instruct-2507','qwen3-30b-thinking']:assert expected[model,r['path']]['sha256']==sha
+  mapping[r['path']]={'evaluated_source_content_sha256':sha,'published_content_sha256':r['content_sha256']}
+
+print('Published task hash mismatches',len(mismatch),flush=True)
+assert not mismatch
+state={'repo':repo,'head':head,'sources':sources,'task_hashes':mapping,'instruction_suffix_normalization':normalizations};(root/'target.json').write_text(json.dumps(state,indent=2)+'\n')

runs/2026-10-07-crosscodeeval-pass16/command_metrics.py

@@ -0,0 +1,196 @@
+"""Profile submitted shell commands for diverse, reproducible pilot sampling."""
+
+# Adapted from tb-science-discovery/analysis/traces/{shell_commands,trajectory_stats}.py.
+from collections import Counter
+import json
+from pathlib import PurePosixPath
+import re
+import shlex
+
+VERSION = 1
+
+
+def command_name(text):
+    try:
+        words = shlex.split(text, comments=True)
+    except ValueError:
+        return '[unknown]'
+    while words and re.match(r'^[A-Za-z_][A-Za-z_0-9]*=', words[0]):
+        words.pop(0)
+    # Simple wrappers only; option-bearing wrappers need a real shell parser.
+    while words and words[0] in ('env', 'sudo', 'command', 'exec', 'nohup'):
+        words.pop(0)
+        while words and re.match(r'^[A-Za-z_][A-Za-z_0-9]*=', words[0]):
+            words.pop(0)
+    if not words:
+        return '[assignment]'
+    head = words[0]
+    if head.startswith(('-', '<', '>')) or any(c in head for c in '$`*?[]'):
+        return '[unknown]'
+    name = PurePosixPath(head).name
+    # Treat versioned interpreters alike, but preserve python -m pytest, etc.
+    if re.fullmatch(r'python(?:\d+(?:\.\d+)*)?', name):
+        name = 'python'
+        if len(words) > 2 and words[1] == '-m':
+            name += '-m:' + words[2]
+    return name or '[unknown]'
+
+
+def profile(steps):
+    if steps is None:
+        return None
+    counts = Counter()
+    submissions = polls = controls = fallback = 0
+    for step in steps:
+        if step.get('source') != 'agent':
+            continue
+        for call in step.get('tool_calls') or []:
+            name = call.get('function_name', 'unknown')
+            if name == 'mark_task_complete':
+                continue
+            arguments = call.get('arguments') or {}
+            if isinstance(arguments, str):
+                try:
+                    arguments = json.loads(arguments)
+                except ValueError:
+                    arguments = None
+            if not isinstance(arguments, dict):
+                counts['[unknown]'] += 1
+                continue
+            command = next((arguments[k] for k in ('keystrokes', 'command', 'cmd')
+                            if isinstance(arguments.get(k), str)), None)
+            if command is None:
+                # Native Read/Edit tools are still meaningful behavioural evidence.
+                label = '[unknown]' if name in ('Bash', 'bash_command', 'exec_command', 'shell_command') else 'tool:' + name
+                counts[label] += 1
+                continue
+            if not command.strip():
+                polls += 1
+                continue
+            if command.strip() in ('C-c', 'C-d', 'C-z', '\x03', '\x04', '\x1a'):
+                controls += 1
+                continue
+            submissions += 1
+            parts, unsupported = split_commands(command)
+            fallback += unsupported
+            if unsupported:
+                counts['[unknown]'] += 1
+            else:
+                counts.update(command_name(part) for part in parts)
+    return {'version': VERSION, 'counts': dict(sorted(counts.items())),
+            'submissions': submissions, 'polls': polls, 'controls': controls,
+            'unsupported_submissions': fallback}
+
+
+def normalized(counts):
+    total = sum(counts.values())
+    return {k: v / total for k, v in sorted(counts.items()) if v > 0} if total else {}
+
+
+def sampling_distribution(command_profile):
+    """Keep missing/entirely unparsed evidence out of distance selection."""
+    if command_profile is None:
+        return None
+    counts = command_profile['counts']
+    known = {k: v for k, v in counts.items() if k != '[unknown]'}
+    if not known:
+        return None if counts else {'[no_commands]': 1.0}
+    return normalized(counts)
+
+
+def mean_distribution(profiles):
+    distributions = [d for p in profiles if (d := sampling_distribution(p)) is not None]
+    if not distributions:
+        return None
+    total = Counter()
+    for distribution in distributions:
+        total.update(distribution)
+    return {k: v / len(distributions) for k, v in sorted(total.items())}
+
+
+def split_commands(text):
+    parts, buf, pending = [], [], []
+    quote = None
+    depth = 0
+    complex_syntax = False
+    i = 0
+
+    def flush():
+        value = ''.join(buf).strip()
+        if value:
+            parts.append(value)
+        buf.clear()
+
+    while i < len(text):
+        c = text[i]
+        if c == '\\' and quote != "'" and i + 1 < len(text):
+            buf.append(text[i:i+2])
+            i += 2
+            continue
+        if quote:
+            buf.append(c)
+            if c == quote:
+                quote = None
+            i += 1
+            continue
+        if c in "'\"`":
+            quote = c
+            buf.append(c)
+            i += 1
+            continue
+        if c == '#' and (i == 0 or text[i-1].isspace()):
+            end = text.find('\n', i)
+            i = len(text) if end < 0 else end
+            continue
+        if text.startswith('<<<', i):
+            buf.append('<<<')
+            i += 3
+            continue
+        if text.startswith('<<', i):
+            m = re.match(r"<<(-?)[ \t]*('([^']*)'|\"([^\"]*)\"|([A-Za-z_][\w]*))", text[i:])
+            if not m:
+                return [text.strip()], True
+            pending.append((m[3] or m[4] or m[5], bool(m[1])))
+            buf.append(m[0])
+            i += len(m[0])
+            continue
+        if c in '({':
+            depth += 1
+            complex_syntax = True
+        elif c in ')}':
+            depth -= 1
+        if c == '\n' and pending and depth == 0:
+            flush()
+            i += 1
+            for delimiter, tabs in pending:
+                while True:
+                    end = text.find('\n', i)
+                    line = text[i:] if end < 0 else text[i:end]
+                    i = len(text) if end < 0 else end + 1
+                    if (line.lstrip('\t') if tabs else line) == delimiter:
+                        break
+                    if end < 0:
+                        return [text.strip()], True
+            pending.clear()
+            continue
+        redirect_amp = c == '&' and ((i > 0 and text[i-1] in '<>') or text[i:i+2] == '&>')
+        if depth == 0 and c in ';|&\n' and not redirect_amp:
+            flush()
+            i += 1
+            if c in '|&' and i < len(text) and text[i] == c:
+                i += 1
+            continue
+        buf.append(c)
+        i += 1
+    flush()
+    if quote or depth or pending or complex_syntax:
+        return [text.strip()], True
+    # These need an AST to distinguish shell syntax from executable commands.
+    for part in parts:
+        try:
+            first = shlex.split(part)[0]
+        except (ValueError, IndexError):
+            return [text.strip()], True
+        if first in ('if', 'then', 'elif', 'else', 'fi', 'for', 'while', 'until', 'do', 'done', 'case', 'esac', 'select', 'function', '!', 'time', '[[', ']]'):
+            return [text.strip()], True
+    return parts, False

runs/2026-10-07-crosscodeeval-pass16/coverage.json

@@ -0,0 +1,558 @@
+{
+  "attempts": 214720,
+  "sources": [
+    {
+      "source": "000",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/qwen3-30b-instruct-2507/part-1/submissions/e9edf9203a68",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 12078,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "001",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/smoke-instruct/submissions/956b1fb754ea",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 112,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "002",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/qwen3-30b-instruct-2507/part-2/submissions/81048c1f948a",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 11661,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "003",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/qwen3-30b-instruct-2507/part-3/submissions/19236405bdba",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3174,
+      "stage7_report": false,
+      "reconstructed": true
+    },
+    {
+      "source": "004",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/qwen3-30b-instruct-2507/part-4/submissions/5e5f883d8061",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3040,
+      "stage7_report": false,
+      "reconstructed": true
+    },
+    {
+      "source": "005",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/qwen3-30b-thinking/part-1/submissions/eb52c16f13fa",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 12401,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "006",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/smoke-thinking/submissions/a91d9cfb4a62",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 112,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "007",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/qwen3-30b-thinking/part-2/submissions/14345a679c88",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 12934,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "008",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/rerun-lost-sessions/run/submissions/89944e4ad6a2",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 2,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "009",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/qwen3-30b-thinking/part-3/submissions/adb5bc9bd9a8",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 12487,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "010",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-20261004/qwen3-30b-thinking/part-4/submissions/82cfb7138e82",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3325,
+      "stage7_report": false,
+      "reconstructed": true
+    },
+    {
+      "source": "011",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/000/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "012",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/001/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3987,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "013",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/002/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "014",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/003/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3983,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "015",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/004/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "016",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/005/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "017",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/006/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "018",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/007/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "019",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/008/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3998,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "020",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/009/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "021",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/010/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "022",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/011/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "023",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/012/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "024",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/013/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3995,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "025",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/014/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "026",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/015/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "027",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/016/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "028",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/017/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 2808,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "029",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/018/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 2007,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "030",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/019/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 1722,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "031",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/020/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 1440,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "032",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/021/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 990,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "033",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/022/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 154,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "034",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/023/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 88,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "035",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/024/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 60,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "036",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/025/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 35,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "037",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-instruct-2507/026/submission",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 15,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "038",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/000/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "039",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/001/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "040",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/002/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3994,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "041",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/003/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "042",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/004/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "043",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/005/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "044",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/006/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "045",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/007/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 4000,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "046",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/008/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3999,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "047",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/009/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3991,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "048",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/010/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "049",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/011/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "050",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/012/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3996,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/013/submission",
+      "selected_attempts": 0,
+      "stage7_report": false
+    },
+    {
+      "source": "052",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/014/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3995,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "053",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/015/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 3200,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "054",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/016/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 1480,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "055",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/017/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 850,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "056",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/018/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 174,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "057",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/019/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 147,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "058",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/020/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 144,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "059",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/021/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 84,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "060",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/022/submission",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 44,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "061",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-retry-20261007/batches/00-00/results/submissions/e8e6e3b86241",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 32,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "062",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-retry-20261007/batches/01-00/results/submissions/1946bc55c7ae",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 16,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "063",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-retry-20261007/batches/02-00/results/submissions/83bfb5485d0f",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 1998,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "064",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-retry-20261007/batches/02-01/results/submissions/4033fdc6d125",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 1998,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "065",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-retry-20261007/pilot-qwen3-30b-instruct-2507/results/submissions/080c83938a2e",
+      "model": "qwen3-30b-instruct-2507",
+      "selected_attempts": 1,
+      "stage7_report": true,
+      "reconstructed": false
+    },
+    {
+      "source": "066",
+      "submission": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-retry-20261007/pilot-qwen3-30b-thinking/results/submissions/7aabffc2d79b",
+      "model": "qwen3-30b-thinking",
+      "selected_attempts": 1,
+      "stage7_report": true,
+      "reconstructed": false
+    }
+  ],
+  "policy_outcomes": [
+    {
+      "model": "qwen3-30b-thinking",
+      "task_id": "crosscodeeval-csharp-0944",
+      "trial": "crosscodeeval-csharp-0944__ary5yNM",
+      "archive": "/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/010/submission/evidence.tar.gz",
+      "member": "results/stage_6_agent_trials/f58f4ddbdcbb/jobs/job/crosscodeeval-csharp-0944__ary5yNM/result.json",
+      "exception": {
+        "exception_type": "ContextManagementInfrastructureError",
+        "exception_message": "The provider rejected a context-recovery request after all copied conversation turns were removed.",
+        "exception_traceback": "Traceback (most recent call last):\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/llms/lite_llm.py\", line 1173, in call\n    raise e\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/llms/lite_llm.py\", line 1140, in call\n    response = await self._dispatch_chat(completion_kwargs)\n               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/pass16-recovery-20261006/batches/qwen3-30b-thinking/010/submission/code/harbor_patches/reasoning_field.py\", line 33, in dispatch\n    return with_reasoning_content(await original(self, *args, **kwargs))\n                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/llms/lite_llm.py\", line 1484, in _dispatch_chat\n    return await self._acreate_chat_raw(completion_kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/llms/lite_llm.py\", line 1435, in _acreate_chat_raw\n    raw_response = await client.chat.completions.with_raw_response.create(\n                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/openai/_legacy_response.py\", line 386, in wrapped\n    return cast(LegacyAPIResponse[R], await func(*args, **kwargs))\n                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/openai/resources/chat/completions/completions.py\", line 2907, in create\n    return await self._post(\n           ^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/openai/_base_client.py\", line 1992, in post\n    return await self.request(cast_to, opts, stream=stream, stream_cls=stream_cls)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/openai/_base_client.py\", line 1777, in request\n    raise self._make_status_error_from_response(err.response) from None\nopenai.BadRequestError: Error code: 400 - {'error': {'message': \"This model's maximum context length is 32768 tokens. However, you requested 4096 output tokens and your prompt contains at least 28673 input tokens, for a total of at least 32769 tokens. Please reduce the length of the input prompt or the number of requested output tokens. (parameter=input_tokens, value=28673)\", 'type': 'BadRequestError', 'param': 'input_tokens', 'code': 400}}\n\nThe above exception was the direct cause of the following exception:\n\nTraceback (most recent call last):\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/context_management.py\", line 64, in call\n    response = await self._llm.call(\n               ^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/tenacity/asyncio/__init__.py\", line 193, in async_wrapped\n    return await copy(fn, *args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/tenacity/asyncio/__init__.py\", line 112, in __call__\n    do = await self.iter(retry_state=retry_state)\n         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/tenacity/asyncio/__init__.py\", line 157, in iter\n    result = await action(retry_state)\n             ^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/tenacity/_utils.py\", line 111, in inner\n    return call(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/tenacity/__init__.py\", line 393, in <lambda>\n    self._add_action_func(lambda rs: rs.outcome.result())\n                                     ^^^^^^^^^^^^^^^^^^^\n  File \"/home/hpc/y500bb/y500bb12/.local/share/uv/python/cpython-3.12.14-linux-x86_64-gnu/lib/python3.12/concurrent/futures/_base.py\", line 449, in result\n    return self.__get_result()\n           ^^^^^^^^^^^^^^^^^^^\n  File \"/home/hpc/y500bb/y500bb12/.local/share/uv/python/cpython-3.12.14-linux-x86_64-gnu/lib/python3.12/concurrent/futures/_base.py\", line 401, in __get_result\n    raise self._exception\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/tenacity/asyncio/__init__.py\", line 116, in __call__\n    result = await fn(*args, **kwargs)\n             ^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/llms/lite_llm.py\", line 1175, in call\n    self._handle_llm_error(\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/llms/lite_llm.py\", line 1682, in _handle_llm_error\n    raise ContextLengthExceededError from e\nharbor.llms.base.ContextLengthExceededError\n\nThe above exception was the direct cause of the following exception:\n\nTraceback (most recent call last):\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/trial/trial.py\", line 251, in run\n    await self._run()\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/trial/single_step.py\", line 36, in _run\n    await self._run_agent()\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/trial/single_step.py\", line 67, in _run_agent\n    await self._run_agent_phase(\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/trial/trial.py\", line 473, in _run_agent_phase\n    await self._await_phase(\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/trial/trial.py\", line 510, in _await_phase\n    return await task\n           ^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 2378, in run\n    await self._run_agent_loop(\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1974, in _run_agent_loop\n    proactive_summary_result = await self._check_proactive_summarization(\n                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1634, in _check_proactive_summarization\n    return await self._summarize_with_retry(chat, original_instruction, session)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1229, in _summarize_with_retry\n    return await self._summarize(\n           ^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1422, in _summarize\n    summary_result = await self._run_subagent(\n                     ^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1149, in _run_subagent\n    managed_response = await self._dispatch_subagent_request(\n                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/agents/terminus_2/terminus_2.py\", line 1105, in _dispatch_subagent_request\n    managed_response = await self._context_window_manager.call(\n                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/hnvme/workspace/y500bb12-crosscodeeval-pilot/validation-smoke/envs/prep/lib/python3.12/site-packages/harbor/context_management.py\", line 74, in call\n    raise ContextManagementInfrastructureError(\nharbor.llms.base.ContextManagementInfrastructureError: The provider rejected a context-recovery request after all copied conversation turns were removed.\n",
+        "occurred_at": "2026-10-06T18:21:30.143521"
+      },
+      "verifier_result": null,
+      "reward": 0,
+      "reward_source": "evaluation_policy",
+      "status": "agent_context_error",
+      "retry_policy": "exclude already-submitted retry from pass@16"
+    }
+  ]
+}

runs/2026-10-07-crosscodeeval-pass16/pass-at-k.json

@@ -0,0 +1,186 @@
+{
+  "qwen3-30b-instruct-2507": {
+    "csharp": {
+      "1": 0.011917960088691795,
+      "2": 0.019419807834441978,
+      "3": 0.02475847323408299,
+      "4": 0.0289060532962972,
+      "5": 0.03232556738654299,
+      "6": 0.03526397844579663,
+      "7": 0.037865349727877445,
+      "8": 0.04021940939901029,
+      "9": 0.04238534931439588,
+      "10": 0.044404320734697675,
+      "11": 0.046306166732996,
+      "12": 0.048113268844976166,
+      "13": 0.049842941611234294,
+      "14": 0.0515089923626509,
+      "15": 0.05312269031781227,
+      "16": 0.05469327420546933
+    },
+    "java": {
+      "1": 0.05448548812664908,
+      "2": 0.07895338610378189,
+      "3": 0.09449962306822465,
+      "4": 0.10587984574791963,
+      "5": 0.11489097973266839,
+      "6": 0.12238421209397463,
+      "7": 0.12882714910419396,
+      "8": 0.13450297576946652,
+      "9": 0.13959522667306307,
+      "10": 0.14422931949317175,
+      "11": 0.14849517237380033,
+      "12": 0.15246020470294874,
+      "13": 0.15617696946852622,
+      "14": 0.1596877748460862,
+      "15": 0.16302770448548812,
+      "16": 0.1662269129287599
+    },
+    "python": {
+      "1": 0.03125,
+      "2": 0.0484375,
+      "3": 0.05991236772486772,
+      "4": 0.06853632478632478,
+      "5": 0.07545723951973952,
+      "6": 0.08125844063344063,
+      "7": 0.08629261363636365,
+      "8": 0.09078912342801232,
+      "9": 0.09489971833721834,
+      "10": 0.09872419247419248,
+      "11": 0.10232721560846561,
+      "12": 0.10575015262515262,
+      "13": 0.10901951058201058,
+      "14": 0.11215277777777778,
+      "15": 0.11516203703703703,
+      "16": 0.11805555555555555
+    },
+    "typescript": {
+      "1": 0.01808993399339934,
+      "2": 0.027816281628162815,
+      "3": 0.03435938236680811,
+      "4": 0.03926921263554927,
+      "5": 0.043210069633336956,
+      "6": 0.04651528339647151,
+      "7": 0.04937492787740312,
+      "8": 0.05190762666010191,
+      "9": 0.054192024971727946,
+      "10": 0.05628245516859379,
+      "11": 0.05821757999975821,
+      "12": 0.060025568490915025,
+      "13": 0.061727369165487975,
+      "14": 0.06333883388338833,
+      "15": 0.06487211221122112,
+      "16": 0.06633663366336634
+    },
+    "All tasks": {
+      "1": 0.027971311475409835,
+      "2": 0.04189269746646796,
+      "3": 0.05105306578667235,
+      "4": 0.05787564894122272,
+      "5": 0.06333516210565392,
+      "6": 0.0679098658420566,
+      "7": 0.07186940481277292,
+      "8": 0.07538018034292251,
+      "9": 0.07855124383812909,
+      "10": 0.08145727223596076,
+      "11": 0.08415129952015198,
+      "12": 0.08667226216406544,
+      "13": 0.0890496593570364,
+      "14": 0.09130650769995032,
+      "15": 0.09346125186289121,
+      "16": 0.09552906110283159
+    }
+  },
+  "qwen3-30b-thinking": {
+    "csharp": {
+      "1": 0.009931633407243163,
+      "2": 0.01701157920670116,
+      "3": 0.02255437651779115,
+      "4": 0.027151303980572274,
+      "5": 0.031093229263960975,
+      "6": 0.03454140094938321,
+      "7": 0.037596909742142554,
+      "8": 0.040330417714009734,
+      "9": 0.04279572976912223,
+      "10": 0.04503626454845967,
+      "11": 0.0470882437345852,
+      "12": 0.04898231849451362,
+      "13": 0.05074437757364587,
+      "14": 0.05239591032273959,
+      "15": 0.05395417590539542,
+      "16": 0.05543237250554324
+    },
+    "java": {
+      "1": 0.053067282321899736,
+      "2": 0.0768117854001759,
+      "3": 0.09180738786279682,
+      "4": 0.10255617733190292,
+      "5": 0.11089852320063402,
+      "6": 0.11771994496268902,
+      "7": 0.1234975644408362,
+      "8": 0.12851186925065552,
+      "9": 0.13294079930623465,
+      "10": 0.1369052451506277,
+      "11": 0.1404922584012294,
+      "12": 0.1437672881208501,
+      "13": 0.14678100263852242,
+      "14": 0.14957343887423044,
+      "15": 0.1521767810026385,
+      "16": 0.1546174142480211
+    },
+    "python": {
+      "1": 0.025607638888888888,
+      "2": 0.04157021604938271,
+      "3": 0.053290343915343916,
+      "4": 0.06278998778998779,
+      "5": 0.07087265550807217,
+      "6": 0.07792525761275762,
+      "7": 0.08418540371665371,
+      "8": 0.08982290857290857,
+      "9": 0.09496527777777777,
+      "10": 0.09970845127095128,
+      "11": 0.10412374084249085,
+      "12": 0.10826338013838013,
+      "13": 0.11216517857142858,
+      "14": 0.11585648148148148,
+      "15": 0.1193576388888889,
+      "16": 0.12268518518518519
+    },
+    "typescript": {
+      "1": 0.018234323432343234,
+      "2": 0.027967546754675464,
+      "3": 0.03472123998114097,
+      "4": 0.03996391397381496,
+      "5": 0.04425162296449425,
+      "6": 0.04786756148142286,
+      "7": 0.05098308869348473,
+      "8": 0.053712037870453706,
+      "9": 0.05613428650557363,
+      "10": 0.05830798739214581,
+      "11": 0.06027665953408528,
+      "12": 0.06207358647952707,
+      "13": 0.06372465818010373,
+      "14": 0.06525027502750275,
+      "15": 0.06666666666666667,
+      "16": 0.06798679867986798
+    },
+    "All tasks": {
+      "1": 0.02687220566318927,
+      "2": 0.0404284649776453,
+      "3": 0.049585373642750695,
+      "4": 0.05652691570724358,
+      "5": 0.06213431377365804,
+      "6": 0.06684295585189773,
+      "7": 0.07090064927620815,
+      "8": 0.07446319204888505,
+      "9": 0.07763596500369974,
+      "10": 0.08049430078118602,
+      "11": 0.08309433053285512,
+      "12": 0.08547911105288154,
+      "13": 0.08768229721098575,
+      "14": 0.08973050173869847,
+      "15": 0.09164493293591654,
+      "16": 0.09344262295081968
+    }
+  }
+}

runs/2026-10-07-crosscodeeval-pass16/prepare.py

@@ -0,0 +1,125 @@
+"""Reconcile accepted pass@16 attempts and aggregate their saved stage-7 measurements."""
+from pathlib import Path
+import sys,json,gzip,collections,tarfile,tempfile,shutil,hashlib,math
+root=Path(__file__).resolve().parent
+repo=Path('/home/hpc/y500bb/y500bb12/OT_next_data');sys.path.insert(0,str(repo))
+from validation.checks.trace_metrics import trial_metrics,aggregate,distribution
+from validation.checks.reward_metrics import pass_at_k,reward_group
+from validation.reporting.plot_results import rates,plot_rates,solved_counts,plot_solved_counts,write_trace_plots
+import pyarrow as pa
+import pyarrow.parquet as pq
+base=root.parent;old=base/'pass16-recovery-20261006';new=base/'pass16-retry-20261007'
+release=root/'release';run='2026-10-07-crosscodeeval-pass16';out=release/'runs'/run;out.mkdir(parents=True,exist_ok=True)
+def read(p):return json.loads(Path(p).read_text())
+def write(p,obj):p.parent.mkdir(parents=True,exist_ok=True);p.write_text(json.dumps(obj,indent=2)+'\n')
+def zipped(p,obj):
+ p.parent.mkdir(parents=True,exist_ok=True)
+ with gzip.open(p,'wt') as f:json.dump(obj,f)
+final=read(new/'combined-task-results.json');expected={(r['model'],r['task_id']):r for r in final};inv=read(old/'inventory.json')
+sources={};selected=collections.defaultdict(dict)
+for r in inv['records']:
+ for t in r['graded_trials']:
+  if 'archive' in t:
+   sub=str(Path(t['archive']).parent);sources[sub]={'model':r['model'],'archive':t['archive'],'selection':'original'}
+   trial=t['member'].split('/jobs/job/')[1].split('/')[0];selected[sub][trial]={'task':r['task_id'],'reward':t['reward'],'member':t['member']}
+  else:
+   sub=str(Path(t['report']).parent.parent);sources[sub]={'model':r['model'],'selection':'scored'}
+for b in read(old/'batches.json'):
+ sub=str(Path(b['request']).parent);sources[sub]={'model':b['model'],'selection':'scored'}
+for j in read(new/'recovery-results.json')['jobs']:
+ sub=str(Path(j['submission']).parent);sources[sub]={'model':j['model'],'selection':'scored','exclude':j.get('excluded_diagnostic_tasks',[])}
+policy=read(new/'agent-failure-outcomes.json')[0]
+policy_trial=policy['trial'];allrows=collections.defaultdict(list);ledger=[];coverage=[]
+for si,(sub,info) in enumerate(sources.items()):
+ p=Path(sub);reports=list((p/'report').glob('stage-7-*.json'));recovered=False
+ if reports:
+  d=read(reports[0]);assert d['complete'] and not d.get('contract_findings');rows=d['trace_metrics']['trajectories']
+  zipped(out/'sources'/f'{si:03d}-stage-7.json.gz',d)
+ else:
+  archive=info.get('archive');rows=[]
+  if not archive:
+   coverage.append({'submission':sub,'selected_attempts':0,'stage7_report':False});continue
+  wanted=selected[sub];results={};trajectories={};scan_error=None
+  try:
+   with tarfile.open(archive,'r|*') as tar:
+    for member in tar:
+     if not member.isfile() or '/jobs/job/' not in member.name:continue
+     suffix=member.name.split('/jobs/job/')[1];name=suffix.split('/')[0]
+     if name not in wanted:continue
+     if member.name==wanted[name]['member']:results[name]=json.load(tar.extractfile(member))
+     if suffix.endswith('/agent/trajectory.json') and (suffix.count('/')==2 or name not in trajectories):trajectories[name]=json.load(tar.extractfile(member))
+  except (EOFError,tarfile.ReadError,OSError) as e:scan_error=str(e)
+  assert set(results)==set(wanted),(sub,len(results),len(wanted))
+  with tempfile.TemporaryDirectory(prefix='cce-metrics-') as tmp:
+   for name,result in results.items():
+    trial=Path(tmp)/name;(trial/'agent').mkdir(parents=True)
+    if name in trajectories:(trial/'agent/trajectory.json').write_text(json.dumps(trajectories[name]))
+    row=trial_metrics(trial,result,context_limit=32768);row['task']=wanted[name]['task'];row['model_call_seconds']=distribution(row['model_call_seconds'] or [])
+    rows.append(row)
+    shutil.rmtree(trial)
+  recovered=True
+  zipped(out/'sources'/f'{si:03d}-reconstructed-stage-7.json.gz',{'stage':7,'complete':True,'archive_scan_error':scan_error,'trajectories':rows})
+ count=0
+ for row in rows:
+  task=row['task'];trial=row['trial'];key=(info['model'],task)
+  if info['selection']=='original':
+   if trial not in selected[sub]:continue
+   assert row['reward']==selected[sub][trial]['reward']
+  else:
+   if task in info.get('exclude',[]):continue
+   if row['reward'] not in (0,1) and trial!=policy_trial:continue
+  if trial==policy_trial:
+   row['verifier_reward']=row['reward'];row['reward']=0;row['reward_source']='evaluation_policy';row['termination']='context_limit'
+   row['recorded_exception_type']='ContextManagementInfrastructureError'
+   for errors in row['errors'].values():
+    if 'ContextManagementInfrastructureError' in errors:errors.remove('ContextManagementInfrastructureError')
+  row['model']=info['model'];row['source']=f'sources/{si:03d}';row.pop('trial_path',None)
+  assert key in expected
+  allrows[info['model']].append(row);count+=1
+  ledger.append({'model':info['model'],'task':task,'trial':trial,'reward':row['reward'],'source':row['source'],'reward_source':row.get('reward_source','verifier')})
+ coverage.append({'source':f'{si:03d}','submission':sub,'model':info['model'],'selected_attempts':count,'stage7_report':bool(reports),'reconstructed':recovered})
+ print(si,info['model'],count,'reconstructed' if recovered else 'saved report',flush=True)
+# Every accepted attempt appears once, with exactly the published reward multiset.
+assert len({(r['model'],r['source'],r['trial']) for r in ledger})==len(ledger)
+observed=collections.defaultdict(list)
+for r in ledger:observed[r['model'],r['task']].append(r['reward'])
+assert set(observed)==set(expected)
+for key,rewards in observed.items():assert sorted(rewards)==sorted(expected[key]['rewards']),(key,len(rewards),len(expected[key]['rewards']))
+write(out/'coverage.json',{'attempts':len(ledger),'sources':coverage,'policy_outcomes':read(new/'agent-failure-outcomes.json')})
+zipped(out/'attempts.json.gz',ledger)
+# Saved per-call distributions cannot recover global quantiles; retain weighted means and maxima.
+def summarize(rows):
+ calls=[r['model_call_seconds'] for r in rows if r.get('model_call_seconds')]
+ clean=[dict(r,model_call_seconds=None) for r in rows];a=aggregate(clean)
+ if calls:
+  n=sum(c['n'] for c in calls);a['latency_seconds']['model_call']={'n':n,'mean':sum(c['mean']*c['n'] for c in calls)/n,'max':max(c['max'] for c in calls),'median':None,'p90':None,'p95':None,'note':'Per-call samples were not retained; pooled quantiles cannot be reconstructed.'}
+ return a
+summary={};panels=[];tables=collections.defaultdict(list);combined={'stage':7,'name':'trace_metrics','complete':True,'scope':'Only attempts counted in the final pass@16 results','models':{},'coverage':'coverage.json'}
+folders={k:f'crosscodeeval-{k}-v{v}' for k,v in [('csharp',5),('java',4),('python',3),('typescript',3)]}
+for model,rows in allrows.items():
+ assert len(rows)==107360
+ tasks=collections.defaultdict(list)
+ for row in rows:tasks[row['task']].append(row)
+ a=summarize(rows);a['trajectory_files_found']=sum(r['trajectory_found'] for r in rows)
+ combined['models'][model]={'all_tasks':a,'tasks':{t:summarize(rr) for t,rr in tasks.items()}}
+ zipped(out/f'{model}-trace-metrics.json.gz',rows)
+ stage7={'stage':7,'complete':True,'trace_metrics':{'trajectories':rows,'all_tasks':a}}
+ temp=root/f'{model}-stage-7.json';write(temp,stage7)
+ for plot in write_trace_plots(temp).values():shutil.move(str(root/plot),out/plot)
+ temp.unlink()
+ mine=[r for r in final if r['model']==model];items=[]
+ for r in mine:
+  assert len(r['rewards'])==16 and r['complete'];n=int(r['solved']);language=r['task_id'].split('-')[1]
+  record={'path':r['task_id'],'content_sha256':r['sha256'],'model':model,'attempts':16,'graded':16,'solved':n,'group':reward_group(r['rewards']),'run':f'runs/{run}.json',**{f'pass@{k}':pass_at_k(16,n,k) for k in range(1,17)}}
+  tables[folders[language]].append(record)
+  items.append({'task':r['task_id'],'rewards':r['rewards'],'metrics':{'attempts':16,'solved':n,'graded':16}})
+ stage6=out/f'{model}-stage-6.json';write(stage6,{'stage':6,'complete':True,'dry_run':False,'model':model,'items':items})
+ panels.append((model,rates(stage6,[1,2,4,8,16])));plot_solved_counts(solved_counts(stage6),out/f'{model}-solved-counts.png')
+ summary[model]={'tasks':len(mine),'attempts':len(rows),**{f'pass@{k}':sum(pass_at_k(16,int(r['solved']),k) for r in mine)/len(mine) for k in range(1,17)},'trace_metrics':a}
+plot_rates(panels,[1,2,4,8,16],out/'pass-rates.png')
+zipped(out/'stage-7-combined.json.gz',combined)
+write(out/'summary.json',summary)
+for folder,rows in tables.items():
+ p=release/folder/'pass_counts.parquet';p.parent.mkdir(parents=True,exist_ok=True);pq.write_table(pa.Table.from_pylist(rows),p)
+write(release/'runs'/f'{run}.json',{'run':run,'kind':'pass_counts_and_trace_metrics','summary':summary,'stage7':f'runs/{run}/stage-7-combined.json.gz','coverage':f'runs/{run}/coverage.json','notes':['Includes original collection and accepted recovery attempts.','One context-limit failure counts as reward 0 under the agreed evaluation policy; its absent verifier score and original error remain recorded.'],'pass_at_k':'1 - comb(16 - solved, k) / comb(16, k)'})
+print('READY',len(ledger),'attempts',flush=True)

runs/2026-10-07-crosscodeeval-pass16/provenance.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61a880df3adcb9b2f3778fec6634857b28824f7d73ab02f16b738d6210c95a25
+size 8560855

runs/2026-10-07-crosscodeeval-pass16/publish.py

@@ -0,0 +1,66 @@
+"""Publish reconciled CrossCodeEval counts, combined stage-7 statistics and plots."""
+from pathlib import Path
+import json,gzip,hashlib,shutil,sys
+import pyarrow as pa
+import pyarrow.parquet as pq
+from huggingface_hub import HfApi,CommitOperationAdd,hf_hub_download
+from huggingface_hub.utils import EntryNotFoundError
+root=Path(__file__).resolve().parent;release=root/'release';target=json.loads((root/'target.json').read_text());repo=target['repo'];api=HfApi();run='2026-10-07-crosscodeeval-pass16';out=release/'runs'/run
+summary=json.loads((out/'summary.json').read_text());coverage=json.loads((out/'coverage.json').read_text())
+# Link results to the published payload while retaining the evaluated source hash.
+for folder in target['sources']:
+ p=release/folder/'pass_counts.parquet';rows=pq.read_table(p).to_pylist()
+ for row in rows:
+  hashes=target['task_hashes'][row['path']];row['evaluated_source_content_sha256']=hashes['evaluated_source_content_sha256'];row['content_sha256']=hashes['published_content_sha256']
+ keys={(r['path'],r['model']) for r in rows}
+ try:
+  previous=hf_hub_download(repo,folder+'/pass_counts.parquet',repo_type='dataset',revision=target['head'],cache_dir=str(root/'hf-cache'))
+  rows += [r for r in pq.read_table(previous).to_pylist() if (r['path'],r['model']) not in keys]
+ except EntryNotFoundError:pass
+ pq.write_table(pa.Table.from_pylist(sorted(rows,key=lambda r:(r['path'],r['model']))),p)
+provenance={'published_revision':target['head'],'task_hashes':target['task_hashes'],'normalization':{'tasks':len(target['instruction_suffix_normalization']),'only_difference':'Published instruction appends the deterministic 900-second time-limit and anti-cheating reminder. Executable files, tests, solutions and resources match.','original_collection_fix_instruction_suffix':True},'sources':[]}
+for source in coverage['sources']:
+ sub=Path(source['submission']);request=json.loads((sub/'request.json').read_text());contract=Path(request['args']['contract']);d=json.loads(contract.read_text());manifest=contract.with_suffix('.tasks.json')
+ entry={'source':source.get('source'),'submission':str(sub),'contract':d,'tasks':json.loads(manifest.read_text())}
+ reports=list((sub/'report').glob('stage-6-*.json'))
+ if reports:entry['stage6']=json.loads(reports[0].read_text())
+ provenance['sources'].append(entry)
+with gzip.open(out/'provenance.json.gz','wt') as f:json.dump(provenance,f)
+for name in ['prepare.py','check_hf.py','publish.py']:shutil.copyfile(root/name,out/name)
+code=Path('/home/hpc/y500bb/y500bb12/OT_next_data')
+for name in ['trace_metrics','command_metrics','reward_metrics']:shutil.copyfile(code/f'validation/checks/{name}.py',out/f'{name}.py')
+missing=sum(v['attempts']-v['trace_metrics']['trajectory_files_found'] for v in summary.values())
+body='''Adds per-task teacher results for all 6,710 CrossCodeEval tasks, with 16 attempts per task for each model (214,720 attempts total).
+
+| Model | Scored attempts | Pass@1 | Pass@16 |
+| --- | ---: | ---: | ---: |
+'''
+for m,label in [('qwen3-30b-instruct-2507','Qwen3-30B-A3B-Instruct-2507'),('qwen3-30b-thinking','Qwen3-30B-A3B (Thinking)')]:
+ s=summary[m];body+=f"| {label} | {s['attempts']:,} | {100*s['pass@1']:.2f}% | {100*s['pass@16']:.2f}% |\n"
+body+='''
+Includes the original collection and recovery runs, with each counted attempt included once.
+
+- Each language folder has `pass_counts.parquet`: one row per task and model, with solved counts and pass@1/2/4/8/16.
+- One combined stage 7 report contains statistics for both models and every task: tokens, turns, runtime, tool use, termination reasons and errors.
+- Plots show pass rates by language, solved-count distributions and trace statistics. Per-attempt metrics and supporting reports are included.
+
+One attempt has no saved trajectory, so its trace measurements remain missing. One context-limit failure counts as reward 0, with its original error and absent verifier score preserved. Global model-call latency quantiles are unavailable because only per-attempt summaries were retained.
+
+Results and plots: `runs/2026-10-07-crosscodeeval-pass16/`. The combined statistics are in `stage-7-combined.json.gz`; `summary.json` provides a smaller overview.
+'''
+(root/'pull-request.md').write_text(body)
+files=sorted(p for p in release.rglob('*') if p.is_file());assert not any(p.name=='README.md' for p in files)
+statefile=root/'hf-publication.json'
+if statefile.exists():state=json.loads(statefile.read_text())
+else:
+ commit=api.create_commit(repo_id=repo,repo_type='dataset',create_pr=True,parent_commit=target['head'],operations=[CommitOperationAdd(path_in_repo=str(p.relative_to(release)),path_or_fileobj=str(p)) for p in files],commit_message='CrossCodeEval: complete pass@16 results and combined trace statistics',commit_description=body)
+ state={'repo':repo,'pr_url':commit.pr_url,'revision':commit.pr_revision,'commit':commit.oid,'files':len(files),'status':'uploaded'};statefile.write_text(json.dumps(state,indent=2)+'\n');print(state['pr_url'],flush=True)
+remote=api.get_paths_info(repo,[str(p.relative_to(release)) for p in files],repo_type='dataset',revision=state['commit']);assert len(remote)==len(files)
+for entry in remote:
+ p=release/entry.path;assert entry.size==p.stat().st_size,entry.path
+ if entry.lfs:
+  sha=entry.lfs['sha256'] if isinstance(entry.lfs,dict) else entry.lfs.sha256
+  assert sha==hashlib.sha256(p.read_bytes()).hexdigest(),entry.path
+for folder in target['sources']:
+ p=hf_hub_download(repo,folder+'/pass_counts.parquet',repo_type='dataset',revision=state['commit'],cache_dir=str(root/'verify-cache'));assert pq.read_table(p).equals(pq.read_table(release/folder/'pass_counts.parquet'))
+state.update(status='verified',tasks=6710,models=2,attempts=214720,merged=False);statefile.write_text(json.dumps(state,indent=2)+'\n');print(json.dumps(state,indent=2),flush=True)

runs/2026-10-07-crosscodeeval-pass16/qwen3-30b-instruct-2507-trace-metrics.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43acc0f6583b2db23cc16dc1c1a564253cb44052753e955b056bbe01ce2b3793
+size 10239527

runs/2026-10-07-crosscodeeval-pass16/qwen3-30b-thinking-trace-metrics.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c7bafa33c9e200ba7bc4e1e3d9ec9da59bd15937be9ab94b884f69a5e1f5eb4
+size 10566312

runs/2026-10-07-crosscodeeval-pass16/reward_metrics.py

@@ -0,0 +1,80 @@
+"""Reward statistics over repeated agent trials: measurements, not acceptance gates."""
+from collections import Counter
+import math
+from pathlib import Path
+
+from math import comb
+
+GROUPS = ('all_solved', 'all_zero', 'constant_partial', 'varying')
+
+
+def pass_at_k(n, c, k):
+    """Unbiased pass@k estimate; unavailable when fewer than k attempts exist."""
+    if n < k:
+        return None
+    return 1.0 if n - c < k else 1.0 - comb(n - c, k) / comb(n, k)
+
+
+def group_of(task):
+    parts = task.split('-')
+    return parts[1] if len(parts) >= 3 else 'all'
+
+
+def trial_reward(result, key='reward'):
+    value = ((result.get('verifier_result') or {}).get('rewards') or {}).get(key)
+    if isinstance(value, bool) or not isinstance(value, (int, float)) or not math.isfinite(value):
+        return None
+    return value
+
+
+def reward_group(rewards):
+    """Classify observed rewards, leaving ungraded tasks unclassified."""
+    if not rewards:
+        return None
+    if len(set(rewards)) > 1:
+        return 'varying'
+    return 'all_solved' if rewards[0] >= 1 else 'all_zero' if rewards[0] == 0 else 'constant_partial'
+
+
+def task_metrics(results, attempts, key='reward'):
+    """One task's attempts. An attempt without a reward counts as unsolved."""
+    rewards = [r for r in (trial_reward(data, key) for _, data in results) if r is not None]
+    n = max(attempts, len(results))
+    solved = sum(r >= 1 for r in rewards)
+    return {'attempts': n, 'rewards': rewards,
+            'reward_counts': {str(value): count for value, count in sorted(Counter(rewards).items())},
+            'mean_reward': sum(rewards) / len(rewards) if rewards else None,
+            'solved': solved, 'partial': sum(0 < r < 1 for r in rewards),
+            'zero': sum(r == 0 for r in rewards), 'no_reward': n - len(rewards),
+            'exceptions': sum(bool(data.get('exception_info')) for _, data in results),
+            'pass@1': solved / n, 'pass@k': pass_at_k(n, solved, n), 'k': n, 'variance_group': reward_group(rewards)}
+
+
+def aggregate(rows):
+    attempts = sum(m['attempts'] for m in rows)
+    means = [m['mean_reward'] for m in rows if m['mean_reward'] is not None]
+    return {'tasks': len(rows), 'attempts': attempts, 'k': sorted({m['k'] for m in rows}),
+            'mean_reward': sum(means) / len(means) if means else None,
+            'pass@1': sum(m['pass@1'] for m in rows) / len(rows),
+            'pass@k': sum(m['pass@k'] for m in rows) / len(rows),
+            'solved_rate': sum(m['solved'] for m in rows) / attempts,
+            'partial_credit_rate': sum(m['partial'] for m in rows) / attempts,
+            'zero_reward_rate': sum(m['zero'] for m in rows) / attempts,
+            'no_reward_rate': sum(m['no_reward'] for m in rows) / attempts}
+
+
+def summarize(items):
+    """All tasks, each family (middle part of the task ID) and variance groups."""
+    measured = {Path(i['task']).name: i['metrics'] for i in items if i.get('metrics')}
+    if not measured:
+        return None
+    families, groups = {}, {key: [] for key in GROUPS}
+    for task, metrics in sorted(measured.items()):
+        families.setdefault(group_of(task), []).append(metrics)
+        if metrics['variance_group']:
+            groups[metrics['variance_group']].append(task)
+    return {'all_tasks': aggregate(list(measured.values())),
+            'families': {name: aggregate(rows) for name, rows in sorted(families.items())},
+            'variance_groups': groups, 'variance_group_counts': {key: len(tasks) for key, tasks in groups.items()},
+            'tasks_without_rewards': sorted(task for task, m in measured.items() if not m['variance_group']),
+            'tasks_without_trials': sorted(Path(i['task']).name for i in items if not i.get('metrics'))}

runs/2026-10-07-crosscodeeval-pass16/sources/000-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83031ff1d603623ee2b5e041a9d2a11eae52a9736116dee883ce98481fe42f14
+size 2119881

runs/2026-10-07-crosscodeeval-pass16/sources/001-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1de69f9997543f42afb636395d4bc155c7b123d7fa886a8949e10a7a6dad4c3e
+size 29502

runs/2026-10-07-crosscodeeval-pass16/sources/002-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a82fc19892f52f46fc61fec488a0df8a6ce9bd761565d980ba3361bc097c2ee2
+size 2156511

runs/2026-10-07-crosscodeeval-pass16/sources/003-reconstructed-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e7869d36b6ee3939d921e14ed66d85946b8ad52a6edc2c6d2083c5c91ab7a34
+size 313693

runs/2026-10-07-crosscodeeval-pass16/sources/004-reconstructed-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef95d33bb7de14b20ea54096a825aa06047a4adffa26511b4bdb212cf40405d9
+size 303697

runs/2026-10-07-crosscodeeval-pass16/sources/005-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df3209da8ba68ca53101d23f07649ee4fe7d47f7b6c0a36edda8d395440020d3
+size 2179121

runs/2026-10-07-crosscodeeval-pass16/sources/006-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a1a33598636b384cf362c75a4a032dcdba350cb153bced5ea8649a0c91a460b
+size 29693

runs/2026-10-07-crosscodeeval-pass16/sources/007-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec452e140687bdb07185f0666979854f9b4cbff9f97b2dcc5ac76f094d05e4b6
+size 2212549

runs/2026-10-07-crosscodeeval-pass16/sources/008-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c34b7be707e6e284f1357e0dc3cd74a8c9e5dd45c2764d3da3a0f2b9ef3d432c
+size 2240

runs/2026-10-07-crosscodeeval-pass16/sources/009-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed2d2987de6384d336e711052788500b7df8a5b0400f8538d73ea44d569a6138
+size 2180087

runs/2026-10-07-crosscodeeval-pass16/sources/010-reconstructed-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:384aea82d1ec9a79d40d6e94becc3380158427a0ec08025e969ac6b0f7afe650
+size 344288

runs/2026-10-07-crosscodeeval-pass16/sources/011-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de5a0eb3553482c9d95c446b3b1016e470f25139025d28720c965e0a082ef787
+size 537753

runs/2026-10-07-crosscodeeval-pass16/sources/012-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cb94f6a30f5f3f34c25b9a36f8c7724132c057c1fdc40b9f1ffd43b111b34ec
+size 469315

runs/2026-10-07-crosscodeeval-pass16/sources/013-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5490160f9589b511db36675798ce0aac1788c1221a61cf33f40902c6ed27789e
+size 466089

runs/2026-10-07-crosscodeeval-pass16/sources/014-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:652db63927c57c7f87022f2e7b29d8165468ce351e4e1f8d19efcc76a634f603
+size 469089

runs/2026-10-07-crosscodeeval-pass16/sources/015-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f4c9db385c6a22fe89e667cbb433a982338890fec5931089bfd34c6f17fc050
+size 466935

runs/2026-10-07-crosscodeeval-pass16/sources/016-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18eb87f013b32055a6a50485c82c39c6f6209e15a43fc4dc305525b96ee19a16
+size 465974

runs/2026-10-07-crosscodeeval-pass16/sources/017-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6ec89a423a6a07ef2a7a1c860e08903bfb26dad6833e18f26800b8d2144076d
+size 468691

runs/2026-10-07-crosscodeeval-pass16/sources/018-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24537f3db58af4ec523038223a903311d082cdc36175334b8abdcfc9b30440ce
+size 466325

runs/2026-10-07-crosscodeeval-pass16/sources/019-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db2926aa9017cc49abec3b05d5858a1b547b95ac71213a8d78c37a14b8bc4cdd
+size 463511

runs/2026-10-07-crosscodeeval-pass16/sources/020-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd568858b38fe870caa92c27bfab3f1de768315a282ad9aedfa81eb4b48f08e7
+size 470632

runs/2026-10-07-crosscodeeval-pass16/sources/021-stage-7.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ecbab32534286a8d56d2ee02f5353c4ed7501f17a529cfa72fbfbe62b91588c
+size 466476